Slashdot Mirror
Seagate to Offer Solid State Drives in 2008
Lucas123 writes "Seagate will introduce drives based on flash memory in various storage capacities across its range of products including desktop and notebook PCs, according to Sumner Lemon at IDG News Service. The drives are expected to consume less power (longer battery life), offer faster data transfer rates and be more rugged than spinning disk, which has moving parts that can be damaged from an impact."
The rewrite issue has been rehashed a million times. It will be fine. I promise.
I've been looking forward to mainstream solid state drives for a long time. Especially in laptops, since the lack of moving parts presumably means no moving parts and therefore lower power consumption, longer battery life, better durability, and so on. It seems surprising that it's taken this long.
A-Bomb
Based on what precisely?
Not trolling, I just havent ever seen hard stats on current flash/solid state durability over time recently.
Ice Cream has no bones.
is projected out in the future? Normal hard drive capacity growth has certainly seemed to level off lately and perhaps is stagnating (so it seems to me). Yes, flash has grown astronomically the past few years, but is it sustainable to the point of meeting and exceeding conventional drives?
If we had the rate of growth in conventional drives that we had a few years back, we would almost certainly be looking at multi-TB drives right now.
The headache now is that most file systems are optimised for mechanical based storage media so wont this also mean we will have to look at changing to new file systems ?
With sizes up to 160 GB (according to TFA), any guesses on how long until the announcement of a new iPod design using these? If they can get to be price compatible, I know it would make me think about an upgrade from my 3G iPod.
Mess not in the affairs of dragons, for you are crunchy and good with ketchup.
I was wondering the same. The last time I checked, a Flash memory cell was good for about 10^6 read/write cycles. I've heard of schemes that distribute the memory load (so that no one cell gets continually re-written), but I would like to see this issue explicitly addressed when it comes to a flash drive substitute for a hard drive.
wants to be the first monkey to touch the monolith
There are different grades of flash chip, with varying amounts of write cycles. The problem with the kind of flash that you get in a usb keyring is not that flash is limited in the number of writes, but that cheap low-end flash is. The kind of solid state storage in a drive can take millions of write cycles, which combined with a file-system that spreads the writes evenly across he chip will give a decent lifespan.
Cost is still a major issue though. The article only has one number in it, that densities will go up to 160Gb. Do you think they'll take a cheque for that, or you do you have to spread and touch your toes in person?
Slashdot: where don knuth is an idiot because he cant grasp the awesome power of php
Most of these systems will be using wear levelling to prevent the certain flash regions being happered too hard. Any system that does not use wear levelling will break down pretty quickly.
Engineering is the art of compromise.
Do the math. When rewrites were a problem, how big were the chips? How big are the chips now? How many more writes are possible now? The amount of data that becomes a problem is astronomical at this point...the 'rewrite problem' will kick in long after a spinning disk has found a reason to die.
Nerd rage is the funniest rage.
Wear levelling. If you have a gigabyte drive, and you can write each byte a million times, then you would have to do 10^15 writes to the disk before you'd start seeing problems. And man, that's just a crazy way to use a hard disk.
No weapon in the arsenals of the world is so formidable as the will and moral courage of free men.-Ronald Reagan
It's a little hard to avoid being redundant when there's no posts yet while I'm typing mine out.
Flash memory can die?
THen Why not ordinary RAM on top of a normal drive?
When I first bought a hard drive it had less capacity than all the RAM on this computer, and it was a big drive. The salesperson laughed at me thinking there was no way to fill it. And I paid more for the drive than I paid for this RAM by a factor of 2 or 3.
UPS + RAM + disk drive - cache all the stuff you use a lot in RAM, and it's all good. I suppose even Flash can be used for the Program Files and Windows folders.
Know your pads. One time pad: good for cryptography. Two timing pad: where to take your mistress.
Nobody mentioned the noise! SSD's are silent.
I can't wait for ssd's. Every hard drive I've owned has been noisy and they drive me nuts.
As for durability... hrmm... maybe in its current state, flash doesn't last that long. But, the potential has got to be better than a constantly-spinning platter of disks. I've never had a RAM stick, or flash card die on me, but I've lost many hard drives.
Also, I think there may be greater potential for memory density. Spinning platters inevitably have wasted space, forming a cylinder in a rectangular prism.
I'd be interested to see the effect of SSD's on prices of normal hard drives. Normal HDD prices have been plummetting rapidly over the last couple of years - I wonder if the lure of flash will push them down further.
I think with capacity being so important, price/MB will be a big determining factor in getting flash into enterprise storage. I think the desktop, and (obviously) laptop markets will lap it up first.
Same here - yet to find an article addressing the lifespan of these solid state drives.
I've got LOADS of working hard drives which won't boot, because its boot sector has wore off.
They're fine when plugged into the computer as slave drives, but that still means that I gotta buy a new hard drive just to boot the ocmputer from.
It would be REALLY annoying to see that happens on these SSDs.
Am I safe in assuming SATA transfer rates are sufficient to handle a SSD?
Will it move choke points elsewhere on the system?
I'd like to know what other practical benefits such would have other than lower power consumption and durability.
Someone hates these cans.
Anandtech recently reviewed a 32GB SSD drive from Mtron, and compared it with a Raptor HDD.0 64&p=6
http://www.anandtech.com/storage/showdoc.aspx?i=3
TLDR: *big* improvement in game loading times, merely ok performance in file copying, encoding, compression. It's also silent, and saves a lot of power. $1500 though. . . should get much cheaper in the next few years.
The same statement can be made for todays mechanical storage devices. Everything mechanical has a finite number of operations until it fails. In theory that number should last longer than the time it takes to make the capacity obsolete.
Only the State obtains its revenue by coercion. - Murray Rothbard
Cost is still a major issue though. The article only has one number in it, that densities will go up to 160Gb. Do you think they'll take a cheque for that, or you do you have to spread and touch your toes in person?
Considering the level of technophilia around here I bet that second proposition sounds reasonable to some in exchange for a 160GB flash drive. I myself would maybe even engage in a little 2nd base action for that kind of payoff.
Free the Quark 3 from asymptotic confinement! Bring your charm! Don't get down! All colours and flavours welcome!
It's not all that bad. If I remember correctly, most flash memory can take 100,000-300,000.. according to wikipedia:
"while high endurance Flash storage is often marketed with endurance of 1-5 million write cycles"
I did a small research project (informational) on flash stuff recently for school, I believe solid state hard drives back in June or so were said to have about 2 million writes.
2 million writes per sector. You can always move the information around, and algorithms are being written to do that.
But, with all that, seems like hybrid drives would be the way to go right now.. after all, there's no limit on READING from solid state drives, just writing.
Not trolling, I just havent ever seen hard stats on current flash/solid state durability over time recently.
Take a 40GB hard drive, and pretend it's Flash memory. If you wrote 40GB worth of data to it every single day (with the circuitry inside a drive to spread writes out over cells evenly), then you would average 1 write per day across each cell. Flash memory can be written to a minimum of 10,000 times before dying, most is even more reliably by an order of magnitude (100,000 writes). Assuming we have crappy 10,000 write limits, we could write 40GB to the drive every day for 10,000 days, or 27 years, before failing is an issue.
Looking at the 40GB drive in one of my machines, the total writes in its uptime comes to about 800MB, which is a shade under 24 hours uptime. That's 800MB worth of writes in a day, 50 times *less* than writing 40GB to the drive every day, so a 40GB flash drive at my current usage rate could be expected to last 27 * 50, or 1350 years.
A lot longer than I have to worry about. The numbers are going to differ for some people, but the initial stats work out - few people would write to every cell every day, and even then that's decades worth of use.
Innodb log files! Yes, put the log files on it and watch the performance go up. It was a neat hack but for $800 a card it wasn't all that practical. The performance was nice, but it wasn't worth the additional investment per machine for the card.
A number of months ago someone blogged about some solid state cards he was looking at. At around the same time I noticed and commented on "IDE" solid state drives coming to market.
Dinner on Tuesday night with Kevin Burton pushed this into my mind again. What was he looking at going with for his data center?
Solid state drives.
This is smart thinking, it makes a lot of sense.
The performance gain for using solid state hard drives for any database, not only MySQL, is a "no argument". Buying performance like this does require some cost analysis. You balance performance with cost. Not everyone buys fiber channel even if it buys performance.
The performance gain does not outweigh the cost.
Capacity though is a requirement. By capacity I mean the ability to put X amount of CPU in a given space.
Data center capacity is not a growing concern, it is an active concern.
Data centers need power, a lot of power. Their capacity is constrained by available power.
Green technology is common sense. Hard drives have moving parts that generate heat, eat electric, and have high failure rates.
Green technology means capacity because data centers can pack in more hardware.
Tom's Hardware gave a price of $25 per gig almost a year ago. Tom was reviewing a 32 gig drive at the time (which... at 64 I don't need a hard drive in my laptop... I keep my mp3 on my iPod not my laptop).
Today we are looking at about $19 a gig, with 128gig drives coming to market.
This is a premium, when you consider SATA half terabyte disks are at $100 (which works out to 19 cents a gig!).
How much of a price tag do you put on capacity?
I'll join you in being O.T. The FP and my post both went up at 8:06 on exactly the same subject.
Kinda annoying, but the mods are actually helping the conversation along by lowering our scores. I hate rereading the same thing over and over throughout a thread.
I have been researching some of the more current SSD drives lately, and I know that they greatly improved the technology/algorithms behind how they write data to the physical memory. Most companies use some kind of wear-leveling techniques that evenly distribute the writes over the entire surface of the disk, maximizing the disk's life span. I have also read that the different-sized memory modules have different physical characteristics such that smaller modules are actually outlived by larger ones.
I can't give exact figures, but I've seen comparisons showing a reasonable life span (>20 years @ 100GB of writes/day) - some of the numbers are even comparable to those of spinning/mechanical hard drives. Considering how often mechanical hard drives seem to fail, it doesn't seem that there will be any major roadblocks in terms of reliability.
