The 1 Terabyte SSD Arrives

An anonymous reader writes "Over recent years Solid State Drives (SSDs) have moved from luxury to affordable additions to one's PC, but mechanical hard drives are still king when it comes to capacity. That was until the revamped Colossus LT series Solid State Drive came along this week. With up to 1TB, the drive offers offers massive storage capacities of the level normally not seen in SSDs. While 1TB of SSD space hits right at the heart of the traditional hard disk market, it comes at a high price — at around $4,000 for the 1TB model, these drives are in the realm of aspirational rather than practical."

Comments

I'll wait a while.

[carlhaagen]

I have a handful of friends who adopted Intel's latest G2 X25-m models at their release. With new firmware, they are all still reporting notably reduced performance over time. Everyone knows what causes it, it is entirely understandable given the storage technology in question, but that doesn't make it any less of a drag. I'll wait and see how things change before doing the switch.

Re:I'll wait a while.

[Nyder]

I have a handful of friends who adopted Intel's latest G2 X25-m models at their release. With new firmware, they are all still reporting notably reduced performance over time. Everyone knows what causes it, it is entirely understandable given the storage technology in question, but that doesn't make it any less of a drag. I'll wait and see how things change before doing the switch.

Everyone knows what causes it huh?

Sorry, that's a really stupid assumption, because, I don't know what causes it.

So I guess not everyone knows what causes it.

Re:I'll wait a while.

[zkrige]

Just defrag it :P

Re:I'll wait a while.

[FredMastro]

I agree with you. Because, I don't follow the SSD market so close. I hear it's a performance gain but slight at that. I've not heard about slower performance over time. Would like to know more. So yes, I don't think an assumption should be made. Maybe, an assumption and then a reference link for those that don't.

Re:I'll wait a while.

[carlhaagen]

They are Windows 7 and Linux users. TRIM seems to just ameliorate temporary.

Re:I'll wait a while.

[carlhaagen]

It's not a file system fragmentation issue. Also, as there is no real seek time, fs fragmentation is a non-problem on SSDs, even with slobs like NTFS/FAT.

Re:I'll wait a while.

[aicrules]

I think maybe it's something like this: http://www.brighthub.com/computing/hardware/articles/43400.aspx

But since he's so mysterious about it, perhaps it's not.

Re:I'll wait a while.

[poetmatt]

at the moment, the only thing that might not support TRIM is OSX, which I suspect does. Linux and windows do.

Re:I'll wait a while.

[Colonel Korn]

They are Windows 7 and Linux users. TRIM seems to just ameliorate temporary.

Your friends aren't benchmarking. Welcome to subjective perceptions. As quantitative data has proven conclusively (see anandtech.com, pcper.com, etc.), TRIM does truly prevent lost performance over time.

Re:I'll wait a while.

[Joce640k]

Ummmm, SSDs map disk sectors to physical flash cells dynamically as part of wear levelling.

Defragging will probably make it much worse.

Re:I'll wait a while.

[hoggoth]

That article was low on logic and common sense.

The article's take-away that SSDs slow down over time may be right, however the reasoning behind the explanations doesn't even make sense.

> "Because they have a two-part write/erase cycle, unlike the single write cycle of mechanical hard drives, they wear out at least twice as fast as their spinning counterparts."

Umm, what? SSD writes are done in two stages, yes, but that has absolutely nothing to do with the way a traditional hard drive does writes. So how could you say SSD's wear out "twice" as fast as traditional drives because they have to write twice? It could be that an SSD could write a thousands times more or a thousand times less than a traditional hard drive before wearing out because they are completely different technologies.

> "This isn't helped by the architecture of most SSDs. Usually, data is laid down within a block of available memory, meaning that it might not take up all the available space--yet will still write to all of it"

Does the author think traditional hard drives write to byte-addressable boundaries? Hard drives write blocks and sectors too and have wasted slack space at the end of their blocks too.

> "Defragmenting or "defragging" a SSD takes up many write/erase cycles... which shortens the lifetime of an SSD, even if it's also cleaning up the drive."

No, defragging is not cleaning up an SSD drive. There is no reason to defrag an SSD because their is no latency getting to a further sector.

> "it's a delicate balance, how often you should defrag your SSD for optimum performance and lifetime"

How about "NEVER"?

> "Only defrag when necessary!"

Argh!

Re:I'll wait a while.

[hoggoth]

Yes, yes I know:

s/their/there/

Re:I'll wait a while.

[obarthelemy]

google: why do ssd get slower over time. first answer: http://www.anandtech.com/show/2738/8

no comment

Re:I'll wait a while.

[gmack]

That seems written by someone who really has little to no idea how SSD drives work. It should take years to see problems caused by flash wearing out even under intense use.

The actual problem involves the way modern SSD drives write your new data to an unused portion of the disk before erasing the old flash to improve speed. If the drives think they are full then you are stuck waiting for the old blocks to be cleared before you can write your data.

TRIM was added to fix this problem by letting the OS tell the drive when blocks become unused but it only works on very recent drives and new operating systems. You are out of luck on that front if your running XP or a Linux kernel older than 2.6.33 but on the upside the problem only affects write speed.

Re:I'll wait a while.

[gmack]

It is only a slight performance improvement for large files. For large amounts of small files it's a huge gain thanks to the lack of head movement.

I picked up a 32gb SSD drive to handle the OS and apps and left my 1TB drive for movies. The difference in boot times and app load times are very noticeable.

Re:I'll wait a while.

[aicrules]

I should have posted a caveat that I just did a google search and picked the first thing that came up. I guess I could have posted a link to the google search rather than a specific article. In any case, I'm impressed with your rather in-depth analysis of a link not actually part of the Slashdot story.

Re:I'll wait a while.

[gmack]

TRIM was only enabled on kernel 2.6.33. I don't know any distros that ship that version yet so unless your friends are custom compiling their kernels they don't have working TRIM.

Re:I'll wait a while.

[hoggoth]

Oh God no... I RTFA...

Do I get banned from Slashdot? Am I an outcast now?

(PS I was not criticizing you, just the linked to article)

Re:I'll wait a while.

[cynyr]

gentoo is shipping both vanilla and "gentoo patched" 2.6.33 sources. :P

Re:I'll wait a while.

[ArsenneLupin]

"This isn't helped by the architecture of most SSDs. Usually, data is laid down within a block of available memory, meaning that it might not take up all the available space--yet will still write to all of it"

Does the author think traditional hard drives write to byte-addressable boundaries? Hard drives write blocks and sectors too and have wasted slack space at the end of their blocks too.

Yes, but these blocks of memory might be much bigger than sectors on a hard disk.

And filesystem code in operating systems knows about the (small) sectors of disks, and might not be able to cope with the large blocks of SSDs. Meaning that the SSD must be sufficiently smart to read the entire block, change whatever range needs to be changed, and rewrites it. And this might happen lots of times, because the higher level code (filesystems) might not be aware of the issue.

There is no reason to defrag an SSD because their is no latency getting to a further sector.

There is no latency, but defragging may be useful for a different purpose: making sure that each memory block is occupied by as few different files as possible (in order to dampen the effect of the phenomenon outlined above).

Re:I'll wait a while.

[StayFrosty]

The Arch testing repository contains the latest kernel usually within a day or two of release.

Re:I'll wait a while.

[Glonoinha]

Does the author think traditional hard drives write to byte-addressable boundaries? Hard drives write blocks and sectors too and have wasted slack space at the end of their blocks too.

Novell was doing block suballocation as long ago as Netware 5.x, which came out in roughly 1998. It's not a completely invalid assertion.

Other than that, excellent rundown on the issues. Thanks.

Re:I'll wait a while.

[MobyDisk]

I recommend reading the Anandtech SSD articles as they are clearer, more informative, and less alarmist.

Re:I'll wait a while.

[toastar]

I have a handful of friends who adopted Intel's latest G2 X25-m models at their release. With new firmware, they are all still reporting notably reduced performance over time. Everyone knows what causes it, it is entirely understandable given the storage technology in question, but that doesn't make it any less of a drag. I'll wait and see how things change before doing the switch.

Um, You bought a MLC drive and are now complaining about drive wear?
You bought a cheap product. If you would of bought an SLC you wouldn't have the same level of drive wear. I want to say the difference is a factor of 10.

next time get a X25-E.

I'd love to know how the X25-V's are shaping up.

Re:I'll wait a while.

[asdf7890]

What OS are they running? Without TRIM support on both the drive and the OS (Windows 7, Windows 2008 at the moment IIRC, or Linux with a kernel version 2.6.33 or above but I'm not aware of any distros carrying that as standard yet - even the beta of Ubuntu 10.04 is still at 2.6.32) the block fragmentation within the drive will cause write performance degradation over time.

I'm told that writing solid blocks of 0s will cause a drive's controller to mark the block as not needing erase-before-write next time (which is what a TRIM call does) - I'm not sure if this is for all drives or if I was reading about a specific range though (or if the "fact" is true at all! - I'll give it a try on my netbook's SSD if/when write performance becomes an issue). If this theory is correct then erasing free space by writing a file from /dev/zero should help or for a full clear do that first, then image the drive to elsewhere, then zero the entire drive, then copy the image back on. This in theory will result in no split blocks, though I've not tested this theory myself yet. If zeroing blocks does not mark them in the same way as TRIM does then you will only be making the problem worse.

Re:I'll wait a while.

[archangel9]

Are they using an O/S that supports TRIM?

I support TRIM by taking Smirnoff Ice sixers to local house parties.

