Slashdot Mirror

Yup, After a short web search I have a solid sightings for you from http://www.storagesearch.com/

1997 - in the SSD market
Bridgeworks designed a RAM SSD with hard drive backup. Sales Director - David Trossell told me - "It was a little ahead of its time and the company dropped it after poor sales."

And later, once Flash was big enough your nice little ramsan ...

Houston, Texas - July 22, 2008 - Texas Memory Systems today launched the world's fastest SSD - the RamSan-440,

The RamSan-440 is a 4U rackmount fibre-channel connected RAM SSD with upto 512GB of storage capacity. It can sustain up to 600,000 random IOPS and over 4GB/second of random read or write bandwidth, with latency of less than 15 microseconds.

the RamSan-440 is a 4U RAM SSD delivering 600,000 random IOPS - click for more info

It's the first RAM SSD to use RAIDed flash memory modules for data backup (instead of hard disk) and the first system to incorporate Texas Memory Systems' patented IO2 (Instant-On Input-Output) technology.

While limited writes are certainly a factor, they probably aren't going to be a major issue for basic consumer use.

Most SSD storage drivers these days automatically spread the writes around the drive, so to hit the write limit you will need to write the equivalent of the capacity of the drive multiplied by the write limit of any particular register. Assuming 2 million write cycles per register, and the low-end 256 GB drive, that's 500,000 TB of writing before you burn out every register. Obviously the user would see some degradation before that, but there's still lots of room to play with.

Some more sample calculations are available here: http://www.storagesearch.com/ssdmyths-endurance.html

As a sanity check - I found some data from Mtron (one of the few SSD oems who do quote endurance in a way that non specialists can understand). In the data sheet for their 32G product - which incidentally has 5 million cycles write endurance - they quote the write endurance for the disk as "greater than 85 years assuming 100G / day erase/write cycles" - which involves overwriting the disk 3 times a day.

That's for old-ish tech and a smallish drive. For consumers, large drives get written to exponentially less. Consider, the vast bulk of consumer "big" drives are and movies. These are big, chunky files that don't get overwritten very much. As a consequence the vast majority of your drive stays clean. For most people, they'll want or need to buy a new hard drive long before the old one wears out. Please read up on the facts before spouting nonsense.

Okay, for everybody's information, I looked it up. According to this recent article, write-cycles are still the limiting factor for most SSDs, rather than MTBF. (Certain RAID configurations are an exception.)

However, the good news is that the article says write endurance has increased a lot in the last few years, with some manufacturers offering SSDs rated at more than 5 million write cycles.

So, the question in my case is not "Will a cheap SSD do the trick?" but "Do I want to spend the money on a high-end SSD?"

Other than a few minor mistypings (SDD instead of SSD, pci-e instead of PCIe), I see nothing wrong.

Yes some do 3.5 http://www.storagesearch.com/ssd-3p5.html

Yes

What you want is cheap 5U rack servers with either OpenFiler or FreeNAS. Personally, I like openfiler better. iSCSI is going to be the way to go unless you want a thick OS on the server and all the other admin issues that come with that. Plus, with openfiler you can still do block level snapshotting and change replication. Also, I've heard good things about Open-e as well. And if you want to mess with ZFS, there's OpenSolaris.

What you do is get yourself a huge (4 or 5U) barebones server from newegg or a cheaper place. Make sure to get a couple of good SATA RAID controllers. Not FakeRAID! SAS would be better, but the drives are a lot more, even for the nearline drives that are basically SATA drives with a SAS interface. Adaptec makes some real SATA raid cards, and there's 3Ware as well. You don't have to worry a lot about the cache, but if it isn't battery backed, you're going to write though it anyway. Who cares, you have 16 spindles! Load it with a bunch of drives. They don't have to be the biggest, anyway more spindles means more performance. 16 500GB drives is going to be fine, for instance, because then you can take 1/3 of that for RAID 6, have some hot spares, etc. Get the slowest drives you can, maybe get a little SSD to use as a boot drive (there are small ones for around $100). You could even boot from a USB key if you feel like the hassle. You don't need a ton of processor. A celeron would probably work, but you probably do want something 64 bit so you can put a bunch of RAM in it as you get more advanced.