I know what I've written is mostly qualitative (apologies on that), but I know the research into how to mitigate the problem of life span has truly advanced in the last few years as interest in SSD has increased. Jim Gray of Microsoft Research fame, predicted that SSD would replace mechanical drives not far off from now. Check out his paper "Flash Disk Opportunity for Server-Applications" for more on that.
SixD
The speed of sequential access is about the same for HD and flash.
Flash is better for random access which is not very important for end-user, that's why end user doesn't care about SCSI, since for them HD is just storage as it supposed to be.
Kernels/filesystems/application aren't just prepared to use the "power" of flash. What's the point of having flash as a disk cache if RAM already does that?
I don't see where the "most used" algorithm of flash disks is different from those used for RAM, so you happen to have a lot data cached twice for no purpose.
RAM is way faster than Flash and 64bits processors give a big limit (32GB) of RAM/processor. Flash is cheaper but, as said, much slower, flash disks will be as commom as buying a 36GB 15k HD to use as cache of your 750GB 7.2k
That isn't the issue. The issue is the pathetic sizes of solid state drives. A measly 160GB? Hell, By that time a true TB HDD will be released for a whole lot less.
I think that may have been a problem on older flash drives, especially b/c fat32 keeps the fat all in the same place. But newer models have built-in wear leveling. The only thing you'd need to do would be to turn off seek-time optimization, as there's no rotating disk.
A more interesting question would be how these things hold up when used for swap.
All modern flash drives use write levelling to ensure writes
are evenly spread across the device.
This article
takes those numbers and using a hypothetical "write logger" app that
continually writes, estimates an average life of 51 years.
MTron specs for their SSDs estimate:
So lets lay this one to rest. SSDs are worth it.
If they're being used as system drives though, how many times will it have to rewrite the same bits over and over again versus being able to constantly apply your rewrites so you only stress each bit once before rewriting the first bit again?
you cant just say "its more bits, so it'll take longer".
Ice Cream has no bones.
I wonder to what extent current high capacity HDs owe their high power consumption to the needs of high performance (low access time and high bandwidth). But if a large flash cache (say 4-16 GB) buffers the HD, then the HD mechanism could be redesigned to a much lower spec. I'd bet that a ultraslow 300 RPM platter with a stepper-motor head (versus the 4200 to 7500 rpm platter + voice coil technology currently used) would provide adequate performance (and low power consumption) if flash handled the vast majority of accesses and high speed read-writes. The physical disk mechanism would only need to support a bandwidth of about 2-3 Mbytes/sec (for a sustained read of an HD video stream) and flash would provide the 80-150 MBytes/sec burst bandwidth to compete with current laptop drives. (Hardcore video editors wouldn't use this device, but then they wouldn't use most of the low-power laptops on the market anyway).
Two wrongs don't make a right, but three lefts do.
Yeah but for someone like me who runs tons of VMware machines I write way more data than that. Way more.
I may write between 100 to 200 GB every few hours. The main cause is swap, each VM has a swap file plus the host computer often needs swap to handle all the VM's.
Say I write a paltry 400 GB per day, that means my hard drive has a lifespan of 2.7 years. Not so good, eh? Consider I write way more data than that so this flash drive would not be usable.
Assuming it never rewrites the same bit twice before having applied a write/rewrite to every other bit.
are the controllers and file systems going to be able to do that?
I still cant find anything worthwhile concerning these types of drives in this respect.
Ice Cream has no bones.
The problem with Flash (~$12:GB) drives as replacements for rotating magnetic hard drives (~$0.18:GB) is that Flash is a lot more expensive, and Flash wears out after relatively few rewrites. RAM (~$35:GB) is even more expensive, but it doesn't wear out and has much faster performance.
For smaller storage (<10GB, for mobiles), what about a Flash drive with a RAM cache? That gets flushed to Flash once every hour or day or so. For that matter, how come we never saw magnetic drives with builtin RAM caches in the GB scale, occasionally written (in parallel) back to the magnetic disc for reliability? We can set up RAMdisks in SW, but that eats main memory and takes extra configuration. Magnetic or Flash drives with big RAM caches would have much higher performance, and HD vendors could diversify in their extremely competitive market.
--
make install -not war
THANK you.
I was trying to find details between issues here at work, but was failing to do so. This actually answers a lot of my previous questions.
Ice Cream has no bones.
Scratch that, Spaghetti Code pasted some articles that DO explain exactly how this works in another reply to my previous post. I didnt realize how solid the wear leveling deal is set up already, let alone for future production.
Ice Cream has no bones.
ZFS will. Traditional RAID may suffer though: all the drives will have the same write pattern. Make sure your flash drives aren't all the same age!
Do you even lift?
These aren't the 'roids you're looking for.
If you are really interested in specific performance metrics, contact the companies that make SSD and ask them - there are a lot of metrics that people have which are under NDA and can't be disclosed, but you can probably obtain yourself for the purposes of your own research! BitMicro (http://www.bitmicro.com) is one company that has a very impressive lineup in 2008 - think huge capacity drives, and think > 100 years life span.
How do you know that the drive will evenly distribute writes per cell? Its more likely that some cells may remain untouched, which other cells may get written or changed much more frequently.
In America we are imprisoned by our fear of them.
"Now there's the kind of warranty I want to include in my products."
But will your girlfriend believe you?
You just need to have some spare space (say 20% of additional capacity) and dynamically remap areas from the 'working' part of the disk.
I did some poking around the net for information on NAND write cycles. They've already been quoted in the comments here (100,000 to 2,000,000) so I'm just going to post this neat white paper I found on Zeus drives that explains the endurance they get from their SSD Drive. http://www.baydel.com/images/gallery/NAND%20flash% 20resilience.pdf
New! Device Legs: These legs will help your poor OEM installed product escape any hamfistedness it may encounter. Ava
The drive controller will do wear leveling, so it will not rewrite the same bits over and over again, even if the OS thinks it does. This has also been rehashed a million times.
If you know Japanese or Korean you can get the 32G mtron drives for less than $1000
Relax, they still sell mechanical hard drives.
They aren't very good for log files, the write throughput is very slow compared to even a single spindle, and terrible when compared to any sort of striping.
I believe there is a some load factoring going on also, to try to spread the wear across the whole disk also.
If they built a small amount of intelligence into the firmware, this would be extremely effective for boot and application start up speeds. That is, have the hard drive cache the most regularly requested pieces of data: the kernel, c libraries, browser executables and libraries, etc. Startups would be much faster due to faster speeds and lower latencies.
...but without the wear leveling.
The simplest technique is the most effective, young grasshopper.
A-Bomb
Just for starters...
No need to defrag the drive - indeed, would me harmful (uses scarce 'write' cycles). Is a waste of time, since flash memory is written to in a 'random walk' pattern to spread the damage evenly. That's one the main reason it's so hard to 'undelete' stuff from flash mem.
More careful OS management of swapping & caching.
etc.
You SURE it's a million? Arf
How big would one of these things be if they were made using vacuum tubes?
What?
Or maybe just a few years with Vista. Most flavors of Windows NT love to thrash the heck out of hard drives, regardless of whether the user is doing something or not. Plus, background virus scans and defragmenting are sure to take their toll.
However, I would love to put a solid state drive in my happy little Mac, something to think about when this puny 60 GB drive dies (my warranty and ProCare plan have long since expired anyway.)
Next thing you'll be saying they still sell RAM sufficient for your working set.
The integrated electronics do it for you, otherwise the flash drive would 'fail' sequentially, in order of cell use, and you'd steadily see your reported usable capacity dropping. Does not happen. In my experience, flash drives just keep on working - even in intensive use - but then just somtimes fail suddenly and totally, with no warning.
on the accompaning announcement from Apple about the new high-capacity diskless ipods.
Current flash technology has 1-5 million *write* cycles MTBF.
All modern flash drives use write levelling to ensure writes
are evenly spread across the device.
This article
takes those numbers and using a hypothetical "write logger" app that
continually writes, estimates an average life of 51 years.
MTron specs for their SSDs estimate:
Write endurance
In the case of 32GB capacity Mtron SSD: >85 years @ 100GByte / day erase/write cycle
So lets lay this one to rest. SSDs are worth it.
yeah, kinda like the 1-5 million hours for the MTBF of current hard drives. With estimated life of 1000 years.
Will this circuitry be present in all flash drives? These calculations work if this circuitry exists, but actual results could be very different if it does not. Will it be possible to write to the same cell every time that the drive is written to? Say you store 1 byte on the drive and immediately delete it afterward. Then repeat it 10,000 times. That would happen pretty quickly and without the circuitry that you mentioned it could wear out a single cell.
Also, what kind of error checking will these drives have? If it writes to a bad cell, will it catch this and try to write to another cell?
Non-volatile memory would do a lot more than that. I'll leave it to the rest of you to figure it out.
..however, who cares? Suppose you have a partition on your flashdrivwe that is for a swap file, and you thrash on it like crazy. After a couple of months, several drive sectors fail, and you run a chkdsk and repair it, just like any other hard drive.
Any media is going to eventually fail. Your brand spankin new hard drive from seagate or maxtor ships from the factory with defects that existed as a result of the manafacturing process that have been scan and tagged as bad.
I'm just hoping that this means that we can finally freakin get an 'instant on' laptop...
HA! I just wasted some of your bandwidth with a frivolous sig!
I agree. GP should be modded redundant, not interesting.
So can you swap out an old HD with a SSD drive nice and easy, or does this require special motherboads?
//TODO: Insert catchy phrase
Will this circuitry be present in all flash drives?
Yes. It already is. No reason not to include it on future flash drives.
Let's explore a worst case scenario:
Let's assume write-speed is 64MB/s, and that the memory is spec'd to one million writes/cell.
If we assume the same standard as plain ol' disks, 512 byte sectors, and fill the disk to the brim, leaving only a single sector free to write to, let's see what happens.
64MB/512 = 131072 (aka 128K) writes/second to that single cell.
1000000 / 131072 ~= 7,629394531
That's the expected lifetime of that cell. Seven point six three seconds.
Don't bullshit me with "will not happen in your lifetime", please.
I may be paging to my swap file multiple times each minute. It might prove hard to level that activity out across the drive as a whole.
"It's the height of ridiculousness to say for those 9 lines you get hundreds of millions."
