SSDs and crunching...reliable?

[123bob]

At the risk of this being discussed before, I have to report that I'm not having very good luck on SSDs lately for cruncher rigs. I own three SSDs that I tried for an experiment. They are all a little less than 2 years old.

I lost two of the three. While I realize this is nowhere near a good statistical sample, I can't help but wonder. What was strange is that both drives failed within 2 weeks of each other. Different rigs, different power protection and UPSs....

The first failure was a Patriot 64gig unit that was part of my main rig, RAID 0 Array. Big mistake, I will NEVER do it again. (second time RAID 0 bit me....) It was dead as a doornail. I did image the rig, but that was last year. Thankfully, I have been so busy that no major stuff was lost on that rig. The image was still pretty up-to-date. It has a RAID mirror on Seagate server drives now.....

The second was on an OCZ 30gig unit. That was Farm-13. I stuck an old hard drive on it and reconstructed the crunching package. No real biggie there either.

The issue I have is that I have hard drives that have been crunching for YEARS, 24/7/365 without fail. In fact, I have never lost a hard drive to crunching. <knocks-on-wood-real-hard>

Anyone else seeing trouble? Remember, I have almost two years on these SSDs 24.7.365. Am I asking too much out of them? IDK.

Regards,
Bob

EDIT: I also realize that two years is a long time ago. Has the reliability improved? I would think so...

[Kryja]

What os were you running bob?

To quote wiki;

"Flash-memory drives have limited lifetimes and will often wear out after 1,000,000 to 2,000,000 P/E cycles (1,000 to 10,000 per cell) for MLC, and up to 5,000,000 P/E cycles (100,000 per cell) for SLC"

Also, depending on your os, lack of trim may have been a factor.

[123bob]

This was before TRIM, and the op sys was Vista 64bit in both cases. That may have been a definite contributing factor.

Bob

[Kryja]

Hm, okay. I've got a magazine somewhere where they battered a number of ssd's with both many large files and numerous small files both with and without trim. I'll dig that out, have a read and report back.

[Jaco]

I think it's recommended to change the setting "write to disk every ... minutes" .
Don't have any SSD's so i may be totally wrong

[Kryja]

Managed to find it. The magazine is Custom pc May 2010, they tested an OCZ vertex 120gb, Corsair P128 128gb & an Intel x25-m 160gb rev 2.

The OCZ uses the Indilinx barefoot drive controller which is also in the Patriot Torqx so that covers your two hopefully.

Give me a minute and I'll read through it all.

[jcool]

I think it's recommended to change the setting "write to disk every ... minutes" .
Don't have any SSD's so i may be totally wrong

You're not wrong Jaco. I change it to "every 600s at most" from the default of 60s with SSDs.
Crunching can definitely shorten an SSD's lifespan. However, if your SSDs were 2 years old an non-Intel, I am going to assume they were all Jmicron-based. I've seen an RMA quota of as high as 60% for those Jmicron SSDs from my customers, so even with normal usage they will most likely fail within 1-2 years.

With any current SSD you'll be safe, at least if you raise that "write to disc" setting by a factor of 10 or more.

[Kryja]

While they go into heavy detail about the benefits of having trim and the negative effects on read, writes, sequential writes etc without, there is no mention of untimely death due to too much use without trim support.

It's quite evident from their testing though that any ssd should be run with trim to ensure longevity.

But as Jcool said, it's probably that yours were using the Jmicron controller as they were bought two years ago.

[123bob]

I ran this with write to disk at 999 seconds on the profile. (EDIT: for the entire two years.) They are pretty old SSDs though. So the old tech idea is probably right.

I only bring this up as a concern/observation here. I would really suggest that if any of you are crunching hard on a main rig with an SSD RAID 0 array, you might want to consider imaging often.... I didn't, and got lucky only because I was too busy doing other stuff to miss my main rig for a year....

My immediate solution was to go way conservative the other way on the main rig. I went with drives that are rated some un-godly MTBF. I had those around, so in they went in a mirror. It is imaged freshly too...