Also check out Storage Search. Not a very well designed site but tons of goof info under iSCSI and SAN and NAS.. If you're rich, you might try out an EqualLogic, they are around $28,000 for 8TB but pretty slick.

If that's indeed the case, then why not simply put the MBR, /boot, /bin, and /usr on the SSD, then mount stuff like /home, /tmp, swap, and the like onto a spindle disk? No algorithm needed, thus no overhead needed to run it, etc.

This sounds like a good way to go.

...only if you want to blow out the SSD wear-limits.

Thanks to wear leveling, your HDD would probably wear out first. This guy did the math.

Read "Flash SSD Application from Hell - the Rogue Data Recorder". And keep in mind it was written in 07 - things are better than that now. You might die of old age before your SSD does, depending on your setup.

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.

I doubt a good SDD will burn out anytime soon:
http://www.storagesearch.com/ssdmyths-endurance.html

I'll sum it up for you, a 64GB drive constantly being written to at 80MB/s will burn out in approximately 51 years.

There is something better. In every measurable way SSD is better. Better power consumption, better shock/drop resistance, lower heat, no noise, near zero seek times, they already can run SATA3, and longer lifetimes. Yes - longer lifetimes. Check this guy's math and you'll see what I mean.

The only thing that keeps them off the mainstream currently is price, which Moore's law will quickly fix.

I've always been a slow adopter of new technology. Hell, in high school I thought CDs were a passing fad. I can admit that. But this though...this one is easy. I don't think I've ever seen a technology upgrade that so completely outshined its predecessor in my life.

I agree that spinning disks will have their place for a while, much in the same way ice houses were still needed when the first refrigerators were being made. Only necessary during the transition to the better tech. Likewise, if the price were to match rotating media dollar for dollar today - you'd never see a spinning hd ever again, except in legacy systems.

SSDs offer speed. Spinning Disk HDDs offer cheap space. Hybrid disks offer a nice compromise until SSDs overtake spinning disks in storage/price.

I mean really, who needs an expensive big SSD for your porn collection? Unless you have 12 monitors running porn simulcasting...SSD speeds are really only needed for heavily accessed files. HDDs offer cheap storage for those not-so-often used files. The solution is relatively inexpensive, and here today

Lot of companies are making SSD's
http://www.storagesearch.com/ssd-2p5.html
Just a few for your desktop.
At the consumer end its a cartel. Nobody is going to drop the $$$$ flow too fast.

That limitation is effectively irrelevant. And that article is from 2006. And it accounts for the case of constant, high speed writing; hibernating out to disk happens a few times a day, and as such is barely noticeable.

Think about it. Lets say you've got a crappy flash drive: Only 1 million writes per sector. And it's exactly the size of your main RAM, so no wear leveling algorithm will help. If woke up and hibernate that machine 100 times per day, it would still last 10,000 days, or 27.4 years. That's slightly longer than I've been alive; I think I could live with a hard drive that "only" lasted that long.

Ah, found one of my original articles, oddly the one corroborating it 404s now. please, read it and make your own conclusions. http://www.storagesearch.com/ssdmyths-endurance.html

No.

Before anyone complains about ssd wearing out quickly, please read here.

No.

Before anyone complains about ssd wearing out quickly, please read here.