Couldn't this be worked around by using RAM instead of the hard drive? I mean, if manufacturers are practically being required to have 2GB of memory installed, I don't see too much need for frequent re-writes....
Remap (or even swap places) cells with heavy usage to other cells.
Say you have the OS in one area that is very rarely written to. As other areas exceed the low end of your usage rate, you remap the low usage data to the high used areas, and vice versa, eventually averaging out their usage.
Basically if a cell is more than a certain threshhold of writes above the lowest (or avg lowest or some other clever metric), exchange them. Repeat as necessary as the different levels of wear move away from average.
One of my former professors told me that's actually one usage of Log structured file systems. Changes are appended to fresh blocks and the pointer is updated.
EvilCON - Made Famous by
As I understand it, solid state drives will have zero seek time (since their is no physical head that has to be moved). In contrast with a HDD you have a mechanical arm that has to whiz about all over the place, sometimes short distances, sometimes longer distances. As a result of this, lots of things are designed to minimize accessing data that's far apart on hard drives, particularly in applications like databases. It sounds like SSD drives with their zero seek times could simplify a lot of software and various design issues.
Somewhere on the order of 1 million erase-write cycles per bit. That should be more than enough even for swapping purposes.
*sigh* back to work...
The chances are astronomical that this floating-point error will ever cause any problems.
Oops, wrong issue.
Find environmentally and socially responsible products on http://buy-right.net
CompactFlash to IDE adapters can be had for $5 or so and work fine with most motherboards.
Other than the fact that a LARGE number of embedded systems use these? Personally, I have a IDE-CF converter on my home server. In it is a 4G CF that I use to store my root. All that goes on the disk is /home, /opt, and a few of the /vars/. Finally, I load squid on a tmpfs partition. The system now runs a LOT faster, quieter, and about 10 degrees cooler. In about a year or so, I will upgrade again. At that time, I will add a second CF (possibly a much faster 4G). But at this rate, I expect the 4GB CF to last at least 5 or more years.
I prefer the "u" in honour as it seems to be missing these days.
"I've been looking forward to mainstream solid state drives for a long time." - by Bombula (670389) on Thursday August 23, @08:09PM (#20337917)
I've had one online here since 2003... you're late man! lol...
CENATEK ROCKET DRIVE I have is 4gb max size, BUT, can be striped with 4 others like it into 1 huge 16gb one!
IT's PCI 2.2 (132mb/sec) + PC133 SDRAM & has a backing powersupply (which I run thru an APC UPS for stability/uptime/security, though I don't place any critical data on it)...
I am STILL using it to this day, almost 5 years later! No moving parts to wear out...
HOWEVER? Let's see a Flash RAM last that long, lol...
E.G./I.E.->
Mine's running strong & doing TONS of repeating writes in a constant stream, by housing my:
1.) The first 1gb partition, NON NTFS COMPRESSED, has Pagefile.sys on it on an NTFS 4096k sector formatted (to match memory transfer sizes used on disk cache, reads, & read ahead for maxing speed).
2.) The second 1gb partition, NOW NTFS COMPRESSED & using 4096kb sector formatting to match memory read + read ahead mechanisms @ the filesystem driver level (& EVEN BLOCK DEVICE DISK DRIVER LEVEL if I wanted by using SuperCache on it as well)
The 2nd partition, & what I use it for, is for webpage caching, temp ops, logging, & housing my %comspec% cmd.exe... & it works in that setup, for better performance.
(Many bennies to this 2 part approach above alone, read on)
HOW? Well, look @ the datatypes I have on it largely:
MOSTLY TEXT BASED DATA, & on a COMPRESSED DISK!
THUS, it gets WAY over 2:1 compression even & by FAR with that kind of data!
Oh, not done, yet...
I gain in speed, YET AGAIN, also because smaller files on disk exist this way as well!
(Due to compression via NTFS filesystem drivers is why (& today's speed of RAM & CPU's way, WAY even offsets the decompression in memory speed, & on disk writes too, only complimenting its read speeds & access speeds more))...
MORE THAN DOUBLING ITS STORAGE RATE really (sure, yes, webpage caching introduces some compressed image data, but still, I gain on the compressible data like text hugely).
(It goes up to 4gb, so you can serve up a website from a 4gb one, FAST reads, 0ms access/seek for example, speed of memory vs. disks 1000's of times slower, play games from it, or movies, or burn them WICKED FAST, from ISO's... man, you NAME it!)
That's "home use stuff, to an extent...
With DB engines (from an article I wrote on this) though?
it was about placing their tables/devices onto them, alongsize pagefiles on another partition & more for TechEd stuff/theories, or using them for smaller DB engines like DBASE/Access/FoxPro OR serving up websites (db driven ones) from it.
ANYHOW - I found out about them, after doing a software one for Windows NT-based systems (NT4/2000/XP/Server 2003)....
That's when I wrote up a review for CENATEK & their RocketDrive (and I got it on a GOOD deal too for doing that article & was featured on their homepage above guys who are fairly famous in this field, & sites article's too, lol)...
The article was based on work I used for a software based/mirror back to backing HDD ware called SuperDisk, by EEC Systems, which took them to the finalist position in the 2000 & 2001 Microsoft Tech Ed, in the toughest category: Boosting SQLServer performance.
(EEC Systems now is a certified Microsoft partner called SuperSpeed.com)
ANYHOW/ANYWAY:
I did that, while I improved the performance of their SuperCache program by up to 40% via an addon ware I got paid for on a contract (whose ideas are now in the mainstream product NOW, via ported code from Delphi, to its C/C++ combination in its driver, & interface, respectively in the current builds).
The article's principals (for databasing alon
Given your comment... what does this do to data recovery, when one DOES fail?
~REZ~ #43301. Who'd fake being me anyway?
JFFS2.
I prefer the "u" in honour as it seems to be missing these days.
4 Years ago we purchased and deployed 12 new workstations at one of my client sites. Over that 4 year period we have had 10 hard drive failures.
While this failure rate may seem exceptional, it's not too far from reality. Today's hard drives have extremely unpredictable lifespans. If you get a bad lot of drives like we did here you are in for a lot of headaches. Sure they were covered by warranty, but time is lost waiting for the overnight replacement part and then re-imaging the new drive.
Frankly, I'm hopeful that eliminating moving parts will remove a serious potential point of failure.
Remember... ZG9uJ3QgZm9yZ2V0IHRvIGRyaW5rIHlvdXIgb3ZhbHRpbmU=
I see some mod thought you were getting a little to interesting. sigh.
Basically, your suggestion might work for a desktop computer, but would be worthless in a laptop.
Laptops are really the market for these sorts of drives, at least for now. In the future as they get better, that may very well change. The resilience and power consumption are much more important than the speed is.
I have heard of people using a setup similar to the one you suggest to speed up the kernel compilation process. Allowing for all of the items to be in ram until they are done with. But it requires a lot of ram, and ram right now is expensive, and at any rate, that wouldn't be very useful for a laptop.
uh, ramsans...........
Some clarifications on my last post, one's you'll like most likely, because they extoll & explain a couple things better + a virtue in hardware I missed:
----
"TODAY, YOU CAN GET BETTER ONES!
(New ones like the Gigabyte IRAM are good, SATA 150 bus based & same bandwidth as that bus, which makes it better than the RocketDrive (PCI 2.2 limited, 132mb/sec transferral rates, vs. SATA1 @ 150gb/sec transferral rates))
The one I want though?
Look this up online -> DDRDrive X1
(PCIExpress people... HUGE more bandwidth memory transfer rates, that BLOW AWAY PCI 2.2, & even SATA1/2 rates)" - APK ----
Oh, NOT ONLY are those newer SSD's faster bus based, but, they also use FASTER RAM THAN MINE DOES (which uses PC-133 SDRAM - the new ones use better like DDR/DDR2 etc.)!
====
AND THIS POINT:
----
"I gain in speed, YET AGAIN, also because smaller files on disk exist this way as well!" - APK
----
By the gain here, it's NOT only doubling in storage on the Ramdrive/SSD, but because the files are tinier in NTFS compressed form, they READ UP FASTER because they are smaller first of all!
(Also again complimenting the 4096k sector formatted gain you get on READS, because they read up like that into RAM, & I also use that on pagefile.sys, it is read in those increments & speed of access/seek that is 0ms & into the nano-second speed of RAM).
Plus, it defrags a 1gb partition in SECONDS... even under use!
Contiguous files in seconds, further speeding it up & QUICKLY!
APK
P.S.=> MORE SPEED... apk
So will harddrives still be called disks when solid state drives are mainstream? Of course people will likely still call them that, but what will they referred to in software (e.g. Disk Manager, Disk Usage Analyzer)?
Would it be possible to image a partition of a solid state drive with an image created from a non-solid-state disk drive?
tasks(723) drafts(105) languages(484) examples(29106)
I may be paging to my swap file multiple times each minute. It might prove hard to level that activity out across the drive as a whole.
perhaps this was a tongue-in-cheek comment... but anyways perhaps this warrants a response for people that may not understand.
ok, so Rule #1: NEVER USE THE FLASH DRIVE FOR PAGING. Use the hard disk drive instead. I can't for the life of me figure out a reasonable answer to completely removing the HDD. Those things are so cheap for so much data density, engineers have taken development of them to an art form. Seriously, paging should be regulated to the 'unreliable' storage medium - that being the HDD of course. What are you afraid of? that moving parts of the HDD are going to fail so consistently - how much abuse are you actually putting your laptop through in the first place?? After all, the only reason why we do paging at all is because we just don't have enough ram. Those things are so close to being disposable memory, and SO much cheaper then flash drives.
IMO I would only use flash drives for only two conditions: critical data that is able to withstand physical abuse of storage medium, and read-data only. So, for the average computer user, that would mostly mean:
that way you reduce the number of writes being done on the drive. Got a bigger HD? You can use it to store your temporary crap: movies you downloaded over BT, your copy of the latest game so you can NOCD it, your porn collection (unless it's VERY important, I suppose..). paging on a flash drive is waste of money - why spend so much on reliable flash ram (more reliable than HDD anyway) so you can use it to PAGE?? That's the equivalent of a chip maker deciding that it's a good idea to make a CPU that contains 512Mb of L1 cache. Or, car analogy time: driving on the highway on first gear.