Re:I'll wait a while.

[Nemyst]

[citation needed]

The wear of defragmentation is never worth the POSSIBLE increase in performance. All manufacturers say it, all tech journalists worth anything say it. Until I'm told by a reputable source (with numbers backing it all up) that defragmentation has a potential benefit for SSDs, I'll listen to the advice given by those who know what they're talking about.

As for filesystem issues, using a modern OS like Windows 7 or recent Linux or OSX releases will mostly take care of that as there have been updates to take advantage of SSDs. Sure, if you're still using Windows 95 with FAT, you may have issues, but anything built in the last few years will take care of that nicely (and continuously improving firmware and drivers further improve performance while diminishing issues).

Re:I'll wait a while.

[Gr8Apes]

I'll still wait a while. With a good HD the OS doesn't lag too much - my boot times are in the sub 30s range even for XP. (about 25s for POST/CD/DVD check, another 25 for the OS to come up)

If I wanted to speed that up - I'd probably add a RAID 1+0 setup and still come out ahead on pricing. (But boot times would probably be slower, since the load time saved would be spent setting up the RAID configuration.)

Since I boot once a month at most, I'll stick with cheap HDs for my desktops for now.

Re:I'll wait a while.

[Alastor187]

TRIM was added to fix this problem by letting the OS tell the drive when blocks become unused but it only works on very recent drives and new operating systems. You are out of luck on that front if your running XP or a Linux kernel older than 2.6.33 but on the upside the problem only affects write speed.

For Intel Drives the new SSD Toolbox supports Trim on older OSes. I have been using it with WinXP...though that probably doesn't help anyone on Linux.

Re:I'll wait a while.

[higuita]

the problem is that the SSD gives you *NO* way to really defrag a file!!

when you tell the SSD to move sector 120 to sector 32 (so it keep close with the begin of a file on sector 31), the SSD will move the data to the next free sector because of the wear levelling.

even if you ask to write on the same sector 120, the SSD WILL MOVE the data to the next free sector.

So defrag a SSD not only is a waste of time, but burns faster the SSD writes cycles and may probably INCREASE the fragmentation (if the free sectors arent continuous)

reading fragmented files on the SSD isnt a problem... writing small fragmented files on a SSD DOES take a performance hit if there is not enough free sectors. TRIM will keep the free sector count high enough so this doesnt became a bigger problem.

Of course, users should not fill up a SSD, as it needs the free space to work well... the more free space, the better :)

Re:I'll wait a while.

[severoon]

I don't get this. Why don't they make the drive so that it periodically moves stuff around, or even a software app that does this, to restore peak performance? In the non-SSD world, this is called defragging...is it such a far-fetched concept that no one's thought of it for SSD? (I assume no...so what's the story?)

Re:I'll wait a while.

[petermgreen]

And this is another reason why trim is important. Without trim the drive does not know what sectors are really in use, only what sectors have been used at some point.

Re:I'll wait a while.

[theskipper]

FYI, the hdparm package includes wiper.sh that can be set up to run every few days. That can be set up to run via cron so the inconvenience isn't too high. Combined with proper partition alignment, my OCZ Vertex has been working flawlessly for a few months now.

Re:I'll wait a while.

[billcopc]

As someone who took the SSD plunge a few months ago, I can tell you that the performance delta entirely depends on two things: your multitasking habits, and the quality of the SSD controller chipset. Some are only good at sequential read/write (like conventional drives), but the better ones are also good at random access, and these are the ones that make your machine zippy. On a spinning platter hard drive, if you have two apps accessing the disk simultaneously, it has to seek back and forth between the two files, reading a little piece each time, and this literally decimates your throughput because the drive spends more time moving the magnetic heads than actually transiting data. This seek time is nil on an SSD, so you regain the full read/write speed, even if you're accessing 100 files in random order. This is what makes the desktop so much faster, because no matter which app you launch, it's always just a microsecond away.

Windows boots in about 10 seconds, apps pop up almost instantly, large compilation jobs finish in a third of the time. I don't run benches but if my boot drive has slowed down in the 4-5 months since I bought it, I haven't noticed at all. I also have four of them in a RAID-0, and I still hit the 700mb/sec writes I've enjoyed since day one. The only detail I still worry about is reliability / longevity. These consumer-grade SSDs are still very new, and we don't yet have any good empirical data on how likely they are to die, or what the real-world wear-out period looks like. I know I beat the crap out of mine, with an inordinate amount of churn these days as I'm ripping thousands of CDs and DVDs back into images for archival, but that's what warranties are for.

Re:I'll wait a while.

[V!NCENT]

FYI: There's a 'bug' in the very design of the SSD that doesn't allow read _and_ write at the same time (IIRC).

There's a firmware update for it so that under certain conditions pieces of software do not completely lock up while waiting for the device but that greatly reduces read and write speed individualy (IIRC).

Re:I'll wait a while.

[V!NCENT]

Wrong! Autists only think of themselves. They assume that what they think, everybody else thinks without communicating while communication is required to get something across.

If somebody says; well everybody know that-... Well maybe not everybody knows.

But the skill in seeking expected balance between common knowledge and needing citations is what many social inapt people lack.

You might want to reconsidder that statement.

Also; getting laid has more to do with "You can't possibly expect me to pick... You?!" anti-social stands than "Hey yeah I heared that last night! I feel so badly for-..." - FAIL!

Re:I'll wait a while.

[obarthelemy]

TRIM. Requires OS and controller support, not available on older SSDS, mostly available on new SSDs and new OSes. Always check, though.

Re:I'll wait a while.

[BikeHelmet]

> "Defragmenting or "defragging" a SSD takes up many write/erase cycles... which shortens the lifetime of an SSD, even if it's also cleaning up the drive."

No, defragging is not cleaning up an SSD drive. There is no reason to defrag an SSD because their is no latency getting to a further sector.

> "it's a delicate balance, how often you should defrag your SSD for optimum performance and lifetime"

How about "NEVER"?

> "Only defrag when necessary!"

Argh!

I see your point - but also consider that NTFS is pretty stupid, and may split a large file into 15000 chunks. There is some performance impact associated with locating all those chunks. It may be on the CPU end rather than SSD end, but it's there. If you want to check how bad it is on your own PC, use a tool like Defraggler - it can tell you how many fragments your worst files are split into.

With a smart filesystem, there would be no reason at all to defragment an SSD.

I feel that even with stupid NTFS, you should avoid defragmenting SSDs.

Re:I'll wait a while.

[severoon]

TRIM doesn't solve the problem, apparently. Why doesn't it just defrag and compact all the files into continous pages on block boundaries? Seems simple enough...

Re:I'll wait a while.

[complete loony]

The block size of flash chips that must be erased all at once is typically larger than the block size of the filesystem. When the filesystem overwrites a logical address, the SSD will relocate that data to a new location and leave the old block in place. Eventually the SSD must reclaim that wasted space by moving all the other remaining blocks to a new location so the entire area can be erased and reused. Unfortunately this process is usually called defragging, leading people to assume it has something to do with normal filesystem defragging.

Re:I'll wait a while.

[maxume]

They aren't noticing drive wear. They are noticing that the drive works faster when it is empty.

It's very much an artifact of the components and design, but it isn't a sign of wear, it is the drive working the way it is supposed to.

People that buy them for fast reads will still be thrilled, people that buy them for fast writes will find out that they are only better than most hard drives.

Re:I'll wait a while.

[JimboFBX]

The problem is how flash has to be written to. You're only allowed to write ONCE to MLC flash pages and up to 4 times with SLC flash pages (in reality you can cheat a bit which some drive manufacturers might do but you are going beyond the manufacturer's specs and what they tested the flash at). Additionally, it's one-directional so changing a 0 back to a 1 requires erasing the whole block, and you're not allowed to program backwards in a block (so if you program page 1 you have to program page 2 next or erase the whole block). Keeping track of what is free and what isn't requires modifying the flash (or maybe in the future a separate PSRAM module). As you might guess, erasing a whole block to mark a single bit is inefficient and likewise so is wasting a whole block to indicate another block is in use is also inefficient. We haven't even gotten into wear leveling yet. Manufacturers end up splitting up the drive into larger segments and then using a bucket array hash table algorithm (as an example, I'm just guessing) to find free blocks and pages within those segments. Those larger segments are marked "completely full", "not completely full", and "empty". Finding what blocks and pages to put data is extremely tricky. So is debugging it. This stuff is slow, it takes weeks to thoroughly test a single drive, and if you update the firmware... guess what, another several weeks of testing.

All SSDs automatically do garbage collection (even those without TRIM), when they find the spare time, which is what defrag basically is. If it weren't for these garbage collection algorithms you would only be able to use a fraction of the drive's space before it ran out of free blocks. TRIM makes the garbage collection significantly more effective however.

Re:I'll wait a while.

[Kleokat]

I just bought one on 60 gigs for approx 300$. I went for the fastest I could find - a whopping > 200 MB/sec. Beats my HDD-stripe. My SuSE linux boots from GRUB to desktop in less than 5 seconds, including all the init-scripts for vmware and mysql and you-name-it. I can now feel how fast my computer really is - now that I don't have to wait for this antique HDD anymore. Performance is incredible. I can only recommend spending the (rather large amount of) money, it's a mind-changing experience. From now on, I regard hard drives as an antique medium, which is good for storing lots of data, but not for installing operating systems. btw: All this defrag speak - it always helps, because there are still read-ahead and similar algorithms, which perform better, when the drive is defragged.

Re:I'll wait a while.