Actually it's not a problem if you use SLC chips. This paper works out the life with wear levelling at 51 years

http://www.storagesearch.com/ssdmyths-endurance.html

Cheap MLC disks don't have a long lifespan if you write to them flat out. Using the same formula the same guy works out a lifetime of 6 months for an MLC disk.

http://www.storagesearch.com/ssd-slc-mlc-notes.html

Actually it's not a problem if you use SLC chips. This paper works out the life with wear levelling at 51 years

http://www.storagesearch.com/ssdmyths-endurance.html

Cheap MLC disks don't have a long lifespan if you write to them flat out. Using the same formula the same guy works out a lifetime of 6 months for an MLC disk.

http://www.storagesearch.com/ssd-slc-mlc-notes.html

Actually even if you do do use it for swap, or some application that writes to out absolutely flat out the lifetime is less than Intel quotes. If I use the formula here

http://www.storagesearch.com/ssdmyths-endurance.html

2 million (write endurance) x 64G (capacity) divided by 80M bytes / sec gives the endurance limited life in seconds.

I can work out how much less.

If you substitute the figures Intel gives for write endurance (100000), capacity(80GB) and write speed(170MB/sec) you only get 1.5 years. Bear in mind that's for an application that writes flat out at 170MB/sec 24 hours a day.

The odd thing is if you compare it to the X25-M. Write endurance is 10x less at 10000, write speed is lower at 70MB/sec. There I get 0,37 years. Mind you with SLC memory being 1/3 the price you could just buy three times as much of it. That way you get 3x the storage and a lifesapce of 1 year absolute worst case. SLC actually seems like a better choice for most people.

Incidentally, this really is a worst case, hopefully no real world application can saturate write bandwidth like this.

It would also make sense to gradually decrease the write bandwidth so the drive slows down in its old age but takes longer to die. Throttling write bandwidth to 70MB/sec on the SLC drive would give a life of 3.7 years. Throttling to 70MB/sec after half the writes were used up for an average write rate of 120MB/sec would give you 2.16 years. You could imagine a sort of Zeno's throttling algorithm (50% bandwidth at 50% life, 25% bandwidth at 75% life and so on) where the write bandwidth keeps dropping so the drive slows down but never actually dies.

These guys are idiots. A few points:

- They 'cheated' on ATTO, only configuring it to start at 8k. Last I checked, default sector size is 512b. Regular day-to-day apps, such as Outlook, perform random sector-level access to the PST when downloading mail.

- If you're going to do an SSD roundup, how about at lest grabbing a few drives off of the SSD top 10. Specifically, Memoright (#1 on that list) makes an SLC drive that competes with the other SLC drives on price, yet outperforms them all: http://www.storagesearch.com/ssd-top10.html

- Disclaimer: I own a Memoright drive. I don't claim to be a fanboy, I just did my research beforehand (along with trying out a few other drives), and found the best thing going at the time.

- The Intel drives, expected to come out this month, are likely to bury everything on that review.

If you have a power failure during a write to an SSD, you are very dependent on the FTL (Flash Translation Layer) between the FS and the device: if it does its job properly it can recover from this, detecting blocks that are invalid because they were partially written. If not, the whole device can be unrecoverable... This is one reason why using an SSD in a laptop (i.e. with battery) or a server with UPS is a good idea.

Having looked at the very long lifetimes of most flash devices (see http://www.storagesearch.com/ssdmyths-endurance.html) I would only use ext2 instead of ext3 if performance was critical and I had good backups on another system. The main issue with ext3 is likely to be performance.

Here's a good discussion of all this from an SSD website that looks at all the developments in this area: http://www.storagesearch.com/ssdmyths-endurance.html

Basically, because write speeds are quite low on flash drives, and capacities are now quite large, even if you continually overwrite all blocks, you can't wear out the whole drive (subject to a wear levelling flash translation layer [FTL]) of course) in less time than a similar hard drive would fail. Writing intensively to just a few blocks is actually easier for the FTL to handle, it just re-maps you to another block.

For enterprise level storage it would be wise to use RAID 1 over the flash drives, just as I would with hard drives, but for other applications you are reasonably safe without mirrors. Of course, you always need backups, to help guard against application bugs, user error, etc.

