SSD Write Endurance 25nm Vs 34nm

[Computurd]

well i think that alot of manufacturers have been bending the rules a bit with the SF issue. From what i have read many are allowing it.

[Christopher]

well i think that alot of manufacturers have been bending the rules a bit with the SF issue. From what i have read many are allowing it.

If there was some kind of assurance that one or two out of every ten SF22xx drives were subject to the error, trying to get a new one wouldn't be a bad idea -- but if data like that exists, no one is talking. I have a few weeks left to make a decision like that, and I know there are many, many SF2281's out there happily working with 1155 boards without incident. There just isn't much real information about it since everyone with inside knowledge is bound by iron clad NDA, but if it's some SF incompatibility with my hardware then drive swapping won't matter. I'm going to roll with it for a while longer before making any decisions, but if it keeps up like this I'll invoke the nuclear option.

EDIT

It finally crashed again, so I tried it in my AMD system. I couldn't even get the system to boot with the drive in the machine, even as a secondary. I don't know what the hell's going on, so I put it in my laptop.

[Christopher]

I just read AnandTech's review on the Intel 710 HET MLC. On the second page I read this:

Thankfully we don't need to just take Intel's word, we can measure ourselves. For the past couple of years Intel has included a couple of counters in the SMART data of its SSDs. SMART attribute E2h gives you an accurate count of how much wear your current workload is putting on the drive's NAND. To measure all you need to do is reset the workload timer (E4h) and run your workload on the drive for at least 60 minutes. Afterwards, take the raw value in E2h, divide by 1024 and you get the percentage of wear your workload put on the drive's NAND. I used smartmontools to reset E4h before running a 60 minute loop of our SQL benchmarks on the drive, simulating about a day of our stats DB workload.

Once the workloads finished looping I measured 0.0145% wear on the drive for a day of our stats DB workload. That works out to be 5.3% of wear per year or around 18.9 years before the NAND is done for. I'd be able to find more storage in my pocket before the 710 died due to NAND wear running our stats DB.

For comparison I ran the same test on an Intel SSD 320 and ended up with a much shorter 4.6 year lifespan. Our stats DB does much more than just these two tasks however - chances are we'd see failure much sooner than 4.6 years on the 320. An even heavier workload would quickly favor the 710's MLC-HET NAND.

I looked at the Smart data for my Intel drives and it looks like it works. I don't know how to reset the attribute with Smartmontools, but I'm working on it. This could be a useful tool for Anvil and OneHertz.

http://www.anandtech.com/show/4902/i...200gb-review/3

[Computurd]

Nice find! Very interesting

[Christopher]

Nice find! Very interesting

I've never used smartmontools, but somehow you can reset certain smart attributes with it. I'm trying to figure that part out now, or whether just disconnecting the drive resets the counter. I did some cross checking and I think its in the data sheet for one of the series of drives (it works for all of them)

[Ao1]

^ Intel talked about this at IDF 2010. The web seminar can be found here. (How do you measure endurance)

[Anvil]

Kingston SSDNow 40GB (X25-V)

358.91TB Host writes
Reallocated sectors : 9
MD5 OK

36.45MiB/s on avg (~12 hours)

--

I've moved the Corsair to an ASUS M4E-Z68 and it can't keep up the same pace as the ASRock MB.

It'll have to do for this weekend though, the Marvell controller on the ASRock was terrible and resulted in an avg of <70MiB/s.
I managed getting it stable though by downloading a new driver and just as important, changing the SATA cable.
(the one I started off with was bad, never used it before though)

Corsair Force 3 120GB

01 90/50 (Raw read error rate)
05 2 (Retired Block count)
B1 47 (Wear range delta)
E6 100 (Life curve status)
E7 80 (SSD Life left)
E9 85962 (Raw writes)
F1 114552 (Host writes)

94.81MiB/s on avg (~11 hours), so, down more than 10MiB/s on avg.

power on hours : 328

[Anvil]

I just read AnandTech's review on the Intel 710 HET MLC. On the second page I read this:

I looked at the Smart data for my Intel drives and it looks like it works. I don't know how to reset the attribute with Smartmontools, but I'm working on it. This could be a useful tool for Anvil and OneHertz.

http://www.anandtech.com/show/4902/i...200gb-review/3

E2h = 5616 (based on >12 hours workload)

5616 / 1024 = 5.484%

It writes about 130GiB per hour

I'll have a look at one that's running 24/7 as a boot drive on a server.

(I did not reset the counters)

[Anvil]

The Force 3 disconnected 30minutes ago. (in the middle of a loop)
In 12 hours it wrote 4026 GiB, avg MiB/s 94.64.