Eventually we'll start seeing smaller drives in laptops that will do nothing but page. Just wait it out and let's see where this goes.
Remember folks, the cost of memory from most expensive to least expensive goes something like this: L1 cache <- L2 cache <- RAM <- HDD. Where does a flash drive fit in? Somewhere between RAM and HDD. Flash isn't the end-all-be-all to memory. We just use flash ram and HDD because we don't have enough ram (and battery life). Which we use because we don't have enough CPU cache (and battery life).
Apples and oranges. A floating-point bug or a long-division bug or any other processor bug affects any and all software run on that machine. It's not a loss of storage space over time but a random, "your computer just barfed at the worst possible time" error. As others have pointed out, it will be at least two or more years before your 300GB flash drive degrades to 280GB. By then you will have probably replaced it with a 500GB or larger flash drive.
If we go back about 20 years, hard drives were for non-volatile fast-access storage, and tape drives were for backup, bulk data storage, archiving and sometimes data transfer (when there was too much for floppies.)
Now that flash is reaching the point where we can contemplate using it for the primary non-volatile storage niche, we may see hard drives being displaced into the backup/bulk storage/archiving niches. If so, expect to see increasing emphasis on ways to hot-plug hard drives into your computer, and increasing emphasis on price/GB and decreasing emphasis on performance and possibly per-drive capacity.
We'll really know we've reached this point when hard drives are used as a medium for delivering software.
Quattuor res in hoc mundo sanctae sunt: libri, liberi, libertas et liberalitas.
tasks(723) drafts(105) languages(484) examples(29106)
They'll just be getting the chips from Samsung or STM, same as everyone else. Seagate's expertise is in making spindles and drive controllers. I'm not saying they only thing they have to add to the technology is their name, but... well actually, that's exactly what I'm saying. I'd much rather buy a SSD (a kind of stupid acronym in itself - there *is* no "disk") from say STM or Samsung or one of the actual manufacturers.
Causation can cause correlation
Will partitioning the media be physical? For example if I have a 20Gb drive, unpartitioned it will last (2x10^6)(10^6) erase-writes balanced across the whole disk. By partitioning won't you physically circumvent the whole wear leveling idea? Maybe not completely circumvent but you would kill erase-write potiential by a factor of 20 in this case, and since a swap partition can get pretty intense you might run out faster then you think.
A loop, by its nature, continues. If that didn't make sense, start reading this sentence again.
Plus the only parts of most OSes that gets written 'over and over' is the swap file. Put that on a different medium and you're golden for many many years. Heck, Vista's Readyboost is a step in that direction anyways. For those of us using good OSes, some 'Vista edition' motherboards are out with semi-integrated USB flash drives, connected to the mobo. Set that whole thing to swap, set the bigger SSD for your OS and programs.
JUST DON'T DEFRAGMENT!
The preceding comment is my own, and in no way construes an opinon of the Emperor of Mankind.
i know i come late in the game on this article. i just thought i'd like to point out
1 day = 86,400 secs.
40gb = 40,000,000,000
40gb / 1 day = 462,962 bytes per second.
now i understand that people will not have their computer on all day.
I am also not a memorizer of drive write statistics.
but approximate a current sata drive at 15mbps then a solid state drive would die out at approximately 1/30 the time you say it will last which is approximately 1 year.
people complain about replacing a iphone battery every year.
it certainly would force people to upgrade more often.
Flash drives are basically goodness by most measures -- more reliable, quiet, less power use, and faster (or at least, the potential to be far faster than mechanical drives).
... POP!". The motherboard was fried, and as it turned out so was the hard drive, probably due to a voltage spike. But a data recovery service was still able to recover the data from the physical platters. I guess with flash drives, those days will be history -- when a flash drive is gone, it's probably gone for good.
But a friend's recent experience with liquid spilled on a laptop illustrates an interesting point about recoverability. The laptop went, and I quote, "bzzzt
How about backing the non-volatile w/ volatile? Not 1:1 necessarily, but 1:2 or 1:3, plus a capacitor and electronics to get the cache onto the non-vol when the power drops. This would effectively negate the max writes problem...and they've probably already implemented like that. Or just shoving data forward in a loop per cell to ensure cells share writes equally, which is probably closer to what they are doing. Great, now I'll have to go read up on them.
Guys, hate to break it to you, but anyone who wants to be running on solid state, is.
3 IDE-CF adapters cost me 8$ including shipment on ebay last week. My game box runs of a 16GB CF card (200$ - new - on ebay, available for months now) with vista (yes, vista on a 22MB/sec CF, though I've gotten it there via ghosting rather than via a regular install), and my living room PC runs XP off a 2GB CF card that cost about 25$ new (again, ebay price, store prices typically a tad higher).
Yes, 20MB/sec is less than the 50-70MB/sec read speed an average harddrive gives, but that is offset by near-zero seek times.
If under windows, make sure you turn off:
* SWAP
* ntfs Access time writes (fs tuning utility, one command from shell, or a reg key)
And if you want to be even more thorough and flash-friendly:
* 8_3 filename writes (in ntfs every file has two filenames one that is backwards-compatible to 8_3 naming. No need to waste CF writes on that)
* Any software that routinely writes stuff to disk.
If you're fanatic, do:
* Event logger
* Indexing
If you want >16GB, you can buy several, then use LVM/dynamic disk/multiple partitions depending on your OS to use that.
I just have the core 16GB (about 8GB occupied) on the game box, and do the rest of the storage (aka keep the Program Files directory) on the RAID5 fileserver over Gigabit LAN, which gives me about 40MB/sec read and write, which is IMHO sufficient. Were I not to rely on that, I'd get another two 16GB cards on a CF-IDE adapter, plonk a RAID0 on them and voilla (assuming you can get windows to make dynamic disks of removable storage, which the CF cards are still recognized as, even when on the IDE bus), which I am by no means certain.
If you're on Linux, no problem there. anything and verything can be raided and LVM'd at will.
A RAID0 of these would cost 400$, give 32GB and give about 40MB/sec performance.
So no need to get overly excited with SSD. They're just an overpriced nicely bundled version of what is already cheaply available, kinda like external harddrives. And they'll keep on being that for a while yet.
-
Every time a story comes up on /. about SSD, this is invariably the first question in the comments. Seriously, I've seen like seven of these now. Yes, the writes are limited... but with efficient algorithms to spread the writes correctly, and operating systems that are aware of the media, we are talking 10-20 years before it becomes an issue.
No company would want the nightmare of releasing a product that is going to fall apart in 2 years. It would tarnish their reputation forever. Think of the notorious IBM Deathstars (now Hitachi). Those were even on warranty and many will never buy them again because of the hassle of returning so many.
Self-referential Sigs are cool on /. these days...
54
Unfortunately, on the 1,000,001th rehash, we began to see read errors. Who knew?
Yeh sure, and 640K ought to be enough for anybody.
Come on, in a perfect world where people backup once a day a million rewrites would be fine, however when people start running apps on these things there can easily be a million rewrites in a few minutes.
And what about the OS updating the time last accessed date for the file?
It would take one sneaky program (malicious or just bad design) to wipe out a flash drive in no time.
I don't remember who makes it, but there is a company that sells a PCI card which connects to a SATA controller as a ram disk. Card has places to put four memory sticks and supports up to 4GB of ram. Has battery backup so it could be used like a standard hard drive. Never used one myself so I don't know if they are any good, but I'd rather use something like that for swap than a flash drive.
You can't really compare hard drive writes to flash writes. Even if you change one bit in flash, you have to erase the whole block and write it out again. Of course, as the size of the writes approaches the block size, this becomes less of an issue.
How do you know that the drive will evenly distribute writes per cell?
You don't. But what we do know is that if you take a balanced 6-sided die and roll it a large number of times, the distribution of faces to come up will be uniform. That is, each face has an equal chance of being selected. So if we randomly choose a sector and write to it, the wear over large numbers of writes will be uniform over all sectors.
Its more likely that some cells may remain untouched, which other cells may get written or changed much more frequently.
That's why if you happen to hit a cell that already has data, you relocate the existing data and write to it anyway. Even though you are using more write cycles, as long as you don't max the capacity the disk will wear out evenly and you won't use up all of the write cycles anytime soon. Assuming a 40GB disk with a poor 10,000 write cycle limit, that would be 400,000GB of data to write before the disk completely fails. That means over one year (365 days) you'd have to write 1095GB of data a day to kill a disk that had the most optimal wear-leveling algorithm. If the algorithm required an average of 2 writes per every 1 write of actual data due to moving around data, then you'd still have to write more than 500GB to kill the disk in a year. The truth is most people don't immediately max the disk until a good year later if at all. Even then, they would only write in the 10s of GBs unless they totally stripped out their ram capacity.
So it's safe to say that the write cycles are nearly unlimited for useful purposes as long as we attempt to do some kind of uniform distribution across all the cells. Most tech only has a max life span of around 10 years so the write cycles for even poor flash cells is pretty much unlimited for it's useful lifespan. In a laptop or portable device, I'm willing to bet your battery will give you problems before any other hardware. Battery recharge cycles are usually around 500 cycles, yet nobody complains about batteries like they do about flash write cycles.
Hard disk manufacturers do this because disks are bound to have bad sectors as they come out of manufacturing, so the drive can remap these on the fly. It has little to do with disk sector longevity when the sectors are bad out of the box.
Thats the great part. With hard drives you only know that a sector has gone bad after a read. With flash you know immediately when a write fails. When this happens the sector will be marked bad and the write will be attempted somewhere else. Eventually enough bad sectors will cause the drive to become full, but you never lose any data. Just boot from another disk and make a backup. Reads do not harm the disk in any way, only writes do. Even though flash has a more limited number of write cycles, the fact that it fails more reliably makes it more reliable over all.
Virus scans read, they don't write.
Defragmenting would essentially be a thing of the past. You don't need to move the bits together to ensure that the drive head can read the file's parts sequentially, since there's no head, and data is accessed 'immediately'.