No, this statement will not be put to bed, because it is based on facts - measured physical quantities. And here's one thing to ponder: if an application writes to the disk 100 times per second, how much will your 4GB SSD going to last? If you have only 1GB of space left, then wear leveling can only count on the blocks that don't contain data. And if the blocksize for the Flash RAM device is 128KB (which is typical, but there are also 256KB Flash RAMs), then the number of blocks you can spread out the writes is 8192. If the SSD is based on MLC Flash (as is, sadly, becoming typical) then you can write up to 10.000 times per block. Assuming perfect wear leveling, the device will last less than 819200 seconds which is 9 days and a few hours.

You fail to consider several things:
1. Static wear levelling/leveling rotate the blocks being written to so both "empty" and "full" blocks are being used, so the amount of free space on the filesystem doesn't matter.
2. The 100.000 writes often quoted are a guaranteed statistical _minimum_, not a average or a maximum. According to some sources the typical cell will endure 200K-1M writes.
http://www.solidkor.com/en/technology/414we.html
3. A typical SSD has spare blocks (just as HDD have spare blocks). So when a block is toast it is just marked as "bad" and a spare block is used instead.
4. Let us not forget ECC schemes that may extend the life of a block significantly.

All this adds up to a considerable lifespan for SSD's.
Let's for arguments sake say that that a SSD has 1 megabyte of spare blocks per 1 gigabyte storage. So if one were to read continuously to one 128 kilobyte block it would take:
(500k writes=assumed lifespan of a block)*(8 the number of spare blocks)=4M writes, and still not a single block lost in the sense that the filesystem/OS still sees 1 gigabyte of storage.

But read more:
http://www.storagesearch.com/ssdmyths-endurance.html

--
Regards

Although the article *is* low on details, there are a few clues as to what the problem is:

The next generation of SSDs will use multilevel cell (MLC) technology, which will require a more sophisticated controller--a crucial component in solid-state drives.

IIRC, MLCs do have a much lower number of allowable writes, so this could actually be the problem!

That used to be true with SLC chips. It's not true with MLC.

http://www.storagesearch.com/ssd-slc-mlc-notes.html

I have seen this web site before, and it is very suspicious. It looks like someone trying to promote SSDs (perhaps because they sell them?) The 10K-100K writes he quotes are true, and available from many sources. But he uses 2 or 3 millions of cycles for his numbers based on "tests" he has done. I have not seen any such "tests" elsewhere.

Why would the manufacturers stick to 100K writes if they could claim millions? And what kinds of tests can he do that the manufacturers can't?

Slashdot makes me laugh. Originally people here worried about flash memory and people worked out the life with wear levelling. And it was indeed fifty years.

http://www.storagesearch.com/ssdmyths-endurance.html

For ages and ages I see people saying "but you can't put swap on flash it will wear out" and I post a link to the above

Now the "it will last for fifty years even if you hammer the shit out of it" meme seems to have penetrated the hive mind. And Intel and Samsung and others have launched flash disks. But those disks are based on MLC flash, not SLC like in the first example. So you need to update the calculation.

http://www.storagesearch.com/ssd-slc-mlc-notes.html

Oh noes! Your data will disappear in 6 months!

Slashdot being slashdot by the time this meme becomes common, something in the calculation will change since the industry knows that 6 months is too short and the worst case lifetime will go up to a couple of decades again. Or more than the 5 year average life of a hard disk at least.

Slashdot makes me laugh. Originally people here worried about flash memory and people worked out the life with wear levelling. And it was indeed fifty years.

http://www.storagesearch.com/ssdmyths-endurance.html

For ages and ages I see people saying "but you can't put swap on flash it will wear out" and I post a link to the above

Now the "it will last for fifty years even if you hammer the shit out of it" meme seems to have penetrated the hive mind. And Intel and Samsung and others have launched flash disks. But those disks are based on MLC flash, not SLC like in the first example. So you need to update the calculation.

http://www.storagesearch.com/ssd-slc-mlc-notes.html

Oh noes! Your data will disappear in 6 months!