[severoon]

So this is great information, thanks so much for taking the time to write it.

So the way to restore a drive to full write speed during garbage collection would be to find 2 or more partial blocks that can fit in a single block, move them all, then completely free those previously partially used blocks. If this is how garbage collection works...why is it that the drives still degrade over time? Just when they fill up a certain amount when it's no longer possible to clear an entire block? (Seems like it should always be possible, though...unless files are required to be always kept in contiguous blocks...)

Re:I'll wait a while.

[JimboFBX]

Drives with TRIM don't really degrade that badly, its just the ones that don't support it or are running on an OS that doesn't support it (such as windows XP). They degrade because every time something is written something else has to be erased first (since the part eventually fills up all of its free space). The read performance doesn't degrade that much.

A watershed moment

[BadAnalogyGuy]

This is the IBM PC Jr of SSDs.

Re:A watershed moment

[Jay Tarbox]

Ouch dude! Leave the PCjr alone! That computer was my geek larval stage. While my buddies played games on their C64's, I was messing around in DOS on my PCjr. Trying to get drivers to load and still have enough RAM to play a "game".

Re:A watershed moment

[Yvan256]

So it's the first SSD to have a 3-voice soundchip?

Re:A watershed moment

[commodore64_love]

And the "chiclet" keyboard. I'll bet that was hella fun. ;-) I'm surprised IBM didn't learn from Atari which made the same mistake in 1979.

Anyway: I was looking for the PCjr on wikipedia and stumbled across the Coleco Adam. Basically a Colecovision upgraded to a computer. However due to poor sales (it cost about twice an Atari or Commodore) and poor design (the computer erased disks & tapes when booted), the Adam was a major flop.

Eventually Coleco went bankrupt. Commodore went bankrupt. Atari went bankrupt. It seems that building a computer in the early 80s was a curse that killed once great companies. Except Apple and RadioShack - they managed to survive (barely).

Re:A watershed moment

[Culture20]

This is the IBM PC Jr of SSDs.

No one will buy it, but it will usher in the age of the SSD. Who are you and what have you done with BadAnalogyGuy?

Re:A watershed moment

[Jay Tarbox]

One of the first things I got my dad to buy for me was the corded, regular keyboard. the infrared was neat but on the kitchen table - at certain times of the day - the sun would keep me from using the computer.

Re:A watershed moment

[ahem]

That was actually the divine presence telling you to go outside and play instead of mouldering away on the computer.

It was either that comment or the ever-trusty: "Sun? What is this 'sun' that you speak of?"

That's a much better deal than a 5GB Hard Drive!

[hellop2]

1000GB for only $23544.00

Speed?

[kingofnexus]

Paying $4000 for a hard drive is one thing, but how fast is it? Slapping what I assume to be a ton of chips together wont make for an impressive benchmark. If I had the cash to blow on this sort of thing I would rather raid together a bunch of and small fast ssd's than 1 big one.

Re:Speed?

[vlm]

Slapping what I assume to be a ton of chips together wont make for an impressive benchmark.

Same as:

... I would rather raid together a bunch of and small fast ssd's than 1 big one.

SSD seek time is zero, there is no multi-spindle advantage. Unless you are trying to exceed a system thruput of 3 gigs/sec, the limit of a single SATA channel...

Re:Speed?

[delinear]

This is exactly what I'm working to with my system at home. Now everyone has laptops and netbooks and WiFi phones, etc, it makes sense to have a big block of storage that's always on and can be nicely hidden away somewhere dark and cool and which anyone can dip into as they come and go.

Yay

[Pojut]

Can we please get an affordable, 60GB one that is actually worth buying now? Last time I checked (two months ago), most of the less expensive drives were real spotty with their reliability.

Any suggestions for a decent 60GB SSD for under $120?

Re:Yay

[IBBoard]

You can get 60GB for under $120? Damnit, I considered an SSD recently and 30/40GB was £100 for the cheapest ones. Didn't get it in the end because of reports of degrading performance over time. That'd be one hell of a downer if you'd bought something that large and expensive!

Re:Yay

[XPeter]

You can get 60GB for under $120? Damnit, I considered an SSD recently and 30/40GB was £100 for the cheapest ones. Didn't get it in the end because of reports of degrading performance over time. That'd be one hell of a downer if you'd bought something that large and expensive!

No, you can't.

http://www.newegg.com/Product/ProductList.aspx?Submit=ENE&N=2010150636%201421439415&name=60GB

The lowest price for a 60GB SSD is $140, and that's from a no-name company. If you want quality for that spec, your wallet will be taking a hit of about $200

Re:Yay

[rubycodez]

yes, in about two more years. hang in there.

and on that day I'll be getting by really cheap, as I only need about 20GB for a laptop

Re:Yay

[Yvanhoe]

And mine will probably have 10 TB. "You got room for only one OS in there ? Do you know that virtualization is all the rage right now ? Be secure, man !"

Re:Yay

[nschubach]

You can get 60GB for under $120?

No, you can't.

Re:Yay

[icebraining]

That would be excellent, if those damn HP people didn't disable the virtualization support of the CPU in the BIOS, with no option to turn it on!

> modprobe kvm-amd
kvm: disabled by bios

Re:Yay

[Yvanhoe]

Heh, everyday I find a new reason why we need open BIOSes as well.

Re:Yay

[mindstrm]

$200 is quite affordable - so for $200 I can drop in one of these, use it as my system drive, and get a massive speed improvement, subjectively, for my $2500 computer.

Sounds like a great deal to me.

Re:Yay

[Khashishi]

Um, you get what you pay for?

Re:Yay

[djnforce9]

Exactly what I was thinking Khashishi.

These can't really take off though until they become just as affordable as conventional HDD's. Once they do though, they sure will make for some excellent backup storage devices at the very least though. It's a royal pain setting up my conventional external HDD's (i.e. bulky power supply+USB cable+space to set the rather large drive down, etc) to backup my data or restore from backup.

I assume or at least hope this drive can be powered by the USB cable itself.

Welcome back to the 90s

[Elledan]

So at roughly $4/GB that'd place us where, back at the late 90s? I'm not sure what part of 'catching up' people seem to think of when they're talking about SSDs replacing HDDs. Yes, they're faster in a number of applications, but HDDs are crazy cheap at $0.10/GB or better, fast enough for most purposes and have a longer life than Flash-based media. I guess I could pull out a stack of punch cards 1 km tall and claim it's got 1 TB storage capacity too, thus having 'caught up' with HDDs.

Considering Flash is reaching the point with its feature sizes (32 nm) where its data retention rate (1 year) and number of write cycles (8,000) is dropping rapidly (enterprise SSDs use 65+ nm SLC Flash instead), it's hard to see how Flash-based SSDs are winning, exactly.

Re:Welcome back to the 90s

[Anonymous Coward]

Um, they are WAY faster. They are also growing in size more rapidly than traditional hard drives. They have gone from like 32gb to 1000gb in just a couple of years. They are also rapidly dropping in price.

Even now, a lot of people only use like 30gb worth of disk space. Sure, they have more, but they don't use it.

32 GB / $125 USD / Sequential Write: 187.5 MB/s / Sequential Read: 294.5 MB/s.
http://www.newegg.com/Product/Product.aspx?Item=N82E16820211419
For a lot of people, that would be the largest upgrade in terms of speed they could possibly give there computer. Maybe reducing the time to load photoshop from 8 seconds to 2. Loading Word for 3 seconds to instant. Simple as that. For $125 dollars.

64 GB / $149 USD / Similar speed
http://www.newegg.com/Product/Product.aspx?Item=N82E16820139132

128 GB / $351 USD / Similar speed
http://www.newegg.com/Product/Product.aspx?Item=N82E16820148319

Again, these are the largest speed improvements that you can possibly give your computer right now. That isn't insignificant. At all.

Sure, they aren't really there YET, but it won't be that long.

Re:Welcome back to the 90s

[MyLongNickName]

i am not a hardware expert. However, I have a few uses for something like that in the small company where I work. $4k is pricey, but for applications that rely on huge file I/O and is sensitive to speed, this is viable.

I've already ordered a SAN solution.... but if I were making the decision again, and the price dropped by a factor of four, I would likely go with the SSD if I could mix and match with traditional hard drives.

Re:Welcome back to the 90s

[Gruturo]

I guess I could pull out a stack of punch cards 1 km tall and claim it's got 1 TB storage capacity too, thus having 'caught up' with HDDs.

This being Slashdot, I'd expect better of you :-)
A 1km-tall stack of cards, which, according to Wikipedia are 0.178mm thick and can storage 64 bytes with the most efficient coding, results in a measly 342.89 megabytes (assuming 1 megabyte= 2^20, which is admittedly uncommon when quoting storage, esp when a vendor does it. They'd use the 10^6 version, so 359.55 megabytes (I'm aware of the kibibyte/mebibyte etc scale, but I don't like using it))

For a full terabyte you're looking at slighly over 3058km worth of stacked punch cards (or 2781.25 km if using the storage vendors' definition)

(Disappointingly, Wolfram Alpha was no help doing the above calculations)

Re:Welcome back to the 90s

[0123456]

32 GB / $125 USD / Sequential Write: 187.5 MB/s / Sequential Read: 294.5 MB/s.
http://www.newegg.com/Product/Product.aspx?Item=N82E16820211419

You do realise that sequential reads and writes are pretty much irrelevant to most people, right? The big benefit of SSDs is _random_ read and write speed, which is where HDDs really suck.