Bob

[jcool]

I've been crunching on an SSD Raid 0 (mainrig) for a while now, no issues or even performance degradation I can see.
These are cutting-edge 3rd generation SSDs though, which (thankfully) have become rather affordable. Just trust me when I say that the old Jmicrap-based SSDs were garbage. I've RMAed more Patriot Warp V1 & V2's than I care to count, and some of them didn't even last half a year. I actually blew through 2 or 3 32GB Warp V2's myself on crunching rigs.

[kjeldoran]

I have been crunching with my 60GB OCZ agility since February. I haven't run into any issues and I don't think I will run into any in the future. The drive has a 3 year warranty and I think by the time my drive does wear out due to too many write cycles I will have purchased a newer model.

[Biker]

I have been running Intel X25-V 40GB drives on my crunching rigs since their launch with no issues.

To maximise longevity I have implemented the following:

1: Only used 30Gb for the partitions to greatly increase spare area.

2: Made sure they are all correctly write block aligned.

3: Disabled the page file.

4: Increased Boinc write to disk to 600 from default.

I am using XP x64 so no TRIM available but the drives are fine without it for this usage profile.

I made a huge thread ages ago about the Jmicron drives and have to agree they were total garbage

I have seen reports and had returns of quite a few Ilindrix and Sandforce drives dying very quickly but have seen very few (one total if my memory is currently working ok ) of the same reports of Intel G2 drives falling over so if I was buying again I would still go for the little Intels...

[Bobsama]

I've done some real damage to the SSD's on an EeePC--these drives were from before the first desktop SSD's became affordable. Paid out my nose and feel like an idiot now, though. They still work, but they're much slower even after clean installs.

[Johnmark]

I have been running Intel X25-V 40GB drives on my crunching rigs since their launch with no issues.

To maximise longevity I have implemented the following:

1: Only used 30Gb for the partitions to greatly increase spare area.

2: Made sure they are all correctly write block aligned.

3: Disabled the page file.

4: Increased Boinc write to disk to 600 from default.

I am using XP x64 so no TRIM available but the drives are fine without it for this usage profile.

I made a huge thread ages ago about the Jmicron drives and have to agree they were total garbage

I have seen reports and had returns of quite a few Ilindrix and Sandforce drives dying very quickly but have seen very few (one total if my memory is currently working ok ) of the same reports of Intel G2 drives falling over so if I was buying again I would still go for the little Intels...

So I finally decide to buy one of these drive's (3 days ago) and now I'm wondering if it was worth it? and/or made any mistakes installing it?

I've held off for a long time due to all the failures I read about, but at 1 million hours MTBF I figured the thing should last at least a few years. After all how are they testing these drives to get such high MTBF, if they are failing after a fraction of that time, they should last for my lifetime?

Saw the Kingston SNV45-S2/128GB on sale, picked her up, used Acronis True Image Home to Automatically copy the image from a 300 gig WD Raptor and am using Win7.

Yup it is definitely fast, but the Raptor is no slouch.

So did I make a mistake, as this is my main machine and a cruncher as well, usually create a backup disk image once every two weeks.

Some thoughts:
-JMicron JMF618 controller but branded as Toshiba.
- Didn't do a fresh install.
- Should I have used special setting in Acronis to transfer the disk image?
- Have not changed any windows or BOINC settings.
- 2: Made sure they are all correctly write block aligned ??

Seems to me after reading this thread that these drives are still more trouble than they are worth!
Heck I've still got 40 gig enterprise drives with 6+ 24/7 years on them and they still work fine.

As this thread has me doubtful about SSD drives for a machine that has a tonne of software installed and is my main rig, I'm thinking I should just go back to the Raptor for this machine and play with the SSD in a cruncher only rig.
Can't really "just" stick the Raptor back in as BOINC has a large que that has changed as well as a few other things.

Is there anything I need to be aware of when transferring the SSD image back to the Raptor?

TIA

[123bob]

It's good to hear that the newer SSDs have ratings like 1M MTBF. I brought this up with the older SSDs to hopefully save some folks grief. It is good to know others are seeing similar failure rates on them.

Just think about risk if you're crunching on the old SSDs with an important rig. At least image the rig.

One of the drives I lost was on a dedicated cruncher. So it was no big deal just to start over with my standard farm package, on a spare drive.