The trouble is unless some kind of special technology is used to redistribute the writes across the entire disk, you'll get the same bytes being written to 1000 times a day in a cache situation. 1350/1000 = 1.35 years which is not so good.
These posts express my own personal views, not those of my employer
First, read this comment
Now I don't want to hear any more of this "flash wears out" FUD.
I'm sure the companies spending tens of millions of dollars on research, who stand to make hundreds of millions if the technology works, have thought about this.
If a sector does happen to reach a remaining lifetime significantly lower than other sectors, a way would exist to find a file or three elsewhere on the drive that have been updated once or twice in the drive lifetime, and move them into the depleted area. Voila. The area you were worried about (which had a measly 5,000 writes left) now holds explorer.exe, which has been updated 12 times in the past 2 years.
Come now, that's like asking whether the designers of an electric car will leave in the gas tank. Of course the controller would have the capability for distributing writes. Why wouldn't it?
There is a fine line between being a cultivated citizen and being someone else's crop. - A. J. Patrick Liszkie
Drive, eh. We are witnessing another one of those anachronism a la TTY in formation.
Flash drives simply don't write the same first bits over and over again. Their firmware is programmed to 'intelligently' spread written data across the entire storage area as fairly as possible.
Between this, massive storage capacity (think: 'dilution') and what will surely be engineering improvements, flash drives should prove to be very reliable.
I for one, welcome out solid state overlords.
Fact: Everything I say is fiction.
I know this music...
!#&*
Does anyone have a good technique about wear leveling at the lower levels? While I'm sure the engineers at Seagate can solve the problem fine, the examples given here of why wear leveling will work --"transferring 2x the capacity of the drive daily gives you 200 years!"-- is a total bullshit argument. That is assuming filling and emptying the entire drive so all sectors get written to. Don't even need wear leveling then.
:)
Also, at the hardware level, how can the wear leveling controller know if the space is free or not. In a file system the entry is just removed from the FAT. So as far as the drive can tell, this data hasn't been 'erased' and cannot be used for wear leveling ?
Again I'm sure Seagate knows what it is doing, but does anyone have any white papers on the lower levels of how this stuff works? I'm an EE so it wont go over my head
It's easier to fight for one's principles than to live up to them.
Can we make sure they're connected via PCI-Express rather than PCI please? PCI is just too slow. I used a solid-state disk that plugged into the PCI bus a while ago and didn't get the expected performance improvements. Good ol' RAM-disks worked tho because they weren't limited by PCI.
Lifetime of the drive is proportional to the ratio of storage capacity to write volume. Higher write volume? Get a bigger drive. Problem solved.
It's not special technology - it's existing technology. That's what happens in a USB thumb drive and it's what will happen in an SSD. We thought of the problem several years ago and made it go away. Distribution of data is not the problem in SSDs. Form factor and driver loading is the problem. But since those are proprietary issues, they're not going to be discussed here by anybody who knows anything.
logfs. It's a log structured filesystem. Perfect for flash.
No weapon in the arsenals of the world is so formidable as the will and moral courage of free men.-Ronald Reagan
If you have enough RAM you never touch swap. Heck, on my Windtunnel G4 I once disabled it. (Things were awesome until the one day I actually needed more than 2GB of RAM. On that day, everything exploded...)
-:sigma.SB
WARN
THERE IS ANOTHER SYSTEM
I'm not sure.
Some vendors apparently have a wear-leveling system that works by knowing enough about FAT32 to shuffle FAT32 files, but has no concept of how to do anything similar for ntfs, ext2fs, hfs...
My blog: http://www.seebs.net/log/ --- My iPhone/iPad app: http://www.seebs.net/seebsfrac/
As late as six months ago, my main machine had 6GiB of storage.
160GiB is plenty for some of us.
-:sigma.SB
WARN
THERE IS ANOTHER SYSTEM
Gigabyte had them. They had support for 4GB of RAM I think, and were accessible thru a SATA 150 interface
It's not about the issue of capacity to wear out a drive in a year, it's about the fact that no one actually writes that much data. 40GB of HD writes is a fairly big amount, particularly if you have atime disabled in your file system(and a flash system will inevitably have this as the case).
No, wear leveling is done at a low enough level that the whole disc is just a giant bag of bits.
Combination - fun iPhone puzzling
Nice to hear that Seagate is going to bring them out by 2008, but drives like that already exist for many years. This is not a new development.
I use one in a media fileserver, for the boot/OS disk. It is made by TRANSCEND, and looks like a largish 40pin connector. It plugs directly in the PATA connector on the motherboard (see, this board still has a PATA connector, it is that old).
They are (sort of) popular in this kind of applications. Building a solid-state system out of PC standard components.
I have also seen adapters that plug in a PATA connector and accept a flash memory card for storage.
Some of those "Linux router" and "plug and play raid storage server" projects use this kind of device as their boot/OS disk. You can even buy them preloaded with software.
A variant of your post has cropped up, *multiple times*, in every single story on every site ever that deals with solid state drives. Hence, redundant.
Well, latest Seagates have MTBF of 1.2 million, which I make to be "only" 136 years...
But I take your point - these numbers are probably unreliable.
Noone gets 136 years from an HDD.
However I suspect the SSD numbers may be less unreliable than the HDD ones,
simply because of a lack of moving parts, and that the degradaton
of flash write cycles is pretty well understood. I've actually
tested a set of flash drives to failure (for a work project - we wanted
to check the manufacturers numbers) and
what we got were certainly within - and often far exceeded - manufacturers
stated MTBF.
We also did some MTBF testing of our own equipment. Mostly it was
putting a large number in an oven and taking it through a series
of temperature cycles over a number of days - to simulate day/night/high load/low load
with the devices at full load. After cycles representing years, we counted the
number of failures (very very small) and got an MTBF figure
"in excess of 300,000 hours". Its an inexact science with solid state
electronics - probably more so with HDDs and moving parts.
"The performance gain for using solid state hard drives for any database, not only MySQL, is a "no argument". Buying performance like this does require some cost analysis. You balance performance with cost. Not everyone buys fiber channel even if it buys performance." - by Lindsay Lohan (847467) on Thursday August 23, @08:28PM (#20338141)
That's true - I wrote the article for that who's ideas & principals were featured alongside one from Mr. John Enck of "Windows IT Pro Magazine" (where he is a technical editor) for EEC Systems (now SuperSpeed.com) when I was improving another of their products, SuperCache, by up to 40% redesigning its parameterization algorithms & configuration via an add on program (whose code was in Delphi, but later ported to C++ & is today integrated into the product itself, not an addon any longer)...
The article I wrote dealt with using a software RAMDISK of theirs called "SuperDisk", which mirrors data back to a backing HDD @ periodic intervals via its driver, & the article I wrote's ideas took them to a finalist position @ Microsoft Tech Ed 2000/2001 iirc, as a finalist in the HARDEST category there was there:
BOOSTING SQLSERVER PERFORMANCE (used on, iirc, SQLServer 6.5 & below, @ that time in those days) & dealt in placing its logging &/or tables devices onto a RAMDRIVE (which nowadays is basically done in SQLServer itself using its OWN "memory mgt." & running its device in RAM afaik)...
The technique, works.
APK
P.S.=> I own a SOLID-STATE RamDisk by CENATEK, called the "rocketdrive", & I use it this way for 'home use':
CENATEK ROCKET DRIVE SSD I have is 4gb max size, BUT, can be striped with 4 others like it into 1 huge 16gb one!
It's PCI 2.2 (132mb/sec) + PC133 SDRAM & has a backing powersupply (which I run thru an APC UPS for stability/uptime/security, though I don't place any critical data on it)...
I am STILL using it to this day, almost 5 years later! No moving parts to wear out...
HOWEVER? Let's see a Flash RAM last that long, lol...
E.G./I.E.->
Mine's running strong & doing TONS of repeating writes in a constant stream, by housing my:
----
1.) The first 1gb partition, NON NTFS COMPRESSED, has Pagefile.sys on it on an NTFS 4096k sector formatted (to match memory transfer sizes used on disk cache, reads, & read ahead for maxing speed).
2.) The second 1gb partition, NOW NTFS COMPRESSED & using 4096kb sector formatting to match memory read + read ahead mechanisms @ the filesystem driver level (& EVEN BLOCK DEVICE DISK DRIVER LEVEL if I wanted by using SuperCache on it as well)
The 2nd partition, & what I use it for, is for webpage caching, temp ops, logging, & housing my %comspec% cmd.exe... & it works in that setup, for better performance.
----
(Many bennies to this 2 part approach above alone, read on)
HOW? Well, look @ the datatypes I have on it largely:
MOSTLY TEXT BASED DATA, & on a COMPRESSED DISK!
THUS, it gets WAY over 2:1 compression even & by FAR with that kind of data!
Oh, not done, yet... Also, I gain in speed, YET AGAIN, also because smaller files on disk exist this way as well!
(Due to compression via NTFS filesystem drivers is why (& today's speed of RAM & CPU's way, WAY even offsets the decompression in memory speed, & on disk writes too, only complimenting its read speeds & access speeds more))...
By the gain here, it's NOT only doubling in storage on the Ramdrive/SSD, but because the files are tinier in NTFS compressed form, they READ UP FASTER because they are smaller first of all!
(Also again complimenting the 4096k sector formatted gain you get on READS, because they read up like that into RAM, & I also use that on pagefile.sys, it is read in those increments & speed of access/seek that is 0ms & into the nano-second spee
Vista has its "ReadyBoost" feature which AFAIK is pretty much like using a USB flash drive as a kind of specialized swap file.
Readyboost is more akin to another layer of disk/filesystem cache, rather than VM/swap.
When I first bought a hard drive it had less capacity than all the RAM on this computer, and it was a big drive. The salesperson laughed at me thinking there was no way to fill it. And I paid more for the drive than I paid for this RAM by a factor of 2 or 3.
I find that difficult to believe. When ?
(& yes, faster ones like the Gigabyte IRAM (SATA 150 bus & DDR2 RAM), & the upcoming DDRDrive X1 (PCI-x based bus & DDR RAM exist in these 2 units vs. mine, since mine is older)??