For a lot of people, that would be the largest upgrade in terms of speed they could possibly give there computer. Maybe reducing the time to load photoshop from 8 seconds to 2.

And how often do you load photoshop? For most people, saving six seconds on something they do once a day is hardly going to be 'the largest upgrade in terms of speed they could possibly give their computer'.

I put an SSD in my new HTPC because I wanted it to boot up fast, and while it probably halves the boot time there it's otherwise pretty underwhelming.

Re:Welcome back to the 90s

[vlm]

So at roughly $4/GB that'd place us where, back at the late 90s? I'm not sure what part of 'catching up' people seem to think of when they're talking about SSDs replacing HDDs.

I deployed a 586 based single board computer using a 4 gig CF as the boot drive about a year ago. Entire system draws about 4 watts total and no moving parts. I would call it vaguely mid 90s ish specifications. If you define HDD as advancing about one year per year, then SSDs seem to be advancing about half a decade per year, thus "catching up" at a rate of about 4 years per calendar year, and currently "about a decade behind" so figure SSD will pass HDD around the end of the world, late 2012-ish. Sign of the Apocalypse?

Re:Welcome back to the 90s

[commodore64_love]

>>>SSDs are faster in a number of applications, but HDDs are crazy cheap at $0.10/GB or better

Precisely. It's why when you buy Mario Kart Wii, it comes on a cheap disc not a solid state cartridge. The disc can be mass-produced and is very simple, which is why it will always be cheaper than the more-complex chip-based storage.

Re:Welcome back to the 90s

[vlm]

I put an SSD in my new HTPC because I wanted it to boot up fast, and while it probably halves the boot time there it's otherwise pretty underwhelming.

Isn't it quieter? When I installed a SSD in my mythtv frontend, hard drive noise went from noticeable, to gone.

Re:Welcome back to the 90s

[Firethorn]

i am not a hardware expert. However, I have a few uses for something like that in the small company where I work. $4k is pricey, but for applications that rely on huge file I/O and is sensitive to speed, this is viable.

True, but I was just shopping for storage and considering a SSD.

Newegg ran a deal on a 2TB HD yesterday for $130. Standard price for many is $140-150. $150 for 7200 RPM.

At $4k, you can afford to have 20 of these drives and still have money for some fancy controllers. Run them in RAID-1 for the DB application to give you the necessary bandwidth/capacity. 20 Drives beat 1 SSD controller. Heck, for most applications 5-10 beat the SSD. Write doesn't matter much because, well, SSDs write about as slow as HDs.

Re:Welcome back to the 90s

[mayberry42]

New applications of technology are always much more expensive than their older counterparts. Why do you think the latest and greatest Nvidia video cards cost $hundreds? (disclaimer: it's been years since i've worked with or built computers, so the details may a bit out of date, but the concept still stands) In part, supply and demand. Part also because they are trying out new techniques which require substantial R&D costs. Once problems are ironed out and people start buying en masse (assuming it's not a shoddy product/tech to begin with, like 3D -- until now), economies of scale will kick in and the "true market value" (for lack of better terminology) of the mass will ultimately dictate whether it's ultimately worth the investment or not. That's the way it was with the introduction of PCs, CDs, video game consoles and any other type of technology conceivable. this will be no different.

If you don't like it now, that's fine (and very understandable), but just give it another few years until the market takes over before making any judgements. In the mean time, sit back, relax, and let the uber-geeks do their thing :-)

Re:Welcome back to the 90s

[bertok]

Seconded.

Nobody who's actually used an SSD talks about how hard drives will remain relevant. Nobody.

It's not even about the streaming speeds. Going from 60 IOPS (typical SATA disk) to 6000 IOPS(*) is a night & day difference. There's just no comparison. Meanwhile, the latest SATA 6 Gbps drives can reputedly do 60,000 IOPS!

To put that in perspective, 60K IOPS is the same as a 350x 15K RPM drives in a SAN storage array. That's the kind of thing that banks buy for $millions, but it's still not as good as the SSD, because a SAN can only do those IOPS if it's receiving many independent requests on many "threads". The latency of any single request is still a couple of milliseconds. Meanwhile, SSDs can not only pull insane aggregate IOPS, but they can do it for a single thread, because the latencies are way lower.

There's "enterprise" SSDs out already that can do over 200K IOPS over PCI-e. Sooner or later, that's going to trickle down to the consumer market.

Hard drives are the new tape.

*) that's what I get from the somewhat outdated SSD in my laptop.

Re:Welcome back to the 90s

[timeOday]

You do realise that sequential reads and writes are pretty much irrelevant to most people, right?

My #1 storage delay is waiting for virtual machines to suspend and resume. That means writing or reading a 1GB (or whatever the VM's RAM is) contiguous file. Before SSD, suspend/resume to disk wasn't even worth the wait, now it is. Most people don't bother with VM's, but suspend-to-disk in general is a feature that millions of home users should be using by default to save power (compared to never shutting down) and reduce waiting (compared to rebooting just to grab an email).

Of course, the fast seek time is great too.

Re:Welcome back to the 90s

[rtb61]

The largest potential advantage for SSD drives, is clever redesign can allow for extreme expandability. Unlike harddisk drives, SSD drives can inherently be designed as more open units, with additional slots for additional memory cards ie, it is readily possible to manufacturer a SSD drive that starts with say 50 gig of memory to which you can add additional memory cards at say 25 gig a piece. All this done before adding the additional cost of another drive with it's own expanding memory slots.

As prices fall, likely target would be 100 gig drives with 100 gig expansion cards to fill out to a terabyte.

Re:Welcome back to the 90s

[zach_the_lizard]

I have an SSD, and I too use it for suspend to disk/resume goodness. It resumes much quicker than with an HDD, and my computer is a bit quieter now, too. A lot of things are just quicker.

Re:Welcome back to the 90s

[Joce640k]

This being Slashdot, you should know that storage manufacturers always exaggerate the data capacity of their products.

The _unformatted_ capacity of 5.6 million punch cards is 998,320,126,812 bytes.

Re:Welcome back to the 90s

[VinylPusher]

You haven't factored in energy costs, physical space requirements or reliability. If you have 20 spinning disks, two will fail within 6 months (most likely) unless you've paid extra for e.g. an Apple fileserver.

Plus: noisy.

Re:Welcome back to the 90s

[value_added]

I deployed a 586 based single board computer using a 4 gig CF as the boot drive about a year ago.

I'll bet it's a Soekris box, and you've added a spinning 2.5 hard drive to it to as well. Did I guess right?

Did something similar about a year ago, too, but I opted for single 16GB Mtron MOBI SLC SSDs. All the units still work as expected.

Re:Welcome back to the 90s

[dylan_-]

So at roughly $4/GB that'd place us where, back at the late 90s?

Around 2002

I'm not sure what part of 'catching up' people seem to think of when they're talking about SSDs replacing HDDs.

They're thinking of SSDs previously being 12 years behind HDDs in $/GB (in 2007!) and now being 7 years behind. That's pretty good "catching up" by any measure!

Re:Welcome back to the 90s

[MrNemesis]

They're not faster in a number of apps; last time I looked (or, indeed, used a mechanical drive as a system disc) SSD's were the performance king by a nautical mile. I'm not aware of any type or number of hard discs that can compete on performance with even a mid-range SSD these days.

In the 90's your only option other than an array the size of a fridge was an enterprise ramdisc (and try getting either of those into your laptop). Depending on which metric is most important to you (and it should be random read/write), SSD's are pretty much an order of magnitude faster that mechanicals.

On top of that, not everyone needs the space. Corporate laptop users especially will boot off a small 30-60GB SSD and have all their documents presented over the network or over VPN. Paying $0.10 a GB is a loss when it's cheaper for you to pay $5 a GB and get $30 more productivity from an employee per day due to faster boots, shorter logins, less thrashing during the mandatory virus scan and fewer mechanical failures (which is often enough to put an employee out of action for a significant fraction of the day, or even week if they're out and about).

IIRC Intel (who make up 2 out of 5 of the SSD's I currently own) say you can write something like 80GB to one of their 160GB drives every day and not start running out of sectors for five years; I don't know of any mechanical discs (especially 2.5" SATA ones) that are capable of such punishment. Flash and SSD controllers (along with a DRAM cache and NCQ) have really come a long way in mitigating the rewrite/longevity problem.

Disclaimer: I don't understand why more people don't love SSD's, especially when the dislike seems to relate solely to their high price.

Re:Welcome back to the 90s

[icebraining]

I put an SSD in my new HTPC because I wanted it to boot up fast, and while it probably halves the boot time there it's otherwise pretty underwhelming.

Which is why I hibernate, because unlike normal booting, it is a sequential read of the memory dump back to RAM (at least, if you reserve a partition for it).

Re:Welcome back to the 90s

[ArsenneLupin]

A 1km-tall stack of cards ... results in a measly 342.89 megabytes

... nice.

But you forgot a nice wording flame:

I could pull out a stack of punch cards 1 km tall and claim it's got 1 TB storage capacity too, thus having 'caught up' with HDDs.

You "catch up" with somebody how had a headstart. Given that HDDs are newer technology than punch cards, it's cute to claim the punch cards could "catch up" with HDDs, even if there was a way to make a 1TB stack of them.

Re:Welcome back to the 90s

[0123456]

Isn't it quieter? When I installed a SSD in my mythtv frontend, hard drive noise went from noticeable, to gone.

True, but my MythTV backend with two hard drives is still sitting beside the TV until I get around to moving it, and while I can hear drive noise close up, I can't hear it from the sofa.