The RAID 0 in the main rig was a different story. I have LOTS of custom stuff on that. It would have been a major PITA to have lost it totally. Fortunately, I had the rig setup for op sys and apps on the RAID 0, and all my data files on a separate hard drive RAID 1 array. I had not made any major additions to apps in the past year, so the old image I restored was fine.



I did some further testing on the OCZ drive I lost. I conducted similar testing on the Patriot drive I lost earlier, with similar results. See Picture below.



This was the standard "tie test". The unit is tested for performance against a standard auto body hammer and railroad tie. It should be noted that this drive did remarkably well in the test. It only shattered when the spike end of the hammer was tested.

I'm thinking it might be time for an RMA?

[Otis11]

I'm thinking it might be time for an RMA?

Send that in... And imagine the look on their faces when they open it.

[jcool]

@Johnmark: The JMF618 is a new controller and supposedly works very well. The old (bad) one was the JMF602, which also suffered from all those stuttering issues.

@Bob: A Vertex? Damn... that one was Indilinx-based, so actually NOT a POS. Must have been bad luck then.. I have gotten 2 or 3 RMAs from Indilinx-based drives, but nowhere near the old Jmicrap ones.

[CaptMorgan]

Should I be worried Bob that is the exact same one I have? I can easily conduct tests like that but i would prefer not too. I really hope it does well for me, crosses my fingers, but I am sure glad the tech is getting better along with the prices

[jcool]

Okay folks, I did a write-up on the subject here: http://www.xtremesystems.org/forums/...d.php?t=260883
I once did something similar right before the XS server crash, but all the info was lost.. I was pretty pissed about that, but figured I could do it again. This time even more detail

[kjeldoran]

So I finally decide to buy one of these drive's (3 days ago) and now I'm wondering if it was worth it? and/or made any mistakes installing it?

I've held off for a long time due to all the failures I read about, but at 1 million hours MTBF I figured the thing should last at least a few years. After all how are they testing these drives to get such high MTBF, if they are failing after a fraction of that time, they should last for my lifetime?

Saw the Kingston SNV45-S2/128GB on sale, picked her up, used Acronis True Image Home to Automatically copy the image from a 300 gig WD Raptor and am using Win7.

Yup it is definitely fast, but the Raptor is no slouch.

So did I make a mistake, as this is my main machine and a cruncher as well, usually create a backup disk image once every two weeks.

Some thoughts:
-JMicron JMF618 controller but branded as Toshiba.
- Didn't do a fresh install.
- Should I have used special setting in Acronis to transfer the disk image?
- Have not changed any windows or BOINC settings.
- 2: Made sure they are all correctly write block aligned ??

Seems to me after reading this thread that these drives are still more trouble than they are worth!
Heck I've still got 40 gig enterprise drives with 6+ 24/7 years on them and they still work fine.

As this thread has me doubtful about SSD drives for a machine that has a tonne of software installed and is my main rig, I'm thinking I should just go back to the Raptor for this machine and play with the SSD in a cruncher only rig.
Can't really "just" stick the Raptor back in as BOINC has a large que that has changed as well as a few other things.

Is there anything I need to be aware of when transferring the SSD image back to the Raptor?

TIA

Well, to be honest I think a fresh install is probably one of the best options. I know I had issues when I swapped motherboards with my win server 08 R2 installation. The SSD was overall slower before a fresh install.

By the way, SSB MTBF is kind of meaningless for consumer level SSDs in my opinion. This is only because MLC based SSDs has a minimum of 10000 write cycles per cell. Even thought most SSDs have 1.5 to 2 million MTBF, I doubt a heavy duty write cycle would allow the MLC NAND to last 4 years. However, when you download small files from WCG you are only writing on certain cells and it is a very small % of the overall drive size. I wouldn't be worried about BOINC killing an SSD with a good controller.

It's good to hear that the newer SSDs have ratings like 1M MTBF. I brought this up with the older SSDs to hopefully save some folks grief. It is good to know others are seeing similar failure rates on them.

Just think about risk if you're crunching on the old SSDs with an important rig. At least image the rig.