You can stripe/span them into 1 huge 16gb unit, since each board on the CENATEK "RocketDrive" I use max out @ 4gb RAM per board...
Writes aren't any slower than on a std. HDD with the type I use!
AND, in fact, I would say that, overall, they are faster since the files are found many orders of magnitude faster than they will be on std. mechanical HDD's...
Moving logs (as I do onto a partition on my SSD, 2nd one, NTFS compressed making the files smaller & faster to readup from it no less) onto it with other things (temp ops, logging, webpage caching) helps speed in ANOTHER way:
It removes the burden on my MAIN C:\ drive, by removing paging operations from it (1st partition on my SSD is for pagefile.sys @ 1gb uncompressed NTFS, & using 4096kb formatted sectors, since this matches the read + read ahead mechanisms of the filesystem driver & in the size of chunks it reads them in with paging too) AND, as you mentioned?
Moving logging, webpage caching, & resetting my systemwide + user level %Temp% & %Tmp% environment variables to it for temp ops??
My main program & OS bearing disk no longer deals with contentions from paging, temp ops, logging, or webpage caches interfering with program & data loadspeeds... making them, faster!
Using NTFS compression on the 1gb partition for temp ops, logging, & webpage caches also literally more than DOUBLES its capacity as well because much of this data is TEXT BASED, & compresses far better than 2:1 ratios even, & the files are tinier to read up from disk, getting into memory faster too!
Also, by moving temp ops, & webpage caching ESPECIALLY to another drive, I avoid fragmentation & clutter they might cause on my main disk as well... increasing speed all the more.
APK
P.S.=> You MIGHT be tempted to say, "NTFS COMPRESSION SLOWS IT DOWN ON WRITES" & it does, this is true... especially for logging! However, today's CPU's & RAM are SO FAST? This is offset imo @ least, by the gains I make using it, noted above (for logging? I can MORE THAN DOUBLE what I can store there, AND access it 1000's of times faster than on std. mechanical disks too)... apk
Yes, the Newton used "soups" (essentially simple databases) for data storage instead of a filesystem, but I never had the impression that this was because of flash memory. I believe it had more to do with reducing the size and complexity of applications by having long-term storage use the same data format as heap storage. On the Newton, you didn't have to "format" data when storing it to a soup or "parse" it when reading it. As far as applications were concerned, soups stored associative arrays just like the rest of NewtonScript did.
Soups were stored in a flat namespace rather than a hierarchical one, largely because the Newton was designed as a single-user device where each application worked with a single data set. The flat namespace and standard data format encouraged data sharing between applications: it was trivially easy to lookup a name in the address book, for example.
The original Palm OS also used "databases" in a flat namespace rather than a filesystem, and for largely the same reasons. Again, I don't think that flash memory had much to do with it. In fact, when Palm OS started supporting external flash memory cards, they used the FAT filesystem (for compatibility with digital cameras and desktop card readers) rather than porting their existing database format.
Form factor and driver loading is the problem.
You can easilly buy wiring/form factor adaptors that take a CF card and fit into a laptop drive bay (though admittedly laptops moving to sata will make this harder). You can then install any OS you like (space permitting but you can easilly buy 16 gigabyte CF cards which should be enough for XP and I suspect if you look arround you can buy ones big enough for vista) with no driver issues at all since as far as the motherboard is concerned it is a standard IDE drive.
note: i'm known as plugwash most places but i screwd up registering that here somehow in the past and now can't register
Thanks - did not help my karma, though!
Still, got loads to burn, so 'troll' this as well, you sad fucks!
numbers aside, what's the warranty on the "drive" ? The warranty period is what companies think is the "least" performance duration. So no need to worry for a year or two ;)
I don't think you realize how big 1 million really is. It's big. And it's for a bit, that once it does die, gets marked and the data moved.
That's a fairly naive thing to say. First of all, the current drives are probably not on the order of 300GB - I believe the Toshiba notebooks are 60GB. So you've taken your "at least two or more years" and dropped it by a factor of 5. 2ndly, what's involved in upgrading a flash drive? The current hard drives in notebooks are easy enough to remove, but I've never replaced one because a) I haven't had one fail and b) reloading a new drive sucks. Lastly, when memory addresses fail, is it graceful or do you lose data silently? If it's the latter, it's every bit as bad as the floating-point bug. If I have to replace my hard drive every couple years, that just might be enough to call it a step backwards.
Let's throw some real numbers into this discussion. I am currently typing this on a MacBook Pro, with a 160GB mechanical hard disk drive and 2GB of RAM. My last reboot was 19 days, 21 hour and 40 minutes ago. Since then, I have written 69.55GB of data to the disk (and read 126.92GB, although that includes playing some DVDs). Let's break these numbers down:
Uptime in seconds: 1,719,600 seconds
Disk writes: 72,928,460.8 kilobytes.
Average write speed: 42KB/s
Time to write the entire disk at this speed: 3,955,945.37 seconds (45.7 days).
Time taken to get through 10,000 write cycles at this rate: 1250 years.
If my hard drive lasts 1250 years, then I will be very happy.
I am TheRaven on Soylent News
Hard disks are block devices too. If you change one bit on a hard disk, you have to write out the whole block again. In most current drives the block size is 512 bytes, in some newer ones it is 4096 bytes.
I am TheRaven on Soylent News
In another post, I worked out that my hard drive, over the last 20 days, has seen an average write speed of 42KB/s (a lot of the time it's writing nothing, of course). At this speed, with perfect wear levelling, a flash drive will last 1250 years if it has 10,000 rewrite cycles (for a 160GB disk; I was comparing it to my current hard disk). With one million rewrite cycles, this comes to 125,000 years. Even with highly sub-optimal wear levelling, the flash drive should last a lot longer than any hard disk I've owned.
I am TheRaven on Soylent News
My last laptop had an 80GB disk, and it was the first hard disk I have not felt was too small by the time I replaced it. My current one has 160GB. I have around 50GB free, and I keep copies of a few video editing projects on this disk that on my previous machine were consigned to an external drive. 160GB is probably going to be big enough for a portable for quite some time. As network bandwidth increases, I'd rather keep important data on a redundant NAS somewhere and only use the local disk as a cache.
I am TheRaven on Soylent News
I would be curious to see numbers on how many people actually use that feature. I have Vista on my machine at work and every time I pop in my usb drive ReadyBoost always pops up as an option and I ignore it.
Hell, if you have a system capable of running Vista at all, it's probably a decent box anyway.
My sig sucks.
Couldn't that problem be at least mitigated by OS-level write caching?
Nobody else has this sig.
Except what that analysis conveniently ignores is that shuffling around the static stored content to ensure even wear from the frequently rewritten content really destroys the performance. Especially when defragmenting so single files are stored more within single blocks readable in bursts, rather than fragmented around many blocks and needing to be filtered for the required bits. And all that shuffling further wears the drive.
Now I don't want to hear any of this "Flash is just like RAM/HD" BS. It's a fantasy.
--
make install -not war
Don't put your swapfile on it, or you will be in for some fun.
Death is life's great reward. R. Hoek
Or maybe another option? Why not have a flash-based hard drive as the primary and a CF/SD slot for the paging and temp partition?
Drop in a 4GB CF or SanDisk card that is used for swap and temp that also has wear leveling and some form of SMART to identify failures. Then if this card starts to wear out, you pop it out and replace it with another one.
No need to have a one-disk-does-all storage solution on a machine.
Probably about the same time as I bought my first hard drive to build my first "home grown" PC - 1995 or thereabouts - before that I never purchased a hard drive as my PC's came "all put together" from other family memebers who had their geek credits long before I did. The system I built that fine spring weekend in 1995 had a 800MB hard drive that I paid the beautiful price of $389 for - the CPU was a 486 dx2 66Mhz that ran me somewhere around $179 including mobo, it had a 4x CD-ROM drive, and 8MB of glorious RAM.
My current desktop PC build (put together 18 mos ago) has 2GB of RAM that ran me somewhere around $130. So - roughly 3 times as much RAM as that first PC had in hard drive space - for roughly 1/3 the price.
As for the solid state hard drives for PC's - especially laptops - there's no reason the drive couldn't include a small NiMH battery, and somewhere around 1GB of standard RAM as part of its controller circuitry - along with its 60, 100, 120, 180, or however many GB of flash RAM storage. It might add $30 or $40 to the price - but it would make for an amazing drive with no worries about write cycles - that RAM could be used as an intelligent write cache for the drive - with the micro-capacity battery included just enough to keep the controller and RAM working for as long as it takes to write that 1GB of cache back to storage. If you figure those writes could be made in 10 seconds, a battery with 4 or 5 minutes of capacity would be plenty. The controller architecture could be smart enough to only write "aged data" from the on-board cache chip. If your drive is "thrashing" due to swap file or whatever - those writes would all be taking place in RAM and the read-backs would come out of there rather than from the flash - only once the blocks of RAM aged a certain amount would they then be written back to the flash RAM on the drive. It could be a "percent time" or a "fixed time" setup - say for example every 500ms the RAM is "aged" and a check is made to see which blocks are the oldest - at that point the oldest 5% of used cache RAM blocks get written back into storage and the cache is "free". If there's no writes at all to the disk, in 10 seconds everything is back on flash and the RAM cache is free. If you have a program swapping in and out constantly because you're too cheap to buy enough system RAM, it's actually swapping to RAM on the drive - slightly slower than memory on your FSB, so there's a small performance hit, but much faster than if it were writing back and forth to Flash, and little to no wear on your flash takes place. If the laptop goes to sleep/standby/hibernation/dead battery/power loss - the small battery in the controller maintains the drive long enough to finish all the writes back. There's other details to be worked out - how to handle writes bigger than the RAM chip on board the controller (pass-through to the drive flash, etc) - but that's for the hardware engineers to figure out - I can't do all their work for them. It's the idea of a large enough RAM cache on board a flash drive to eliminate writing "often written data" that is what should happen here.
And to think - there's almost as much L2 Cache on my CPU now as there was hard drive space my first "IBM PC" computer I ever used (not built) had (4MB of on die L2 cache now, VS a 5MB Corvus drive in my 8088 - circa 1983).