Re:Welcome back to the 90s

[DoofusOfDeath]

So at roughly $4/GB that'd place us where, back at the late 90s?

Why yes, if you completely ignore the performance difference, which is the main selling point for flash drives.

It would be similar if I complained about how my tricked out Honda Odyssey is no cheaper than a new 1980's Chevy van was.

Re:Welcome back to the 90s

[radish]

You know what I do a lot? Install apps. Install updates. Compile stuff. The huge increase in random access speed from an SSD makes all these things so much faster I hate going back to machines with spinning disks.

Re:Welcome back to the 90s

[iamhassi]

""Compare how fast 3.5" capacity went from 1G->500G to how relatively slowly its inched from 500G->2000G."

First 1GB hard drive came out around 1994 or 1995. It took 10 years until the first 500GB HD came out in 2005. Then the first 2TB drive came out in 2009, 4 years later. So basically, what the hell are you talking about?"

First, I need links.
First 100gb, 2001
First 500gb, 2005
First 2tb, 2009

I'd love to find older stories but any page before 2000 doesn't rank well on Google.

Anyway, Quantum you're right, we went from 500gb to 2tb in 4 years, while the trip from 1gb to 500gb took a little over ten years.

Re:Welcome back to the 90s

[0123456]

My #1 storage delay is waiting for virtual machines to suspend and resume.

Then you're not most people. Most people boot up the PC, do some web browsing and email, watch some youtube video their mate on myspacebook sent them and then shut down the PC. Booting faster is a benefit, but since that's largely random reads, a higher sequential read rate might save them a few seconds of boot time... they'd save a lot more by wiping Windows and installing Linux.

Most people don't bother with VM's, but suspend-to-disk in general is a feature that millions of home users should be using by default to save power (compared to never shutting down) and reduce waiting (compared to rebooting just to grab an email).

So again you're talking about something people might do once or twice a day being slightly faster because the sequential read and write time is faster. The rest of the time they'll see no benefit because PCs rarely perform hundreds of megabytes of sequential reads and writes in normal operation... for the kind of things most people do on their PC even the improved random reads and writes really only affect application startup and games which stream data from disk, it won't make web pages load noticeably faster or youtube videos play better.

Re:Welcome back to the 90s

[iamhassi]

"Even now, a lot of people only use like 30gb worth of disk space. Sure, they have more, but they don't use it."

Your source? A single game now days takes over 10gb. Grand Theft Auto IV, which was released for PC in 2008, requires 16gb, Red Alert 3, also released in 2008, required 6 to 12 gb, Assassin's Creed, released 2008, required 12gb.

Even if you're not playing games, with 12+ MP images and HD video cameras common now days you're going to burn through 30gb in no time, with blu-ray HD being 40mbps.

So while it might be true your grandma only needs 30gb, anyone that has installed a game made in the last 2 years or takes the occasional picture or video is going to need a bit more.

Re:Welcome back to the 90s

[TheRaven64]

7 is still a bit far, but 4 is probably close to the tipping point. My current laptop is 3.5 years old, and its hard disk is still not filled. More space is always nice, but given the choice I'd go with an SSD the same size rather than a bigger disk, if the cost were the same.

Re:Welcome back to the 90s

[AK Marc]

At $4k, you can afford to have 20 of these drives and still have money for some fancy controllers.

You'd be hard pressed to get "fancy" controllers for 20 drives for the $4k. And good luck getting them all to fit in a 1RU server.

Re:Welcome back to the 90s

[Firethorn]

Well, no I haven't. Still, even 5 discs would be able to output more requests on average outside of applications/tests almost uniquely optimised for SSD, and a HD won't use more than $100 of electricity in a year, again, outside of special circumstances.

And the fact that the file server is an apple won't magically decrease the odds of a disk failure. It's not like they have their own HD factory.

SSDs do go bad. If the data is that critical, the multiple spinning platters means a graceful degregation of service if one fails, vs a complete failure.

10 discs, each with a 10% chance of failure within a given period of time, gives you a .00000001% of all of them failing.

Re:Welcome back to the 90s

[idontgno]

No, actually, I think the formatting overhead has been included in GPP's calculations.

They're only storing 64 bytes per card. Each Hollerith zone is one byte. Therefore, each card has 80, not 64, bytes of capacity.

The remaining 16 bytes? Formatting. Go ahead and include a 16-digit card number. There's your format. And also, when the computer operator drops your deck, you'll have card numbers to use in sorting the deck back into correct order.

Re:Welcome back to the 90s

[david_thornley]

Sheesh. A standard punch card had 80 columns, each with 12 possible punches. (There were other forms of punch cards, but not nearly as popular.) There may have been octet-per-column encodings, but that's 120 bytes of raw storage. (I'm going to disregard the problems with data integrity in one-heavy (i.e., lace) punch cards.)

Therefore, 5.6 million punch cards can contain something like 750 megabytes (that's the base-10 megabyte).

Now, unformatted, or blank, I suppose you could make up a process using a laser that will make zillions of potential little holes in the card. If you can get something like 150,000 holes per card, you'd be getting up to a terabyte. Since that's probably possible to build, I suppose the stack could have an unformatted capacity of a terabyte.

The protection of a kilometer stack against droppage and mixing up the order is left as an exercise for the reader.

Now, if you haven't used punch cards, get off my lawn!

Re:Welcome back to the 90s

[Cowclops]

Why would you store images and HD video on an expensive-per-gigabyte SSD? They don't benefit from high iops. You already have a platter drive thats presumably big enough for your media storage needs - why would you move it off your hypothetical ~1TB drive and back to the expensive 40GB SSD?

SSDs are just another intermediate step in the memory hierarchy. HDs are slower but cheaper than SSD, SSDs are slower but cheaper than ram, ram is slower but cheaper than L3 cache, L3 cache is slower but cheaper than L2 cache, L2 cache is slower but cheaper than L1 cache, L1 cache is slower but cheaper than registers.

When processors started coming with L3 cache, people weren't posting on Slashdot saying "Why do I want L3 6MB of cache, thats not even big enough for firefox!" Adding more in-between faster but lower capacity memory speeds up your computer without rendering the slower-but-cheaper memory obsolete.

Re:Welcome back to the 90s

[idontgno]

Shoot, don't lecture me about cards. My first professional software projects were on cards. My first professional software projects were floor-sorted by computer operators, too, so I know of what I speak in this context. :)

I'm conforming to the storage scheme described earlier. 64 bytes per card is very low-density, given the total number of holes available. But avoiding lace cards is an admirable goal, unless you want to piss off the machine room operators. Which you may want to do in retaliation for floor-sorting your 1-3 km tall deck.

BTW, the lawn I'm standing on is my own.

Re:Welcome back to the 90s

[Joce640k]

You're all assuming 80 column punched cards with one character per column.

That's one way of "formatting" them, sure, but hardly the highest density possible.

{whoosh}

Re:Welcome back to the 90s

[maxume]

They aren't going to connect together with magic.

A 1 TB drive with the chips wired to the controller is going to be fantastically cheaper than 1 TB of chips wired to pins, inserted into plugs wired to the controller.

And you don't even have to mess with it.

Re:Welcome back to the 90s

[rtb61]

You could skip the whole controller, put the functional directly into the CPU and just have slots on the mobo or a daughter board for the memory, cheapest solution, no hdisk controller at all.

Re:Welcome back to the 90s

[meringuoid]

Ah, but a single statk of cards 1km high would just tip over. You'd need many stacks clustered together to reach such a height.

Try filling the Burj Khalifa with punched cards. Total floor space of 464,511 square metres. Let each floor be three metres high, that's a volume of 1.4 million cubic metres. Taking the figures from your cited Wikipedia article, a card is 2.7 millionths of a cubic metre. I make it about 5E11 cards, which at 64 bytes per card gives us... getting on for 30 terabytes.

Re:Welcome back to the 90s

[vlm]

I'll bet it's a Soekris box, and you've added a spinning 2.5 hard drive to it to as well. Did I guess right?

Yup meet my IP PBX (and other services). Love the Soekris stuff, it just works.

Re:Welcome back to the 90s

[mitgib]

At $4k, you can afford to have 20 of these drives and still have money for some fancy controllers.
You'd be hard pressed to get "fancy" controllers for 20 drives for the $4k. And good luck getting them all to fit in a 1RU server.

You'd be surprised what $4k will buy. Now I didn't meet your 20 drive or 1U, but the following is a totally workable SOHO solution

3ware 9690SA-4I ~$300
Chenbro CK12803 ~$300
16 * 2tb HDD @ $150 $2400
Coolmaster mid tower $60
3 5-in-3 Hot swap bays ~$300
750W 80% Power supply ~$100

~$3500 leaving $500 to fill out the mobo/ram/cpu. These are quick rough numbers, but reality, this is quiet, fault tolerant, and in my closet. I store all my movies/music on this as well as archive all the CD's that come with the things I buy, like driver disc's and manuals. This is for my home, not some enterprise solution, so the risk of flooding the controller is minimal, but I can expand this to 128 devices, and for a household streaming media, I can see it serving that purpose for many years to come. Anything new I build will most likely use SSD for the things needed on the local machine, then the NAS will be sitting there at the ready for all the large storage needs.

I've used WHS as the platform running my NAS, mostly for the backup solution it provides for all the client machines, at ~$100, it is one of the few values Microsoft has on the market. If your household is like most, running Windows XP/Vista/7 then WHS is definatly worth a look, HP even has some small boxes on the market starting at something like $400 and expandable, I don't care for them, but they might be a good fit for others.