Slashdot being slashdot by the time this meme becomes common, something in the calculation will change since the industry knows that 6 months is too short and the worst case lifetime will go up to a couple of decades again. Or more than the 5 year average life of a hard disk at least.

Regular flash works just fine for swap. If you write nonstop at top speed to a standard chip, you'll wear I'd out in about fifty years. Thus I don't understand why we should care about an even longer lifetime.

That used to be true with SLC chips. It's not true with MLC.

http://www.storagesearch.com/ssd-slc-mlc-notes.html

It's a simple matter to plug new data for MLCs into the calculation I did for the worst case wear-out process for flash SSDs - which I called the Rogue Data Recorder.

Instead of the 64GB example I used then, I'll assume the MLC SSD has 128GB capacity. MLC SSDs have more capacity than SLC. And more capacity means longer operating life - before cells wear out.

I'll still use the 80M bytes / sec sustained write speed - because the fastest MLC products (in Feb 2008) can already do that. (Meanwhile the fastest SLC products have moved up in the world and are about 50% faster.)

The next factor is where we hit the big problem... Instead of a write endurance rating of 2 million cycles (for the best SLC) - I can only use a figure of 10,000 for MLC. MLC has a much lower rating due to the complex interaction of discriminating multiple logic levels reliably coupled with the intrinsic failure mechanism of wear-out.

Plugging these numbers in the same calculation gives an estimated MLC flash SSD operating life (at max write throughput) which is 6 months! (instead of 51 years for a 64GB SLC SSD).

All the affordable SSDs I've seen from Intel and Samsung are based on MLC flash because it costs much less per bit. Down to $2 per GB in fact. SLC currently costs 2-4x as much. E.g.

Here are the average prices for flash

32Gb 4Gx8 MLC 9.27
16Gb 2Gx8 SLC 15.61
16Gb 2Gx8 MLC 3.97
8Gb 1Gx8 SLC 6.31
8Gb 1Gx8 MLC 2.34

SLC is 2.7x more expensive for 1Gx8 and 3.9x more expensive for 2Gx8. So it's not surprising that most SSDs are MLC based. But if you write at full speed to them they will die very quickly.

Incidentally look at the price of 4Gx8 MLC. $2.31 per gigabyte. Pretty damn cheap.

Actually SSD wear is really not a big issue as long as you have a reasonable FTL (flash translation layer) in the SSD, which all reputable ones do. Basically you'd have to be writing continuously to the entire disk for many months or years, 24/7, and in reality people and apps don't work like this - people sleep, and apps read far more than they write.

See http://www.storagesearch.com/ssdmyths-endurance.html and also the eee forums which have an FAQ on this - http://wiki.eeeuser.com/ssd_write_limit has some good eee specific stuff.

The fact that flash drives have low write speeds is a saving grace incidentally - as they speed up this may be more of an issue, but they are also getting larger meaning more spare blocks for the FTL to swap in (as with hard disk bad block management, sort of).

Try this list: http://www.storagesearch.com/ssd-fastest.html

A typical source is something like this. This is not looking at typical use: it's looking at writing to the drive constantly. In the worst case scenario, the drive would last 51 years. That article is out of date too. Flash drives could potentially last hundreds of years in the worst case these days.

This is not to say that modern flash drives last hundreds of years, just that the number of rewrites is not the limiting factor. It doesn't matter how little free space there is on the drive, what access patterns you use, etc. Take a modern flash drive with no free blocks and write and rewrite to it non stop. It should last at least a few decades before the number of rewrites starts to become an issue.

Actually the eee community has discussed this many times, and the flash world has examined this in depth - it seems that with realistic usage patterns and reasonably good wear levelling there's much less risk of flash wearing out (see http://forum.eeeuser.com/viewtopic.php?id=7077 and in particular the link in first post to http://storagesearch.com/ ).

