Intel 320 SSD bug causes forum despair Users are reporting that Intel 320 SSDs are dying, and power cycle limitations appear to be at fault. The company had not issued a fix at time of writing. There is an active forum about the faulty solid state drives at the Intel Support Community website. "Poster Goose" wrote a message on 1 June, quoting from a PC Review …
The write lifetime "issue" is swept under the carpet because it's a non issue. For any reasonably large SSD (read around the 128GB mark) the point at which they'll hit the write cycle cap at *maximum write speed* continuously is around about 5-6 years down the line. Given that that's about the average lifetime of an HDD anyway, and a vast underestimate of life time, that's fine.
Their last generation of drives had the lowest reported rate of problems, though the new ones are of course new and it's early days to judge.
The write lifetime issue is a complete non-issue given realistic expectations about how long a user will have the SSD and how much will be written to it.
An SSD is the single best upgrade most users can make. Once you try one, it's impossible to go back ;)
where firmware for their SSDs is concerned - they had to pull a firmware update for their X.25 Gen II drives[1] as updated drives began to fail immediately afterwards.
It seems Intel are still getting to grips with the tricky software side of their hardware business.
1. http://www.pcper.com/news/Storage/Intel-halts-downloads-new-X25-M-firmware-due-corruption
You're absolutely right about that firmware problem, which was a pretty bad mis-step. Overall, though, they lead the table for reliability:
SSD failure rates
* Intel 0,59%
* Corsair 2,17%
* Crucial 2,25%
* Kingston 2,39%
* OCZ 2,93%
(source http://www.hardmac.com/news/2010/12/08/ssd-failure-rate-first-data-on-ssd-reliability)
I just checked... over the last two weeks, I've written an average of 3.6 GB/day to my 64 GB SSD. Assuming a conservative maximum write cycle count of 5000 per cell, that means I can write (64 * 5000 = ) 320,000 GB before the drive wears out.
So at this rate, my drive's flash memory will last at least 243 years.
I think I might upgrade before then.
SSDs are a bloody nightmare. They're expensive, they have small capacities, limited write cycles and you can't security wipe them (well, not easily).
Sure, they may be faster but the demand that I'm seeing is for users who have laden their PCs with so much shit that they hardly boot at all with a normal drive. The problem is that senior management get hold of them and go "wow" and have absolutely zero idea of the drawbacks. Until they lose all their data that is.
I've been using a 256GB Crucial drive since September 2009 and, compared to the hard drive setup I had before, it is an absolute dream; far and away the best upgrade I've ever performed.
As for write cycle limitations, then it's not going to be an issue that affects typical PC users. In any case, hard drives are prone to sudden and unpredictable failures (as I've found to my cost).
I should add that I use the SSD for system and other areas with high random I/O demands. For bulk data I use HDDs.
So one thing I can say for sure, my experience of an SSD is anything but a nightmare. It has transformed the usability of my PC.
Pretty sure this irrelevant for boot volumes. They don't get written to very much anyway. Even things like page files don't get many writes. There are system log files and things but that ought to be a very small %ge of the content. As long as the drive performs wear levelling they should last for many years.
Now for data volumes it might be slightly more concerning but even there I suspect an SSD would last as long as an HDD even if you were performing a lot of writes.
Since you appear to be running Windows, the short answer is "You don't want a swap file.". If you can afford an SSD, you can afford enough RAM to fill your motherboard and then you won't need a swap file.
There was a time when the high cost of RAM made it almost reasonable to use a hard disc to emulate a bit more of it. This hasn't been true for a decade or more.
I beg to differ. eeePC 901. 1Gb RAM, SSDs. Used for normal domestic use, watching videos, and programming.
No swap file, is perfectly stable. The only thing that tends to be crashy is Firefox, which is a memory hog at the best of times. But I get advance notice (the icons disappear) so I can close some tabs, or kill off another task or two to release more memory. Either way, in this configuration, Windows (XP, ho ho), is NOT "unstable".
I have test machines with just about every recent version of Windows. None of them have swap files. They're all what I'd call stable. (No blue screens. No unexpected application failures.) Perhaps your problems were caused by something else.
That's not to say there aren't badly written programs that go all whiny and crash if you don't let them "malloc forever". (The JVM springs to mind.) But that's an application bug rather than an instability and swap space merely postpones the problem, with the final few hours of death spent trying to saw the outer rim of your platter off.