Cheaper than 1TB FC drives, EMC SSD

[symbolset]

Hey, $4K/TB isn't that expensive. What's the performance and reliability like?

Re:Cheaper than 1TB FC drives, EMC SSD

[Urza9814]

Yea, it's only _40 times_ the cost of a regular hard drive.

Why not buy a crapload of regular drives...then you can RAID a bunch of them and have a crapload more left over for when any of them fail. Guarantee 40 TB regular drives could be set up to be both faster and last longer than a 1TB SSD. Would be a bit bigger though....

Re:Cheaper than 1TB FC drives, EMC SSD

[VinylPusher]

Maybe you want to deploy it somewhere you don't visit regularly...

Perhaps you have space constraints.

Perhaps you want to put it in a music production environment (which needs to be quiet).

There are any number of uses for $4k/TB silent, low-power storage. Not all of them will fit your idea of sensible, but some are still valid.

Re:Cheaper than 1TB FC drives, EMC SSD

[ShadowRangerRIT]

Well, it's MLC flash, so its lifespan will be shorter and its write speed will be substantially lower than an SLC flash drive. Should roughly match a high speed platter drive (slightly slower at sequential write, faster at random write), but you won't see the jaw dropping numbers that SLC drives can pull.

Re:Cheaper than 1TB FC drives, EMC SSD

[mindstrm]

The SSD will still win on seek time.

Re:Cheaper than 1TB FC drives, EMC SSD

[silas_moeckel]

The raid controller to get anywhere near the IOPs of this is prohibitive. I run 24 drive software raid 6's in storage servers all the time the cost of the 24 port raid card is more than the server chassis power supply and motherboard add int he fact that having 24 GB of cache does great things as well. With enterprise class drives we loose a drive in one of those chassis every month and a half or so per chassis (36 months average drive lifespan roughly). Conversely I also run a lot of DB's and over the last year we have been using the Intel SSD's only one failed (infant mortality) the hardest problem is getting some form of raid that does not cripple there IOPS.

Solid State of the Art

[sackvillian]

It's too bad that I won't be able to take this baby for a spin...

Re:Solid State of the Art

[theY4Kman]

I think that all the time at the park!

great scot!

[rarel]

This sucker's electrical. But I need a nuclear reaction to compress the data to the 1 terabyte of capacity I need.

Not so sure about the "aspirational" line

[vikingpower]

"aspirational more than affordable" ? For business ( on-site programming ) purposes I just ordered a new laptop with two 256-Gb SSD drives. Only a few hundred bucks more expensive than one with disks. Wait a year or two, and 1 Tb SSD drives will be perfectly normal items on a medium to high end computer.

Re:Not so sure about the "aspirational" line

[Urza9814]

Yea. Only a few hundred bucks more expensive for the entire laptop, which may not seem like much...but is it really worth it? Of all the things you could have spent that money on, it went to the hard drives? You're adding a few hundred bucks for a part that is usually under a hundred bucks.

Re:Not so sure about the "aspirational" line

[vikingpower]

That is an Alienware M17x. I live in Europe, ordered it in the US. May sound funny, to use a gaming laptop for work. But it works ! Clients and prospect are astonished to see the thing, and gawk at its speed. Gawking prospect = nearly 100% a new client.

Re:Not so sure about the "aspirational" line

[badran]

-Lower Temps.
-Lower Noise.
-Lower power users, hence longer battery life.
-Lighter.
-Faster bootup. If you are paying 60 USD/hour or 1 USD/min, and saving 1 min every reboot or startup. You will make up the difference in 1 year or so.

Re:Not so sure about the "aspirational" line

[vikingpower]

but is it really worth it? Of all the things you could have spent that money on, it went to the hard drives?

Yep. I work on-site at my client's ( code generation, compiling, then running & sending data to my remote server ). Together with an i7 XM processor, SSDs make an enormous difference.

Re:Not so sure about the "aspirational" line

[badran]

*Lower power usage*

Uhmm....

[sonicmerlin]

This has been on newegg for a very long time: http://www.newegg.com/Product/Product.aspx?Item=N82E16820227500

I've been waiting forever for its price to drop, but nothing seems to be happening. I don't think SSDs will be of any consequence to mainstream users before memristors become all the rage.

Re:Uhmm....

[The MAZZTer]

But look, it's only $1.99 shipping! What a deal!

Re:Uhmm....

[haruchai]

Considering how much variety and mindshare there's been around SSDs and the competition with HDDs, I find that the prices have
stayed much too high for far too long.
I wonder if we are going to hear about another PC/IT price-fixing scheme.

That's a lot of money.....

[commodore64_love]

.....just to store downloaded episodes of Star Trek, Stargate, and Galactica (plud other unmentionables). I'll buy the DVDs instead.

Re:That's a lot of money.....

[CastrTroy]

You don't need a solid state drive for storing media. It's completely useless. There's only a few really good uses of these things. Mostly in places where you have a lot of reads all over the disk in a very short amount of time. Mostly for things like Databases and stuff. For personal use, it really only makes sense to store your programs and OS on it. There's no reason to store things like movies and MP3s on there. Get a second drive spinning platter drive for that.

Re:That's a lot of money.....

[Firethorn]

Yeah, like my video server trying to feed multiple HD streams at the same time.

How many HD streams at the same time?

Right now it'd be cheaper for you to get 2 drives in raid 1 and a raid controller smart enough to queue requests seperately to the two drives.

Unless you don't have enough movies to make HD's economical - HD's scale up well, Flash scales down better.

Re:That's a lot of money.....

[misexistentialist]

Not a bad idea if you have 1TB of torrents running 24/7...

Re:That's a lot of money.....

[icebraining]

It would probably be much less expensive to buy some HDs (four 250GB, instead of a 1TB SSD) and use raid0.

Re:That's a lot of money.....

[TheRaven64]

Yeah, like my video server trying to feed multiple HD streams at the same time. Zero seek time sure would be nice. Good thing video is not media, because then it would be useless.

If you have an OS from the last two decades and a decent amount of RAM, then a mechanical disk is still a better bet. The read-ahead caching will read a few MB from one file, store it in RAM, read a few MB from the other file, store it in RAM, and so on. With a one second buffer per client (around 4MB for HD), you only need one IOPS per client, so a typical mechanical hard disk can serve 50+ clients quite easily in terms of seek time, although probably not in terms of throughput (an SSD will have problems there too though).

Re:That's a lot of money.....

[G00F]

No reason?

How about power, sound, heat, and not having drives go bad because of age alone?

My home server has 8 HD's. 6 in RAID5, 1 OS, 1 backup(cron job tar up certain directories). It's mostly a glorified samba/web server using a 12 watt CPU, the drives are what use the power.

Re:That's a lot of money.....

[Khashishi]

Yet, they are a perfect medium for passing movies and MP3s on to your friend.

I have a 512GB SSD

[Anonymous Coward]

And it is awesome. My MacBook Pro flies. I've also lost a lot of weight since I don't have money for food anymore. Highly recommended.

Not 400x

[radaos]

Maths fail in article. $4000 / $100 != 400x

Re:Not 400x

[keithpreston]

You just don't know where to find the $10 1TB hard drives. I'm going to buy them out and sell them on Ebay!

Re:Not 400x

[commodore64_love]

I got my terabyte USB drive for $50 at a staples sale. So $4000/50 == almost 100x cheaper.

Snow Leopard

[blackfrancis75]

It's a real pity OSX 10.6 failed to add TRIM support. With Win7, this is the first time I've seen MS cut Apple's lunch.

Re:Snow Leopard

[anaesthetica]

Apparently TRIM support is already in Mac OS code in both 10.5.7 and 10.6.*. It's called IOStorage::discard. It's just not turned on.

Re:Snow Leopard

[blackfrancis75]

That's interesting, but it's not really TRIM support. If you read TFA, It's *most* of the implementation of TRIM, implying that it will be supported at some future date. Perhaps the day after Duke Nukem Forever is released.

Re:I can seem some enterprise paying for this.

[ircmaxell]

Well, a normal 15k RPM SAS drive costs about $1400 per TB ($700 for a 500gb drive) and draws around 16 watts of power (for a Seagate Cheetah at least). Let's assume these SSD's will be like the others and draw around 1 watt. So that's a difference of $2600 and 31 watts (Because you need 2 SAS drives per SSD). So every hour, each SSD will consume 31 watts less. So with a price of $0.12 / kWh, every hour the SSD will save about $0.0036. Over the course of a year, that will add up to about $31.44 in power savings. So you'd need to run the drives for around 82 years to recoup the added cost from power savings (A higher electricty cost will lower this, but even at $0.50 per kWh, you're looking at nearly 20 years). Needless to say, that's well beyond the life span of the drive. So no, a prudent company won't buy these for power savings...

Re:I can seem some enterprise paying for this.

[PopeRatzo]

Imagine the power savings and time savings replacing existing storage with these.

As a media producer who uses DAW and video editing apps, solid state storage is a dream for me. I'm using much smaller SSDs now and although the power savings don't mean much to me, they are certainly quieter and faster than magnetic or optical media.

My 15k rpm drives are too loud and too warm.

When a 1TB SSD hits $1000, I'm in for two.