However, a system that minimises flash writes will also perform much better, since flash is rather slow at present for writing compared to hard disk or RAM.

Hmm...
I just found this:

Unlike DRAM, flash memory chips have a limited lifespan. Further, different flash chips have a different number of write cycles before errors start to occur. Flash chips with 300,000 write cycles are common, and currently the best flash chips are rated at 1,000,000 write cycles per block (with 8,000 blocks per chip). Now, just because a flash chip has a given write cycle rating, it doesn't mean that the chip will self-destruct as soon as that threshold is reached. It means that a flash chip with a 1 million Erase/Write endurance threshold limit will have only 0.02 percent of the sample population turn into a bad block when the write threshold is reached for that block. The better flash SSD manufacturers have two ways to increase the longevity of the drives: First, a "balancing" algorithm is used. This monitors how many times each disk block has been written. This will greatly extend the life of the drive. The better manufacturers have "wear-leveling" algorithms that balance the data intelligently, avoiding both exacerbating the wearing of the blocks and "thrashing" of the disk: When a given block has been written above a certain percentage threshold, the SSD will (in the background, avoiding performance decreases) swap the data in that block with the data in a block that has exhibited a "read-only-like" characteristic. Second, should bad blocks occur, they are mapped out as they would be on a rotating disk. With usage patterns of writing gigabytes per day, each flash-based SSD should last hundreds of years, depending on capacity. If it has a DRAM cache, it'll last even longer.

1997 called, it wants its caveat back.
http://www.storagesearch.com/ssdmyths-endurance.html http://www.bitmicro.com/press_resources_debunking.php

Please stop spreading FUD. An out-of-date flash drive, being written to CONSTANTLY (under MUCH harder conditions than just swapping) should last for no less than 51 years. For modern flash drives, that number should be in the hundreds of years.

This is not to say that flash drives last forever, or even that they last 51 years. This is to say that, no matter what you're doing, there is nothing in this world perverse enough you could be doing which will make a modern flash drive level out. As mentioned in a parent post, the limiting factor in flash drives these days is NOT how many writes it can do. It's probably not too hard to abuse a flash drive to render it useless, but writes are NOT the limiting factor.

Will this ever die? The write cycle counts in modern flash is in the millions now. Doing the math you very easily get 20+ years before write cycle wear is a concern: http://www.storagesearch.com/ssdmyths-endurance.html

How many heavily used spinning drives do you know that last even 10+ years?

I'm nearly positive that the read/write cycle issues have long been moot. http://www.storagesearch.com/ssdmyths-endurance.html I realize this is an industry sponsored site but even with taking very pessimistic views of their numbers a flash drive will last far longer than most disk based drives on the market will.

This is the link you want. Basically, your problem is assuming only 100k writes. 100k writes was typical for 1997; the page I linked to gives a figure of 2M writes, but I would say even that is very conservative these days.

So the link I gave gave a (conservative) number of 51 years. The point isn't that a flash drive is going to last 51 years. The point is that a modern flash drive is not going to wear out from write exhaustion. Basically flash memory today has reached a point where the number of rewrites allowed is not the limiting factor in its engineering.

It appears as though I am behind the times.
We are both talking about wear levelling, however I was referring to something known as Dynamic wear levelling which takes into account only the free portions of the disk whilst you are talking about a more advanced Static wear levelling.

Static wear levelling as you have described will move blocks of allocated data around and give ~4 times longer lifespan than the dynamic (space sensitive) levelling.

I read a very nice report about it here.

According to this article, it would take decades before the write limits are reached on today's SSDs.

Write endurance

In the case of 32GB capacity Mtron SSD: >85 years @ 100GByte / day erase/write cycles

http://www.storagesearch.com/ssd.html

The way that SSD oems deal with the management of write endurance internally within their products varies but they all have the common theme of scoring how many times a block of memory has been written to, and then reallocating physical blocks to logical blocks dynamically and transparently to spread the laod across the whole disk. In a well designed flash SSD you would have to write to the whole disk the endurance number of cycles to be in danger.