"""you can't security wipe them (well, not easily)."""
I do many times a day, for various reasons, using the secure erase command from some ATA protocol version or another. It's actually easier to securely erase an SSD than a magnetic drive, because the bits don't have any memory like magnetic bits do, which eliminates the need to overwrite many times. And since NAND flash is actually pretty quick at erasing itself, it only takes 2 minutes, as opposed to many hours for your average spinning disk.
"""Until they lose all their data that is."""
Well, since you never trust any important data to one device, this isn't a problem. I mean you use some sort of mirrored raid and backups, right? I'm sure you've got a huge amount of experience with a large number of SSDs, and you've determined in a statistically significant manner that they're less reliable than spinning disks. Because I've done those things, and I must say that, including firmware issues, SSDs aren't significantly worse than enterprise spinning disks (Which actually aren't that much cheaper per GB than comprable SSDs.)
Sure SSDs are different, and they've got some problems, but there are certain places where they certainly make a whole lot of economic sense, and they have done for years, even before prices dropped as much as they have.
You're correct in general about the failure modes of SSDs. However, given that it is essentially a big lump of flash, it ought to be possible to sector-read from the remaining (working) cells for data recovery. I'm not sure whether or not controllers support such a thing.
Conversely, I've had harddiscs fail. Usually at boot-up, but one "just died" on me in the middle of use (started thrashing, and this was on a system that does no VM whatsoever, and went into a endless loop of spindown-spinup-spindown). Anyway, harddiscs can and do fail, and *all* the data on said disc is lost. How are SSDs different in this respect? Drive failure is drive failure...
Recipe for faster, smaller and more reliable drive:
1) Take one fundamentally dodgy technology. Ideally, this should exhibit at least 2 complex failure modes: i) random failures which exist at the point of manufacture, ii) random failures which develop over random time.
2) Add a complex software layer to mask failures from application.
3) Slowly realise that complexity reduces reliability and speed.
4) Suddenly realise that software layer makes secure deletion impossible.
5) Stop and wait for something better.
Take one fundamentally dodgy technology. Ideally, this should exhibit at least 2 complex failure modes: i) random failures which exist at the point of manufacture, ii) random failures which develop over random time.
So drive head failures, motor failures, platter damage..etc etc..
2) Add a complex software layer to mask failures from application.
Of course traditional disk drives have no error masking, that's right, you just carry on writing to bad sectors, oh hold on no you don't
3) Slowly realise that complexity reduces reliability and speed.
Yup as does bits outlined in point 1 & 2.
4) Suddenly realise that software layer makes secure deletion impossible.
Welcome to 2009.
5) Stop and wait for something better.
Yes I hear the new 286 is being launched soon.
are well understood, and by and large unrelated to usage patterns. OTOH writing to a flash cell accelerates its failure, making it inherently harder to predict. Wear levelling slows things down, adds complexity, and virtually removes any hope of secure deletion. It's a bodge to fix a poor technology.
1) You're overreacting, if anything regular HDDs are even worse at that
2) The only thing that would be better is to move the write leveling stuff and error recovery up in the stack to file system. That would require a new file system for the "most widely used" OS. As they couldn't fix their current one, I doubt they would be able to create a new one from scratch with such advanced features in any reasonable time frame.
>That would require a new file system for the "most widely used" OS. As they couldn't fix their >current one, I doubt they would be able to create a new one from scratch
Are you referring to NTFS? If so I'd like to know what it is you think needs fixing. It's incredibly reliable and resilient. I can't remember the last time I've ever had to run ChkDsk or ever lost data. Personally I'd rate NTFS as one of the best things to ever come out of Microsoft. Probably because it Dave 'I'm God' Cutler was involved :)
What does "best to ever come out of MS" say? Mmm?
Anyway, here are a few limitations:
1) 2TB practical limit.
"In theory, the maximum NTFS volume size is 2 64 clusters. However, there are limitations to the maximum size of a volume, such as volume tables. By industry standards, volume tables are limited to 2 32 sectors.