Not so bad, compared to:

[Ancient_Hacker]

Paying $4000 for a thousand gigabytes is not so bad. Some of us have worked on:

DEC DF-32: 32K 12-bit words for around $5000 (1971)

DEC RKO5- 2.5 megabytes for $10,000 ( 1973 )

Mac HD-20: 20 megabytes for $1000 ( 1985 )

All those were like, 1000x or more per byte. AND WE WERE PERFECTLY HAPPY. (Well, a little cramped on the DF32)

Re:Not so bad, compared to:

[nschubach]

Is this a new play on "When I was young..."? ;)

Re:Not so bad, compared to:

[Junior J. Junior III]

Paying $4000 for a thousand gigabytes is not so bad. Some of us have worked on:

DEC DF-32: 32K 12-bit words for around $5000 (1971)

DEC RKO5- 2.5 megabytes for $10,000 ( 1973 )

Mac HD-20: 20 megabytes for $1000 ( 1985 )

All those were like, 1000x or more per byte. AND WE WERE PERFECTLY HAPPY. (Well, a little cramped on the DF32)

Wow. I will be getting of your lawn now, sir.

Re:Not so bad, compared to:

[sanosuke001]

I'm not all that old (26) but I remember when I bought my first hard drive (2.1gb. Not my first HDD but the first I paid for myself) out of a catalog, the most expensive ones in there were SCSI 27GB for like $25000. I couldn't imagine what I'd do with 27GB at whatever age I was (11-12 or something)

Now I have a file server with 6TB and it doesn't seem endless anymore...

Re:I can seem some enterprise paying for this.

[fuzzyfuzzyfungus]

The only real "enterprise" case for SSDs(besides just making the boss's laptop quieter and more responsive for a few hundred bucks extra) is IOPS.

As a mass storage option, SSDs are pretty pitiful. As you note, even 15k RPM SAS stuff, hardly the cheap seats, is substantially cheaper per gigabyte. If you can step down to 10K RPM, or even the nicer grade of 7200RPM SATA(SAS/SATA compatibility can be quite convenient), the difference gets even starker.

If you are talking IOPS/$, though, SSDs passed the "economically viable" point some time ago and were last seen running for a location somewhere between "not even fair" and "Good God, man, it's like curb-stomping a puppy!" in their competition with even the zippiest of mechanical drives.

Affordable

[Mr_Silver]

Over recent years Solid State Drives (SSDs) have moved from luxury to affordable additions to one's PC

When I can get a 1TB 3.5" SATA drive for £61.33 (approx $94.58), I'm not sure how something which is 42 times more expensive can be considered "affordable".

Maybe I have a different definition of the word.

Re:Affordable

[hesiod]

Yes, you do have a different definition; your definition is entirely wrong. One item's affordability is in no way affected by the price of another item, regardless how similar it may (or may not) be.

Many SSDs are quite affordable for many people. A $100 1TB drive is just cheap -- in price and, I suspect, in quality.

Re:Affordable

[Ksevio]

And at that price point, you could buy a few drives plus the RAID controller and still have money left over for power costs and replacements for several years.

Re:Affordable

[Just Some Guy]

When I can get a 1TB 3.5" SATA drive [ebuyer.com] for £61.33 (approx $94.58), I'm not sure how something which is 42 times more expensive can be considered "affordable".

For you and me? It isn't. For the database guy or the NAS admin who's installing one to act as as ZFS L2 cache, it's pretty dead simple.

Price.

[Urza9814]

Sweet! Maybe now I can replace the 1.3TB of hard drives in my desktop with some solid state!

Oh wait. Nope. I can barely afford a 30GB drive. Let me know when SSDs are less than $1/GB.

Re:Price.

[bbn]

Come back in a year or so.

we paid that for a megabyte in late 1970s

[peter303]

Its all relative, folks.

Re:I can seem some enterprise paying for this.

[Joce640k]

It might save them on power infrastructure as well as electrons...

(By switching to SSDs they might be able to add a couple more servers without changing any cables and/or upgrading the UPS and backup generators).

Reference Articles

[mindbrane]

I don't pretend to even an elementary working knowledge of this stuff but the Anandtech articles seem to be the most frequently cited reference starting points. The SSD Anthology: Understanding SSDs and New Drives from OCZ and The SSD Relapse. I've a rudimentary understanding of the problems but have yet to come across anything that speaks to whether a SSD can be "refurbished" at the end of it's relatively short life, or, if a technology could be developed that would be profitable to refurbish SSDs at the end of their life. Just to underscore how little I know about this, I'm not at all sure what I mean when I say "refurbish" a SSD.

Not the right use of this tech

[DCFusor]

I'd simply want to put the "read mostly" parts of say a linux distro on one, to improve boot and app startup times, and leave the rest on good old spinning disks, even though I'm severely power limited here and SSD's would help with that. Fairly easy to do in Linux with mounting points and so on, not so easy with other opsys. Of course, the more bloatier apps still churn a bunch of writing on startup due to config files being updated and the like....

But this way you don't write the SSD much, it should live long and not get real fragmented. Also, you don't need that big size to pull it off. To save power (but not annoyance) you then set the spinning disks to power down real quick when not in use...

Seems like a good plan for my home main server which has a ton of read-mostly data on it -- now that could use up some SSD capacity indeed. But nearly all of that read mostly data could do as well on a normal hard drive -- music, data sheets, scientific papers and such like don't need blazing fast reads to work well. I can't see putting my MySQL databases on SSD anytime soon though.

disapointing

[petermgreen]

You could already get half this capacity in a laptop sized drive and a desktop drive is more than twice the volume of a laptop drive.

Ahem

[dreamer.redeemer]

Ignoring as much as possible the confounding composition of this summary, there's something very wrong with this bit:

"...but mechanical hard drives are still king when it comes to capacity. That was until the revamped Colossus LT series Solid State Drive came along this week. With up to 1TB..."

Given that standard desktop form factor hard drives with a capacity of 2 TB are readily available for purchase, it doesn't seem that the arrival of the Revamped 1 TB Colossus LT Solid State Drive represents even a slight advantage for SSDs regarding capacity. Furthermore, as others have pointed out, instead of this single SSD, 20 traditional 1 TB hard drives could be purchased with enough budget left for a server board, processor, ram and a few discrete RAID cards.

I'm surprised to see this publicly available, usually such premium-priced products are exclusive to industries with more dollars than sense--film and medical come to mind.

Re:I can seem some enterprise paying for this.

[aminorex]

It's 4000 AUSTRALIAN dollars, not human dollars.

Use as cache?

[HuguesT]

Can't we use a smallish SSD acting as cache in front of a large spinning disk? This is old technology, that may need to be modified somewhat to take the particulars of SSDs into account, but surely this is feasible? Any reason why not?

Re:Use as cache?

[dbIII]

That is pretty well exactly what mirror.aanet.edu.au do. They have the most requested content on SSD as a front end to their main disk based storage.

Re:Use as cache?

[Marcika]

Can't we use a smallish SSD acting as cache in front of a large spinning disk? This is old technology, that may need to be modified somewhat to take the particulars of SSDs into account, but surely this is feasible? Any reason why not?

There are drives that do this (from Samsung, apparently) and it is supported by Vista and Win7 drivers.

They seem to be much more expensive than pure HDDs and apparently not all that much faster, especially for writes -- but I wonder whether a better OS-side driver implementation could change that...

Re:I can seem some enterprise paying for this.

[iamhassi]

"So no, a prudent company won't buy these for power savings..."

Bah! You and your "prudent company"! Obviously you have no understanding of the marketing power of being able to say your company is "green". Check out this video from Google and just imagine the millions spent on the solar panels and free electric cars for employee use. Like she says "socially responsible company" is the hot buzzword right now.

Re:I can seem some enterprise paying for this.

[ircmaxell]

Oh, I'm not saying that they don't have a good place in the market. I am saying that it would be foolish to buy them from a power consumption standpoint alone. If you need IOPS, then the added cost can definitely be worth it...

The other thing you need to look at is lifespan. A 15k drive should last a company at least 3 years (I know some companies replace them yearly, but a typical rotation is 3 to 5 years based on what I've seen). Can an SSD (that's under high I/O) last that long? Or are you going to be replacing them yearly because of wear leveling issues? Again, I'm not saying that they are not worth the money. All I am saying is that it's far from a simple math problem to determine if they are the right fit for an enterprise...

Re:I can seem some enterprise paying for this.

[b0bby]

Well, you'd need to factor in the additional cooling costs associated with the higher power drives too, so you'd recoup in closer to 41 years...

Worth it!

[AndyS2]

Have you ever used a notebook with a good ssd? (Intel Postville 80GB, for example) I have seen one in action (don't own one right now, but will soon), and I can tell you that browsing the web with firefox and 'using' that system is such a dream compared to using a normal notebook with a harddrive that you can't really imagine going back to your old system. Just think about how often you really (re)start a program or click that 'Open File' button in a program, just to wait 1 - 4 seconds until you can continue working. Of course I'm not entirely sure if it really speeds up browser usage, but starting firefox / opera with a few tabs saved is much faster, and IIRC the caching mechanism of firefox with all the slow background filesystem syncing stops you from using your browser for a while, too, so maybe that is the reason why browser usage feels faster, too. You might be able to do all this with a ramdisk and enough RAM, btw (I've heard of people putting their firefox cache in one, for example), but I never tried preloading all 'common' applications when booting my OS because I find it too complex or slow when booting (There is some linux software that does it for you by learning what programs you like to use, though, and Windows Vista can do it too IIRC) -- Now, what parts would you be throwing those 200 - 400 Euros at when not choosing a faster storage device for a laptop? When assuming the 1500 Euro price range because we were talking about a dual harddrive notebook, I'd rather have the "2,0 GHz Intel Centrino dual core something" instead of a 2,5GHz one if that means I can get one SSD + harddrive / two SSDs while staying in my budget. And considering the performance gain, I'd consider doing this in the 800 - 1000 Euro range, too, even when not considering the better data safety if you drop your notebook or the possiblity of slightly less energy consumption (I'm not too sure on that one, though).