I have faith we will see storage drives with Petabyte capacities around 2020 - from 1983 to 1996 (13yrs) went from 5MB to 5GB - I'm pretty sure by 2009 we'll see 5TB drives as we are already seeing 1TB drives in 2007. 13 years should see us at 1000x today's capacity - 1 petabyte - and we'll be talking about using terabyte flash and having 2 terabytes of RAM in our PC. Hmmm... Microsoft better really add some bloat to the next version of its OS to drive that demand!!!!!!!!!! That's even more than Vista with a service pack will use!
Ok enough going on topic and off topic. I'm all for solid state drives. Less power consumption. Faster access. Quiet. Cool. Extremely impact sho
Hi guys, I am a little new to this stuff so bear with me on this one. If I understand the tech correctly, an SSD drive would eliminate the need for swap space under windows right? would the same hold true for Linux? Would an SSD work under the current kernel - I thought it just worked like a hard drive?
Bizzare. Seems like for years, the tuning idea is find the biggest sector size you can live with. Benchmark with different sector sizes (going up and up) and find the diminishing return. This because seeks are slow and reads are fast.
I now benefit from SMALLER sector sizes. Reads are slow and seeks are fast. So that means I need to tune my filesystems and databases DOWNWARD to, say, 128 byte blocks. Or 64 bytes blocks? CPU overhead of 10 times the number of reads is probably very little compared to that tedious pull time.
Can somebody with a math bent try to quantify this?
>>How long of a warranty will these have? Doesn't flash memory break down after a good number of rewrites?
e _frm/thread/136aad9133d01bb7/9c5bce4697be4e1f?lnk= st&q=&rnum=8#9c5bce4697be4e1f
But it's better than having to park my hard drive heads every time before I shut down. Sometimes I forget, and then that data is corrupt. Maybe one day Hard Drives will park themselves at shutdown.
Reference;
http://groups.google.com/group/net.micro.pc/brows
(Tone:Sarcastic/Funny)
This wear leveling controller may help to more evenly distribute the read/writes, so that the drive can last longer, but it is also can be a boon to the folks into forensics or stealing data from "wiped" drives.
Since each write will be allocated to a different section, one does not really know if he is overwriting information that should be declassified. This may make it easier to recover information from the drive. See Forensic Data Recovery from Flash Memory by Marcel Breeuwsma, Martien de Jongh, Coert Klaver, Ronald van der Knijff and Mark Roeloffs.
you are correct. under your disk writing pattern you will not need to worry about this issue. And you maybe in the 80% normal usage pattern. however i feel that both arguments are valid.
under normal usage you will not have a problem.
under high usage you will have to replace a drive once a year.
#!/usr/bin/perl my $is_fine; my $i = 0; while ($i++ 1000000){ $is_fine = hash($is_fine); } print (($is_fine) ? "Wow! Everything _is_ fine" : "LIAR!!!!!!");
Good reply from AC, just to add to that.
1. In my experience, flash memory can sometimes fail totally. This may be due to it being often removable, and accessed in rather non-robust ways, (USB ports, card readers). Hence (presumably) gets nuked by static etc. My attempts at recovering such 'dead' flash devices have not been great, so far. When it's dead, it's dead...even re-format does not work sometimes.
Presumably, internal flash 'disk drive replacements' would be rather more robust.
2. When flash drives first came out, 'classical' data-recovery tools seem to have difficulty recovering from acidental deletes and formats etc., since they seemed (I'm not an expert) to be looking for HDD-like behaviour. I remember reading an interesting paper long ago about the consequences of 'random walk' data storage for recovery... Since then, things have improved, and a lot of tools claim to be/are able to recover data from flash. Of course, I never need these, since I have good backups, ahem.
BTW, I was recently at a client site (for once without my PC and DVDs, CDs, flash drives etc. stuffed with tools) when the sales manager wiped his hard disk. Their in-house IT support was - as usual - no help. I download one of my fav. simple tools,
http://www.snapfiles.com/get/restoration.html
ran it from a USB, copied the undeleted files to a USB HDD and bingo! Another happy customer.
Check it out - if the PC boots (into windoz) it does the job...
how big were the chips? How big are the chips now?
Do the math indeed.
I have filled my hard disks and they're always sitting nearly full - that will reduce the number of available bits for rewriting. But I don't rewrite every single file. Some files are updated a lot for a span of weeks, but as my focus is on these files the other files just sit. If there's enough Flash memory, files that I know will almost never require rewriting should be made candidates for shifting around. That would make even more bits available for rewriting. Otherwise if the disk is always near full, it will be the free bits ready for rewriting over and over, and these will wear out.
Look for reviews that test with almost full drives, and speculate a file shifting mechanism may become available.
If the drive is almost empty, even if I swap big time every day, and that would be ultra painful because it seems that if I use more than the available RAM on my Vista, the swapping chews up my computer and makes it so slow that everything crawls - it's like hundreds of Megs are swapped and swapped and swapped for what I don't know - even so, that's about writing 10 times a day, 3650 times a year, which is much much less than 1,000,000. If the disk is an itty bitty 20 Gb, and wear leveling scatters 1 Gb of swap daily, it's an average of 0.1 writes per day for the whole disk and only 36.5 times a year.
I say let me swap on one of these because the rotating disk is a bitch to swap on. But by the time I can get to swap on such a drive, I'll get a computer with more RAM I hope, and I'll try to force that machine to its knees too.
Come to think of it, if the swap if fast, what will happen to the RAM market? Some people don't need gazigabytes and if the hard drive is a cheap RAM alternative, will RAM makers reduce R&D? Also, having less RAM equals having more swap -- good for making you replace your hard drive. The truth of the matter is that I don't need to use so much RAM at any one time. If I'm typing that takes very little memory. However, I manipulate the occasional image on MS Paint and have noticed that on the biggest files more than 1 Gb of RAM is consumed to work on a 100 Mb file. Even this little RAM eater would be ok with swapping to a Flash memory. Having 2 Gb of RAM with such a fast swap would be good enough for most people as long as the swap space expands with technology and is significantly less expensive. We may also hope that more R&D will be spent on Flash to remove the rewrite limit as well as speed up access.
Know your pads. One time pad: good for cryptography. Two timing pad: where to take your mistress.
I hope this development will help lower the price some more. Assuming we have crappy 10,000 write limits, we could write 40GB to the drive every day for 10,000 days, or 27 years, before failing is an issue. I agree it'd take a long time, but there's a significant chance it'd last more or less than 10K days. *Murphy's law says less.
As the flash begins to fail - your total SSD capacity decreases; you'd be writing multiple times to some parts, per day. (writing 40GB to the remaining 20GB would be two writes) Which would make the remainder fail that much faster.
A bit wouldn't be the smallest unit internally - it'd be some chunk of bits that would have a part of itself dedicated to an error correcting code - when that failed the whole chunk would be marked bad.
If we call that internal unit a block of 32 bits (4 bytes) - then 40GB/4=10G blocks or 10 billion total flash failures possible for 40 billion bytes.
That's where the math starts to get painful, but I think it's like calculating interest over time. Just estimating I'd say 25% more or less than 27 years +/- 6.75 years. (but I'm probably FoC here)
Anyway, so long as it had great performance and I didn't have to pay more than $200 for the drive I'd be happy even if it only lasted a decade.
If I ever have the time and money, I think it'd be fun to build my own SSD. (or an external ram cache for a HDD)
With 10% or more of a 10GB SSD cached with RAM, it'd be blazing fast and that small drive's useful life might meet or exceed it's MTBF.
My ideal 10GB SSD would be about $80 for flash (5 - 2GB SD cards @ $16), 1GB of old ram $40, and the additional parts (CPLD, SD slots, pcb, etc) would be roughly $60... $180 total and capable of 100MB/s (burst read & cached).
Gigabyte makes a neat device using battery backed up RAM plugged into PCI slot that has a SATA hard drive interface wire hanging out of it. Since the whole thing is plugged into your SATA interface and presented to the OS as a hard drive, it performs like a SATA drive that always has the next bits in cache. i.e. fast and with no fragmentation or write limits.
Very popular to serve mailq queue directories. i.e. tons of reads/writes and ludicrously small seek times, and static so a reboot or a hard-crash doesn't wipe it out. 4 GB is plenty big enough for a mailq or even a whole OS image. And this device with 2 GB chips (8 GB capacity) makes a really good LDAP or other small database drive, or a storage area for larger database indexes... where the real data is stored on slower traditional arrays, and this device merely stores the indexes. Works great if your indexes are easy to rebuild.
Two problems here with your idea:
1: Paging is the most disk intensive part of computing, and therefore offers the greatest performance increase in going to an SSD. To just use the SSD for ordinary program storage negates much of the advantage of having a faster drive.
2: The idea of an SSD is to get away from a mechanical drive due to power requirements, its much more fragile nature, and slower speeds. You want to include BOTH.
"It's the height of ridiculousness to say for those 9 lines you get hundreds of millions."
It sounds like you're hardware problems are a result of an ID10T error. Read a book or something.
Sounds similar to to the life span of a hard disk. Do you think you could take a consumer hard disk, max it out every day, and get it to last more than one year?
Gamingmuseum.com: Give your 3D accelerator a rest.
See:r amdisk/
http://www.bit-tech.net/news/2005/05/31/gigabyte_
.
== WolfriderV6 == I'm willing to admit that *I just might* be wrong... Are you??
--Defragging could still be useful, even on flash drives. If all your files are kept sequential, there's no need to skip around looking for the next filepart.
--Won't save HUGE amounts of time, or may not even be very noticeable - but every little bit (pun intended) helps...
.
== WolfriderV6 == I'm willing to admit that *I just might* be wrong... Are you??
When flash fails, it fails to write. You can still read the data.
My Freakin Blog
...has anyone managed to build a fanless media extension box that does 1080p decoding (H.264, VC-1, MPEG2) including normal trailers off the web etc., not just HD DVD/Blu-Ray? I've been thinking of putting one together, with a fanless CPU, fanless graphics card and CF drive + network for bulk data access. Anyone made a project like that which works well?