So, no improvement on the other Z68 rig, will be moving it to a X58.

[bulanula]

Well Christopher, at least we now firmly know that SF2 has not been fixed. 10% failure due to incompatibility alone sounds ridiculously high to me. Lesson to learn : avoid SF drives until a major vendor jumps onto it. Avoid for now unless you like to RMA all day long back and forth ( one guy on OCZ forums RMA his drive 3 times and it still failed LOL ).

[Anvil]

The Force 3 is now up and running on my super-rig
(X58, 3Gb/s ICH10R, iaStor 9.6.0.1014)

Speed looks to be on par or better than the M4E-Z 68 rig but let's give it a few loops...

[sergiu]

Well Christopher, at least we now firmly know that SF2 has not been fixed. 10% failure due to incompatibility alone sounds ridiculously high to me. Lesson to learn : avoid SF drives until a major vendor jumps onto it. Avoid for now unless you like to RMA all day long back and forth ( one guy on OCZ forums RMA his drive 3 times and it still failed LOL ).

Avoid is too harsh to say considering that for each drive with issues there are 9 drive that are working. A better advice would be "Evaluate carefully your options! If you cannot afford to risk, then consider SSDs with other controllers". I know Sandforce based SSDs deployed in desktops that are used as servers (24/24 usage) and in 3 months there was only one restart having root cause an unstable overclocking during which one drive was temporary not recognized.

[B.A.T]

Update with a milestone
m4:
574.7837 TiB
2117 hours
Avg speed 88.77 MiB/s.
AD gone from 27 to 22.
P/E 10044.
MD5 OK.
Still no reallocated sectors


Kingston V+100
92.3074 TiB
367 hours
Avg speed 76.74 MiB/s.
AD gone from 114 to 105.
P/E ?.
MD5 OK.

[bluestang]

Update with a milestone
m4:
574.7837 TiB
2117 hours
Avg speed 88.77 MiB/s.
AD gone from 27 to 22.
P/E 10044.
MD5 OK.
Still no reallocated sectors

Bravo!

[Christopher]

My Mushkin has run without incident on my C2D laptop (I think it's the ICH8M). Speeds drop from 123MBs Avg on my 1155 system to 94MBs Avg. I'm using the RST drivers as well.

The Sata II ports on the 1155 rig were good for just as much speed as the Sata III ports, so maybe the much slower laptop helps stability. I couldn't even get my AMD system to boot with the drive attached to a controller... which is strange (it's the 710 SB).

I certainly don't think it's every SF2281 drive, but I think some hardware just doesn't work with the drives due to defect, or of out of spec hardware, or one of twenty other reasons. Until and unless this drive doesn't work with any hardware I have, I won't be giving up on it. So if it's not stable in this laptop, then I get a PCIe Marvel 6G controller. If that doesn't work, then I start thinking about more extreme options.

[johnw]

I know Sandforce based SSDs deployed in desktops that are used as servers (24/24 usage) and in 3 months there was only one restart having root cause an unstable overclocking during which one drive was temporary not recognized.

A server was being run with overclocking, and not even known stable overclocking at that? I am not surprised that someone who would be crazy enough to run a server like that was also crazy enough to put Sandforce SSDs in it...

[Christopher]

I heard that it if you put a SATA bus analyzer in the chain to try and analyze SF stability issues, you end up stopping the instability you're trying to detect. Maybe I can find one cheap on ebay

EDIT
Forget that... they're seriously expensive... who would have thought?

PS
Here is the link to the Intel 320 PDF for enterprise use if you want to see what Intel has to say on the matter of workload timers.
http://www.intel.com/content/dam/doc...n-addendum.pdf

[sergiu]

A server was being run with overclocking, and not even known stable overclocking at that? I am not surprised that someone who would be crazy enough to run a server like that was also crazy enough to put Sandforce SSDs in it...

Well, the overclocked server + Sandforce SSD was my idea, but I did not implemented the solution by myself, as it is over 1000Km away. I could only "crash" the server remotely, nothing more . Either way, the guys are very happy with the choice. They did not complain about problems and they noticed that swapping takes place very fast. Having an i7 2600K @4.8Ghz, 16GB ram and a Vertex 3 240GB hosting VMs (in house usage) for half the price that they would have paid for a real server with slow HDDs is a good deal for them.

[Computurd]

im not sure how much stock i would put into the analyzer causing the issue to stop. If they cant even explain how it is happening, or make it replicable, how would they be sure the analyzer is stopping it?