One of the drives I lost was on a dedicated cruncher. So it was no big deal just to start over with my standard farm package, on a spare drive.

The RAID 0 in the main rig was a different story. I have LOTS of custom stuff on that. It would have been a major PITA to have lost it totally. Fortunately, I had the rig setup for op sys and apps on the RAID 0, and all my data files on a separate hard drive RAID 1 array. I had not made any major additions to apps in the past year, so the old image I restored was fine.



I did some further testing on the OCZ drive I lost. I conducted similar testing on the Patriot drive I lost earlier, with similar results. See Picture below.



This was the standard "tie test". The unit is tested for performance against a standard auto body hammer and railroad tie. It should be noted that this drive did remarkably well in the test. It only shattered when the spike end of the hammer was tested.

I'm thinking it might be time for an RMA?

LOL, I would RMA it.

Should I be worried Bob that is the exact same one I have? I can easily conduct tests like that but i would prefer not too. I really hope it does well for me, crosses my fingers, but I am sure glad the tech is getting better along with the prices

I think that the issues (failures) with most SSDs is related to the controller failing or the controller having issues. I have an Agility 60GB which has been solid for me since the day I have purchased it. I think you'll be fine.

[Sparky]

SSDs have always been too expensive for me to justify even in my main PC. I can't imagine using them in a cruncher... What would be the purpose of doing so?

[kjeldoran]

SSDs have always been too expensive for me to justify even in my main PC. I can't imagine using them in a cruncher... What would be the purpose of doing so?

The only reason I could see would be silence, and that is only if the rest of the system is silent. Power savings aren't worth it due to the pay off factor when compared to a standard hard drive. I don't see the improved speed or access times as making a huge points difference.

However, for all the machines that I will make that are for more than crunching, an SSD will be going into the machine. It is like night and day when using a system with a standard HDD.

[123bob]

SSDs have always been too expensive for me to justify even in my main PC. I can't imagine using them in a cruncher... What would be the purpose of doing so?

I only ran the vertex in a pure cruncher rig because A) it was small in capacity, and B) it was free to me. I won it in a contest. At the time I built the 860 rig, had it available, so in it went....

It is for sure that it was dead. Tested on several rigs. I even tried the same trick that works for RAM on occasion. I threw it in the freezer overnight. Still gone.

Okay folks, I did a write-up on the subject here: http://www.xtremesystems.org/forums/...d.php?t=260883
I once did something similar right before the XS server crash, but all the info was lost.. I was pretty pissed about that, but figured I could do it again. This time even more detail

I'd pay attention to this man's write-ups. I rely on it for most things. That's why I posted up. There was very good stuff before the "great server crash" a while ago (and after that too....). It's still a shame that was totally lost. A tip for any of this is to write up important posts in word docs, then post them. That way, you always have them in one place, or another.

On the other hand, the server crash caused us to set up the XS refugee camp at Tech Power Up, meet some great folks, and in my case, caused me to set up another team on another forum. Not all bad.

Regards,
Bob

[sierra_bound]

I have been running Intel X25-V 40GB drives on my crunching rigs since their launch with no issues.

To maximise longevity I have implemented the following:

1: Only used 30Gb for the partitions to greatly increase spare area.

2: Made sure they are all correctly write block aligned.

3: Disabled the page file.

4: Increased Boinc write to disk to 600 from default.

I am using XP x64 so no TRIM available but the drives are fine without it for this usage profile.

I made a huge thread ages ago about the Jmicron drives and have to agree they were total garbage

I have seen reports and had returns of quite a few Ilindrix and Sandforce drives dying very quickly but have seen very few (one total if my memory is currently working ok ) of the same reports of Intel G2 drives falling over so if I was buying again I would still go for the little Intels...

Rather than disabling the page file completely, I would recommend moving it to a second drive (if there is one) or second partition. I do this with my main PC which crunches and performs day-to-day tasks.

I've been running a Sandforce SSD for awhile and haven't had any major problems (knock on wood).

[Movieman]

All I can say about SSD's is they look nice for a OS drive for a main machine but not for the constant read/writes with a cruncher.
Did I say I just grabbed a 150gig Velociraptor for $45.00 shipped?