Live today, because you never know what tomorrow brings
Every manufacturer's firmware does wear levelling, and none of them know or care about filesystems. Its simply moving frequently written logical addresses around to different physical addresses. There is nothing complex, difficult or magical about it.
--Thayat's some good thinkin' there, Clem. ;-) Makes sense.
;-)
--HOWEVER, if people would just put more actual RAM in their boxes, lots of swap space becomes redundant. (If you have 4GB+ RAM and a 64-bit host OS, you can prolly get away with 64-512MB of Swap.)
--Remember that the disk-swapping code was originally written **because of** lack of sufficient RAM...
.
== WolfriderV6 == I'm willing to admit that *I just might* be wrong... Are you??
as this drives dont have mobile parts that can be damaged, engineers are currently furiously working on new and effective ways you can break/damage this flash devices.
Dad, why do they call them disks?
Dinomite.net
Thanks for the info, and the link. Never hurts to have one more tool in the kit.
I haven't had a flash unit fail yet myself, but I don't use 'em a lot (mainly just my camera's xD card). It does sometimes play dead and pretend it wants reformatting, but I think that's disagreement between itself and XP's notion of a USB1.x driver (old machine). I ignore its complaints, and after a couple more access attempts the problem goes away.
Someone did torture tests a couple years ago and found that most consumer-grade flash units survive everything you can throw at them physically (microwaving, hammering, one even survived having a nail driven through it) but I don't recall if they tested for static shocks -- which are more likely to be encountered in Everyday Life[tm].
The "death without warning" issue is what would bother me the most. My W.D. HDs invariably have given me months of notice (sometimes years) before going totally tits-up; there's ALWAYS time for one final backup. Works-today, Dead-tomorrow is more incentive for backups, yeah, but there's always the backup you didn't quite get to yet....
On the plus side, flash HDs would lend themselves handily to having multiple removable drives -- don't even need to redesign the PC case, just make a unit that fits in a standard 5" drive bay, that could take a dozen or so, the size of current thumb drives, plugged in side by side. That would also help with the size vs cost issue -- a dozen 100GB drives being usually as good a solution as one 1000GB drive, and much safer from a data recovery standpoint, plus you could start with one and add as needed very easily. With an IDE or SATA adapter, it could even be used with an old machine.
~REZ~ #43301. Who'd fake being me anyway?
Using compactflash to IDE adapters is inexpensive and readily available, but the real solid state drives perform much better, last much longer, and are directly comparable to their mechanical counterparts.
Actually, solid-state hard drives with SCSI and IDE interfaces already exist (BitMicro and Sandisk both have them, Sandisk having bought one of the original SSD vendors M-Systems). There are also some drives with SATA 1.5 GB/s interfaces, but I'm personally waiting for the next generation of drives (which is currently being developed) that has SATA 3.0Gb/s interfaces, and are standard sizes (3.5"). Those are going to be expensive at first, but they'll be amazing!
You can already swap out a 2.5" SATA laptop mechanical drive with a 2.5" SATA solid-state one, for example. They just cost too much (small drives cost a few thousand!).
SixD
My first hard drive was a Micropolis 1300 20Mb $1200 Employee discount.
I put it on my S-100 system for which I had to write the BIOS in assembler.
That was about 1985
Nothing is foolproof, fools are too ingenious. - Murphy
That's not what I meant. The form factor problem is the one of being able to fit enough chips, passives and PCB into a 1.8" or 2.5" enclosure and still be able to have a decent amount of storage (and not burn up).
In your example, yes, CF will get the job done for now, but flash transfer rates are increasing rapidly, latencies are decreasing rapidly and we should be seeing SSDs by the end of next year that contain purpose-built components designed for high speed, low parasitic loading and low latencies. Even now a 32GB or 64GB SATA SSD is a much more elegant solution than a bunch of CF cards plugged into adapters. And in the case of a notebook, a 64GB drive can easily be the only mass storage device you need.
So if we randomly choose a sector and write to it, the wear over large numbers of writes will be uniform over all sectors.
Ahem.
FILE SYSTEMS DO NOT CHOOSE FREAKING RANDOM SECTORS TO WRITE TO.
If you examine the real distribution of writes to a disk with a real filesystem under real usage scenarios, you will find that some sectors are overwritten many orders of magnitude than others. Filesystems have metadata structures. These structures MUST be updated every time files change. Ergo, they get overwritten. A lot.
For example, many filesystems use some form of a bitmap to track which sectors are in use. Every time a sector is allocated or deallocated, the bitmap must be changed. This means the sectors storing the bitmap are overwritten at least once every time any file is created, deleted, or changes length.
So please. Stop with the ignorant analysis which assumes a perfectly random distribution of writes over the disk's sector address range. It's dumb. No real world FS behaves like that.
(There are some filesystems designed to work better with flash. But none of them are likely to be well suited to what you want to do with a personal computer, as the existing flash filesystems are designed for embedded systems.)
Time taken to get through 10,000 write cycles at this rate: 1250 years.
WRONG.
You see, the important metric isn't the amount of data you write. It's the number of times any one sector is written to.
For example: UNIX-like operating systems such as MacOS X record the last time a file was accessed. The particular sector storing the access time for a specific file must be written every time that file is accessed. Betcha there's at least one file the OS has accessed 10k times during 1,719,600 seconds of uptime... that's only 1 access per ~172 seconds.
Your argument is flawed. Sure, if you have a 50GB disk and you can rewrite it a million times, you have a total write capacity of 50 petabyte, which indeed is a lot longer than a moving parts driven hard-disk's could do in its lifespan. But this is taken into account the distribution of your writes is 100% perfect and in normal usage patterns some bits will be "hit harder" than others (ie. some data *has* to be rewritten by a filesystem). Also, you seem to assume that bits can just be moved around everywhere at the hardware level, so your data ends up in random order on the disk. Think about it, no filesystem nor the hardware could handle putting all those pieces back again (not without a lot of meta-data, which would have to be stored somewhere else).
Granted, I'm not saying that solid state disks are crap, probably with reasonable usage you will get a decent life out of them. And I'm sure eventually that's what we will all be using.
Why do I keep seeing people say not to run sawp on the CF/SSD? Why is there some 'wear leveling' thing? Obvisouly, I'm not understanding something, but in my ignorance I assume that a solid state device, save for heat, should last forever. So if someone would be so kind as to explain..? I'd love to run my swap on SSD if it meant my system would run faster! The biggest bottleneck I currently have is hard drive throughput with mechanical drives.
Knowledge is like ignorance.. too much can be just as bad as not enough.
"This will be pretty similar to how flat screens took over the market from CRT monitors."
Not for those who give a damn about color accuracy or running at something other than native resolution.
Your post raises a good issue, but misses part of the point. Yes the drives are capable of much faster writing than the GP considered, but no one actually comes close to maxing out their hard drive bandwidth on a continuous basis. Even relying heavily on swap files, (and any high use computer should have adequate RAM not too, while computers than don't have plenty of RAM seldom seldom see demanding use), there should be a lot less data than that being written.
However, the concern that a person should draw from this is how does swap file usage affect this. Sectors assigned to virtual memory could get written to far more often than other sectors. What rate? I don't know, but if memory life is a problem, this is where it would show up.
Fortunately, the swap file is just temporary data. If it goes caput, the original data is theoretically still safe and retrievable in other sectors.
True, its too expensive to replace SATA. Nevermind performance compares. Just run your OS on SSD and keep your data seperate. It'll make your computer much faster, won't cost much, yet makes your setup also more flexible. I wanted to run an OS on an USB stick, but alas it was bad supported by my OS so I ended up with a SSD/PATA. Ironically, those have to reside inside the case whereas a USB stick is plugged on the outside. The latter is in some setups more flexible: have a backup copy or 1:1 copy of USB easily available for the server. If one needs replacement its sticking out and in a key (and pressing a button). Even the secretary can do such simple tasks.
WE DON'T NEED NO BLOG CONTROL.
My current desktop PC build (put together 18 mos ago) has 2GB of RAM that ran me somewhere around $130. So - roughly 3 times as much RAM as that first PC had in hard drive space - for roughly 1/3 the price.
Misinterpretation on my part, I though the OP was saying when he got his first hard disk, his machine's RAM size was larger than it.
An idea of mine way, WAY back then in 1997 (a decade ago, seems hard to believe now it was that long because it seems like yesterday, lol) took them to a finalist position @ Microsoft Tech Ed 2000 & 2001, & 2002 iirc!
(In the hardest category there - SQLServer Performance Enhancement, & the ideas where to place devices DB engines use into RAM on ramdrives, then a mirroring back to backing HDD software-based device driver driven one, called SuperDisk that SuperSpeed.com produced & retailed of unlimited sizing (to RAM amount present)).
I was doing a contract job for them improving their SuperCache product by up to 40% in performance, & decided to write them on my "theories" (some of which you touch on in fact) as to ramdisks being used to enhance performance... & even got into Windows NT magazine as well back in 1996 for this stuff, which was WAY cool (forerunner of today's
See some of HOW I personally use an SSD, @ home (logging as you mention is just one & more things that might interest you as well), are here:
http://hardware.slashdot.org/comments.pl?sid=2782
(Enjoy the read!)
APK
P.S.=> I also wrote up an article that CENATEK later featured on the front page of their site, when I evaluated an SSD (the one I bought, & it is still running STRONG TO THIS DAY mind you (let's see a FLASH based disk go 5++ years) in the CENATEK "RocketDrive"), I applied the SAME BASE IDEAS to the SSD from the mirroring unlimited sized software-based RamDisk from EEC ideas mentioned above, but now to hardware based ones (they are great)... better/faster ones exist today, using faster busses than mine (PCI 2.2 132mb/sec max) & faster RAM too (PC-133 SDRAM on mine)...
E.G.-> Gigabyte IRAM SATA I 150mb/sec max, & DDR2 RAM, or hopefully soon (I want one is why I state that), the DDRDrive X1... PCI-x & DDR ram! That last one has WAY larger transfer/burst than even the IRAM... & certainly more than mine does! apk
The PCI slot is just for power, there is no communication over the PCI slot. And the battery backup is good for some hours without power. But overall, neat device.