Re:I can seem some enterprise paying for this.

[siride]

You have a strange sense of economics. Reduced demand usually means reduced prices and vice versa. Reduced consumption means less need for more power-plants and other related expenses. On the flip side, much increased demand means that the power companies have to build new plants or upgrade old ones, or upgrade infrastructure. The fact is, using resources costs money and the more you use, the more it costs somewhere. Reducing consumption does not make costs go up unless there's a false economy created by imprudent decisions on the part of the power companies or some sort of insane government involvement that keeps fixed costs high (which I can buy).

Ahh, why bother? Let's just burn through all our natural resources like there's no tomorrow so that the status quo can be preserved at all costs. I'm sure that'll work out somehow.

With cooling you pay for that power twice

[dbIII]

Still it isn't much. Space and interface simplicity is going to be a consideration before that.
For instance, I have a single SSD in one machine because the only other way to get the same performance is to have yet another file server with a few more mechanical drives and a decent interface card (cheaper now than it was). In that case it was just cheaper to get one SSD, add it to the existing box (pulled out the slimline DVD) and make sure it isn't the only place where that data lives.
At relatively small volume sizes SSDs win on capital cost.

Re:I can seem some enterprise paying for this.

[mindstrm]

High-school algebra - no harder than any other technology decision regarding performance upgrades.

Re:I can seem some enterprise paying for this.

[Anpheus]

He doesn't care about that, he wants random access performance.

And yes, your RAID0 of four Velociraptor drives will have pitiful performance.

Here's a performance review that included IOPS:

http://www.pcper.com/article.php?aid=892&type=expert&pid=7

Everyone says these drives get 300-400IOPS peak. An Intel SSD will get you 8,000-12,000 easy, and the X25-E model will get you 15,000-20,000 easy. To compare:

http://www.legitreviews.com/article/1208/13/

That's the top two SSDs in the market, using Intel's controller and Sandforce's controller. For reads and a large request size the limitation is the SATA bus, you'll notice that each SSD maxes out the bus at 260MB/s. For small reads, it gets increasingly more ridiculous.

Four 600GB velociraptors in RAID0: $1120 and approximately 1200 IOPS. For 1.07IOPS/$.
One X25-M or Sandforce SSD in the 80-120GB range: $300 and approximately 8000 IOPS. For 26.67IOPS/$. IOPS actually grows as block size decreases, and both the X25-M and Sandforce SSDs see more than double the performance for 512B requests.

FUD FUD FUD and more FUD and more FUD

[celtic_hackr]

I don't know where you're getting your write cycle BS from, but here have some facts to go with your FUD.
Real SSDs today, will far outlast any HD you buy. Lifetimes are in decades, and will outlast the usability of every component in the computer it is initially built for.
The "guaranteed" writes of 100,000 is just meaningless. These suckers can do millions of writes.
Sure some of the no name models may not be as good. They may have bad design in write leveling. Which could impact the life of the device.

Re:FUD FUD FUD and more FUD and more FUD

[Elledan]

I got my 'FUD' from industry sources, such as:

According to Akihito Nishikawa, senior manager, Memory Division, SSD Application Engineering Dept of Toshiba, "The memory cell is assured for about 3,000 rewrites. If the target SSD capacity is 128GB, then the total bit capacity is 128GB times 3,000 cycles divided by 1.5 (our figure for rewrite efficiency), or 256TB. Toshiba research indicates that actual PC users generally don't rewrite more than 20GB, max, per day. If these numbers are used to estimate the SSD service life, it works out to 256TB divided by 20GB divided by 365 days, or about 35 years.

Source: http://techon.nikkeibp.co.jp/article/HONSHI/20090528/170920/

Re:FUD FUD FUD and more FUD and more FUD

[celtic_hackr]

Interesting and informative. But are your implying there is any disk, outside of SAS drives, that could last this long? While, disc drives, aren't rated like this, I'd like to see you make any, equivalent to the quoted device, OTC SATA last for that many writes, That's well outside the MTBF for any commercial SATA drive.

Re:Tax day is tomorrow...

[RyuuzakiTetsuya]

The President does control his party as well as signs and enacts spending legislation(yes, overturning v. from 2001 to 2006, it was President Bush who ran his party to run huge deficits, engage in two costly wars, and cut taxes. From 2006 to 2009, when Democrats took power, Republicans dragged their feet on a lot of things, and from 2009 until present, they've drug their feet on *EVERYTHING*. Even pay-as-you-go legislation rules that worked in the 90's got filibustered.

This kind of meltdown of our economic and political system can only come from unintentional incompetence.

Re:I can seem some enterprise paying for this.

[ircmaxell]

True, but my whole point is that the power savings is minimal, even including cooling costs. So 1TB of SSD would save you 60 Watts of power (including the traditional 2*power cooling estimate). So for every 5TB, you could add another 1U server. The savings is minimal in terms of power draw (at least compared to the investment required)... Sure, you could upgrade your 15 disk array to SSD, and save about 225 Watts, but it'd cost you over $45000 to save that 225 watts. For that cost, you could replace all your servers with blades and save over 3000 Watts (considering a typical 1U server uses about 300 Watts, and a typical 14 blade setup uses around 1kW). That's my point. Not that you can't save power by switching to SSD, but that you're not going to spend the $$$ for SSD just to save power...

Prove it

[celtic_hackr]

I'd like to know your source that 20 years is outside the life of any current SSD drive. I'd put the MTBF for such devices at 50+ years. Far outside the 3+ years for an OTS HD. Now once you go in WD Black drives and SAS drives, well that's a whole new category. Reliability comes at a cost. I'm not saying the $4000 SSD is a good buy. It's an intro price and as such is a premium. SSDs can make sense for a number of applications. When they come down more in price, I'll probably go that route. Also, there good and bad SSDs, just as in anything else in tech.

Re:Prove it

[celtic_hackr]

PS: If I was looking for reliable and fast storage, it'd be 15K SAS drives at 4-10x the price of such things as Seagate Cheetahs. There's still nothing out there to beat a good reliable 15k SCSI drive. When they can make an SSD in that price range, then they'll have a market.

Re:I can seem some enterprise paying for this.

[commport1]

*ahem* 1 USD = 1.078 AUD

Re:I can seem some enterprise paying for this.

[MartinSchou]

A 15k drive should last a company at least 3 years [...] Can an SSD (that's under high I/O) last that long?

Well, the fastest 15k SAS drive I can find clocks in at just under 400 I/O per second. That's 3.79 × 10^10 operations over its lifetime

Intel's Enterprise drives clock in at 3,300 IOPS for writes and 35,000 for reads. For the same work load, the Intel SSD would only have to survive constant writing for 133 days, or 12.5 days for reads.

Suppose for a moment, that the Intel SSD only lasts for a year before it becomes a read-only device. That's a third the life time of the Seagate Cheetah, but in that time it will have delivered between 1.04 × 10^11 and 1.1 × 10^12 operations. That's between 2.75 and 29.16 times as much work in a third of the life time.

As is obvious, the SSD option is a complete waste of money and resources. I mean - who would ever want to spend more money on something and only get upwards of 30 times the performance?

Re:I can seem some enterprise paying for this.

[ircmaxell]

I never said it was a waste of money or resources. I said that it's a great option if I/O is important. But that doesn't mean that it's great in general... If most of your needs are simply for bulk storage or the added I/O isn't worth the considerable difference in price, then SSDs may not be for you. And 30 times the performance is only worth the price if you need that performance... My whole point was that there's more to consider than just purchase price or thermal savings when looking into SSD storage...

Re:I can seem some enterprise paying for this.

[mjwx]

It's 4000 AUSTRALIAN dollars, not human dollars.

Which at the current exchange rate is only 98,309,992 US Peso's.

MM Drives are cheap today

[symbolset]

Although I endorse this approach for people with big storage needs, space and power budgets, I had in mind an application that would that would RAID 45 of them for an obscenely high IOPS + bandwidth FC node for media content storage for video work. The kind of thing James Cameron would use for shipping his in-progress movies on. I might actually go with something else, like this instead since it supports up to 70 TB in 5U and now is certified to work with normal SAS controllers instead of a proprietary switch.

Naturally at five racks instead of 5U your suggestion lacks a certain perfomance density for this application - though admittedly you do have the advantage in the $/TB area, that's not always the only consideration.

Over time SSD will become cheaper than spinning disc, and as performant as RAM. That will change many of the market dynamics and may cause some unpleasantness.

Re:Stupid

[CyberDragon777]

No, you are not paying 4K for 1 TB.

You are paying 4K for 1 TB and

* Read: Up to 260 MB/s
        * Write: Up to 260 MB/s
        * Sustained Write: 220 MB/s
        * Max IOPS: 15,000 (4K random)

What RAID under 4K can do that?

Obligatory car analogy:

Why spend a bunch of money on a Formula 1 race car when it only fits 1 driver?
You could buy several 18 wheeler trucks for that price!

Re:I can seem some enterprise paying for this.

[Jake Griffin]

US Peso's? Currently, 4000 Australian dollars = 45 559.0081 Mexican pesos so you weren't talking about those... What the crap is a US Peso?