Sector size, another limitation, is typically 512 bytes. While sector sizes might increase in the future, the current size puts a limit on a single volume of 2 terabytes (2 32 * 512 bytes, or 2 41 bytes). For now, 2 terabytes is considered the practical limit for both physical and logical volumes using NTFS."
source: http://technet.microsoft.com/en-us/library/cc938937.aspx
2) "NTFS compression [or encryption] will only work on volumes with a 4KB cluster size."
source: http://blogs.msdn.com/b/ntdebugging/archive/2011/06/28/ntfs-and-4k-disks.aspx
For a list of other benefits better filesystems provide, learn about ZFS and cry.
Actually, I would be careful trusting NTFS as it is quite unsafe. Research shows that NTFS, XFS, JFS, ReiserFS, etc are all not safe.
Have you heard about ECC RAM? Why do you use it? Well, there might be spontaneous flipping bits in RAM. ECC corrects those flips. This problem also exists in filesystems:
http://www.zdnet.com/blog/storage/how-microsoft-puts-your-data-at-risk/169
And also, research shows that hardware raid has the same problems. For instance, there are papers from CERN confirming this.
Here is more reading if your data is important to you, you should read it. Here are lots of research papers:
http://en.wikipedia.org/wiki/ZFS#Data_Integrity
You've pretty much described how a modern HDD works there. The drives start life with a number of bad sectors that are mapped out by the firmware and lots of error correction bits to keep what's on there reliable. Over time sectors degrade and are mapped out from a pool of spair sectors. Mapped sectors aren't in sequence so the drive slows down when they are accessed. You can never wipe mapped out sectors.
What were you saying about SSDs again?
My experience is that the majority of hard drive failures (maybe 75%) are recoverable to a greater or lesser extent using a read-retry recovery tool like DDrescue. Also about half the failing disk drives I've seen flag up their deterioration through SMART, prior to any data loss at all.
i don't have enough SSD experence but I'd expect them to turn from datastore to brick in a microsecond, with absolutely no hope of recovering anything.
Of course, you've got backups anyway ;-)
I wish they'd stop trying to make bigger SSDs and market a small fast cheap one (under £30). I'd like to run the O/S off an SSD but keep the users' data on HD. And in my environment, that's TB RAID HD arrays on servers. No backup of the SSD or HD in the system box needed - just replace and reinstall.
I'm a firm fan of SSD and have reviewed a number of drives for Reg Hardware.
In my own PC I use a 256GB SSD for Windows 7 and applications with two HDD (1TB and 2TB) for data and am very happy with the set-up.
From a personal stand point I detest RAID on the desktop as it adds complexity and failure points however if you want the best of both worlds from SSD and HDD you might consider the Intel Z68 Core i5/i7 chipset. Z68 introduces Intel Smart Response which is a form of RAID that uses a 20GB SSD as a cache drive with software that learns how you use your PC and then copies files from the HDD to SSD to speed up access.
The technology relies on a compatible BIOS and Intel integrated RAID drivers and isn't my personal cup of tea however it works surprisingly well if you want extra performance on the (relatively) cheap.
According to this (rather good) interview with one of the guys from Kingston, when SSDs fail they tend to revert to "read-only" mode. http://www.tomshardware.com/reviews/kingston-ssdnow-ssd,2550.html
That was the final convincer for me upgrading our server to RAID-1 on SSD. Belt and braces man, me.
"TB RAID HD arrays on servers. No backup of the SSD or HD in the system box needed - just replace and reinstall."
Ha ha ha ha ha ha ha haaaaaaaaaaaaaaa
Sorry, got to love your optimism....nothing like a good flood, fire, power supply explosion,raid controller f**k up, accidental deletion, etc to test that theory.
You should learn to read before you comment.
From your quote:
"No backup of the SSD or HD in the system box needed - just replace and reinstall.""
No data to be lost on the system means that a "flood, fire, power supply explosion,raid controller f**k up, accidental deletion, etc" on the individual users' boxes won't affect the data that's all being held on the ""TB RAID HD arrays on servers." The RAID array would get backed up on its own and has nothing to do with the desktop boxes where you're putting the SSDs to good use.
Notice the issue is when the drive loses power. Keep this in your notebook with its battery charged, or in your desktop with a UPS and you're fine.
I have a Intel 320-series 160GB SSD in my MacBook Pro and I'm not concerned one bit. First, it's a laptop so it's unlikely to lose power while writing. Second, I set up wireless Time Machine backups to my NAS. Even if I lose the entire contents of the SSD, restoring it is painless.