So much FUD coming out around this issue! Hard to sift out the truth.

if the problem cant be located via a bus analyzer, then it will be impossible to fix.

[Christopher]

im not sure how much stock i would put into the analyzer causing the issue to stop. If they cant even explain how it is happening, or make it replicable, how would they be sure the analyzer is stopping it?

So much FUD coming out around this issue! Hard to sift out the truth.

if the problem cant be located via a bus analyzer, then it will be impossible to fix.

coincidence?

I was joking about that

Seriously, that's what SandForce said to AnandTech a while back. I don't believe anyone who claims to know what's going on with any sort of specificity. Every other controller seems to work okay (with maybe the Kingston 100+'s being some kind of exception) with every 1155 board and X58 and AMD southbridge. I'm going out on a limb here, but I think maybe the problem is with SandForce

Here's some more on the Intel workload timers

Example for Intel X18-M/X25-M G2 SSDs only: The subcommand 0x40 ('-t vendor,0x40') clears the timed workload related SMART attributes (226, 227, 228).

From the Smartctl MAN page
http://smartmontools.sourceforge.net...martctl.8.html

[Anvil]

Kingston SSDNow 40GB (X25-V)

360.19TB Host writes
Reallocated sectors : 10 (up 1 again)
MD5 OK

34.42MiB/s on avg (~24 hours)

--

Corsair Force 3 120GB

01 90/50 (Raw read error rate)
05 2 (Retired Block count)
B1 47 (Wear range delta)
E6 100 (Life curve status)
E7 79 (SSD Life left)
E9 88562 (Raw writes)
F1 118022 (Host writes)

94.70MiB/s on avg (~9 hours)

power on hours : 339

Let's see how long it lasts on the X58.

--

I'm happy with my SF based drives, including the SF2 based ones. (and I've got a few )

[Christopher]

Mushkin Chronos Deluxe 60 Update

05 2 <<-----This has increased from 0
Retired Block Count

Also Program Fail and Erase Fail (B5 and B6) increased to 1, likely increasing retired block count.

B1 25 (Up from 20)
Wear Range Delta

F1 92732
Host Writes

E9 71505
NAND Writes

E6 100
Life Curve

E7 64
Life Left

Average 93.23MBs.

239Hours Work
Time 9 days, 23 hours

Last 19.7hrs on 3gps port (C2D Laptop)

12GiB Minimum Free Space

SSDlife expects 16 days to 0 MWI


It looks like it will run forever in the laptop, but it's much (soul-crushingly) slower that the P67's 3gbps ports.
We'll see, but if(when) it crashes I'm trying something else.

EDIT: I wasn't even paying attention. At some point I had two reallocations with one program fail and one erase fail. I'll keep an eye on it.

** Apparently, I'm not S.M.A.R.T. enough to reset the workload timer on my Intels with Smartctl... **

[B.A.T]

Update
m4:
581.5452 TiB
2138 hours
Avg speed 88.74 MiB/s.
AD gone from 22 to 18.
P/E 10156.
MD5 OK.
Still no reallocated sectors


Kingston V+100
98.1974 TiB
389 hours
Avg speed 76.91 MiB/s.
AD gone from 105 to 98.
P/E ?.
MD5 OK.

[devsk]

Anybody wants to bet the m4 goes all the way to 1PB?

[Ao1]

** Apparently, I'm not S.M.A.R.T. enough to reset the workload timer on my Intels with Smartctl... **

Did you try this command to reset the values?

smartctl -t vendor,0x40 /dev/pd0

(pd0 is the drive indentify and is user variable. In my case pd0 = C drive on SATA Port 0)

If the command works you should see this:

C:\Program Files (x86)\smartmontools\bin>smartctl -t vendor,0x40 /dev/pd0
smartctl 5.41 2011-06-09 r3365 [i686-w64-mingw32-win7(64)-sp1] (sf-win32-5.41-1)

Copyright (C) 2002-11 by Bruce Allen, http://smartmontools.sourceforge.net

=== START OF OFFLINE IMMEDIATE AND SELF-TEST SECTION ===
Sending command: "SMART EXECUTE OFF-LINE IMMEDIATE subcommand 0x40".
Drive command "SMART EXECUTE OFF-LINE IMMEDIATE subcommand 0x40" successful.

After an hour the values should change. I just ran it to make sure it worked. I only ran it for 80mins and the amount of writes were so low they did not even register. (This is on an X25-M btw)

(E2) 226 – 0 (0/1024 = 0.000%).
(E3) 227 – 79 (79% Reads)
(E4) 228 – 80 (80 minutes)