For this illustrative calculation I'm going to pick the following parameters:-
Configuration -- a single flash SSD. (Using more disks in an array could increase the operating life.)
Write endurance rating -- 2 million cycles. (The typical range today for flash SSDs is from 1 to 5 million. The technology trend has been for this to get better.)
Sustained write speed -- 80M bytes / sec (That's the fastest for a flash SSD available today and assumes that the data is being written in big DMA blocks.)
Capacity -- 64G bytes (Actually single flash SSDs are available with 160G capacity in 2.5" form factor from Adtron and 155G in a 3.5" form factor from BiTMICRO Networks. The bigger the capacity - the longer the operating life - in the write endurance context.)

To get that very high speed the process will have to write big blocks (which also simplifies the calculation).

We assume perfect wear levelling which means we need to fill the disk 2 million times to get to the write endurance limit.

2 million (write endurance) x 64G (capacity) divided by 80M bytes / sec gives the endurance limited life in seconds.

That's a meaningless number - which needs to be divided by seconds in an hour, hours in a day etc etc to give...

The end result of 51 years for the absolute worst case possible.

I believe you're right. Lets do some simple calculation:

This HD has 600 000 hr MTBF max and transfer rate of 300MB/s.
Say you transfer at this rate for all its MTBF life thats
600 000 hr * 3600s/hr * 300MB/s = 648 000 000 000 MB or 6.48 petabytes max transfer in its life

MTBF(sec.) * Transfer rate = Max lifetime transfer

Normal flash has 300,000 write cycles/block amortized among N GB blocks (block = byte) so thats 300,000 * N cycles with (comparable)max rate of 300MB/s or 90 000 000 * N MB max transfer in its lifetime.

(WC * n blocks) * Transfer rate = Max lifetime transfer

MTBF = WC * N blocks

N = MTBF/WC -> higher write-cycle flash needs less space to approach MTBF of HDs

N would need to be about 64 800/9 MB or ~7.2 GB capacity to approach the lifetime transfer of a hard drive. I realize that with each failing block, size of the flash would be shrinking, but the total failure of the flash would occur when no blocks could be written to. This is just a rough comparison of the two. Also I assumed the same transfer rate as the HD which is not very realistic, but its there to show a rough comparison.

As you can see a 100GB or so flash disk with a 1,000,000 block write-cycle would more than surpass the MTBF in current Hard Drives.

I think the ultimate point will be the $/GB. I don't think that Flash can compete with Hard Drives in that department, not yet anyway. Perhaps we will see them as a buffer(it's already being used in Vista) between Hard Drive and main memory. And eventually(!?) overtake HDs. We can always hope..

"The typical access time for a Flash based SSD is about 35 - 100 micro-seconds, whereas that of a rotating disk is around 5,000 - 10,000 micro-seconds. That makes a Flash-based SSD approximately 100 times faster than a rotating disk. " http://www.storagesearch.com/bitmicro-art3.html

"The SSD found files more than twice as fast, and accelerated boot-up. Its cumulative speed advantage over the other two drives was an impressive 25 percent" http://www.pcworld.com/article/id,126833/article.h tml

Solid state drives (flash drives) are not slower. If they were, hard drive manufacturers would not be using flash to make hybrid drives.

No, you're wrong. The write leveling is done by the controller on the USB thumb drive itself, not the computer it's attached too.

Ref: http://www.storagesearch.com/siliconsys-art1.html

so if you are actually using most of your 20GB drive (which isnt very difficult to achieve), then you will wear out one section a lot faster.

Wear-leveling algorithms can take this problem into account. According to this article on Solid State Disks (SSD):

When a given block has been written above a certain percentage threshold, the SSD will (in the background, avoiding performance decreases) swap the data in that block with the data in a block that has exhibited a "read-only-like" characteristic.