Do you have 'reliable' references for the write lifetimes you are claiming? A pint for the first proper analysis. HDDs are well understood (they work by magick these days), flash is new and long term experience is limited. The effects of excess temperature on HDDs, for example, are documented if you know where to look. What happens if the datacentre aircon fails and the SDDs get too hot ?
high temp data retention is much smaller than ambient temp data retention - keeping your SSD cool is always a good idea.
I have a little experience with Flash in automotive applications, so your mileage may vary in SSD applications and commercial temp ranges :).
You can reasonably expect data retention to go down as your number of write cycles goes up. Flash memory specs list two separate numbers: max number of write cycles, and guaranteed data retention for a virgin device (less then 10 erase/reprogram cycles). The devices I worked with could do 1M (million) erase/reprogram cycles, and keep the data for 10 years at 150C or 20 years at 125C (die temp, not ambient), but as far as I know nobody lists any guaranteed data retention time for a device subjected to 1M erase/reprogram cycles.....
ECC may help, but if you do not check your SMART counters on a regular basis, of if the firmware does not track ECC correction cycles, you may have failures with no warnings.
"What happens if the datacentre aircon fails and the SDDs get too hot ?"
I take it from this statement that you know nothing of SSDs. However, to feed the conversation for those honestly seeking answers, SSDs can work well within the range a server room would hit if the "AirCon" goes out. I'd be more worried about your CPUs tipping over and sending garbage to the SSDs than that the SSDs would continue to function normally.
"I take it from this statement that you know nothing of SSDs."
I take it that you don't know the difference between a statement and a question, and don't have a background in solid state physics?
I know enough about solid state physics to know that behaviour of both semiconductors (SSD) and magnetic materials (HDD) is substantially temperature dependent, sometimes at not particularly high temperatures. I suspect, but have no definitive evidence to prove, that the temperatures where bad things start happening are possibly higher for HDDs than for SSDs. I know that the high temperature behaviour of HDDs is a matter of factual record. Definitive equivalent info on SSDs would be nice, especially if folks are proposing using this SSD stuff for anything important rather than for l33t gaming type stuff. If you know of anything definitive, feel free to post it.
In a bet-your-business situation I'd take gherone's answer in preference to yours, as it seems to be more fact-based than yours. But facts in general about SSD behaviour in unwanted but not that unusual circumstances seem to be in short supply so far.
For SSD write lifetimes take a look at:
http://www.xtremesystems.org/forums/showthread.php?271063-SSD-Write-Endurance-25nm-Vs-34nm
They have been performing 24/7 writing to numerous SSDs to determine just what type of lifetime they are actually getting in the real world. The results are impressive and interesting. Most of the drives are up past 200TB of writes and counting...
When deployed properly, the average SSD should easily give many years of life writing GBs per day. SSDs *DO* typically live up to their write lifetimes, if the rest of the drive lasts that long, the average computer user will run into other hardware failure or upgrades long before they typically out-write their SSD.
When purchasing SSDs, the write lifetime is not something that I am concerned about since the numbers are correct, it is the other SSD issues that worry me: compatibility with motherboard and OS, the seemingly bad firmware, the high drive controller/memory failure rate.
I am also a very skeptical SSD user. If other computer hardware failed as much as SSDs the computer industry would be in trouble. My company has almost a dozen computers and the only hardware failure in the past 10 years has been one WD 500GB hard drive and one Monitor.
I own one SSD right now, but so far I still have not replaced all of the system OS drives with SSDs just because I don't trust them at all for consistent long-term use without failure, whereas the WD hard drives in my experience are a known quantity.
I'll take that pint now...
My experience is that if the cheap Chinese junk works, for some value of "works"defined by yourself, it holds together as well as anything else. I have a 4-port KVM+Printer switch that cost £6 and works as well as a £100 one for my needs. Ditto my Tesco mouse (a Sunday distress purchase that I'd come to like better than the original by Monday). The thing is, if these things stop working, I just trash them and buy another.
No good for valuable data, though. 99.99% isn't good enough and you don't notice the 0.01% data manglement until it's too late.
When an HDD has trouble reading, there are established techniques and suppliers who will do their bestest to get the data off. They might charge, but that's the cost of not having proper backups.
When an SSD has trouble reading, what happens? In particular what happens if it's the only copy of the data?
Nothing against SSDs in the right place, e.g. for read-mostly devices and data.
Not a panacea though.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2025
