SSD S.M.A.R.T attributes

[Ao1]

This tread is to get a better understanding of the S.M.A.R.T attributes of SSD’s. It is not a thread about how long before my ssd craps out, or why I’m not bothered when my ssd craps out.

First off it doesn’t seem that there is an SSD standard for S.M.A.R.T attributes. Some of the ID’s seem new and others have been borrowed from HDD ID attributes that are of no relevance to ssd’s (like spin up time.)

Some of the ID’s below maybe incorrect and some of the S.M.A.R.T tools don’t always report the ID’s correctly.

There are loads of tools out there that can report S.M.A.R.T

I like Hard Disk Sentinel as it has a number of other handy features like disk performance monitoring.

Indilinx SSD’s

ID Hex Attribute
----------------------
01 (01) Read error rate
09 (09) Power on Hours
0C (12) Device power cycle count
C7 (199) Write Sectors Total Count
CD (205) Max PE Count Spec
CE (206) Min Erase Count
CF (207) Max Erase Count
D0 (208) Erase count average
D1 (209) Remaining drive life in % by Erase count

There is a really useful post that explains the above.
To find out total writes: ID# 199 x 512
To find out the % of write cycles used: ID# 208 x 100/ID# 205 = % of write cycles used.
To estimate remaining hours of use: ID# 9 (100/7.26) – 1 (x ID# 9)
To convert the above to years: / 8760

Intel SSD’s

ID Hex Attribute
----------------------
05 Re-allocated sector count
09 (09) Power on Hours
0C (12) Device power cycle count
CO Unsafe shutdown count
E1 (225) Host writes
E8 (232) Available Reserved Space
E9 (233) Media wear out indictor
B8 End to end error detection count

Info on Intel S.M.A.R.T attributes can be found here. For G2's (with the latest f/w) S.M.A.R.T data can be obtained from the Intel Drive Toolbox.

As far as I know all ID’s are the same between a G1 and a G2 apart from B8, which is G2 only.

Key ID’s
E1 – HostWrites
This attribute reports the total number of sectors written by the host system. The raw value is increased by 1 for every 65,536 sectors written by the host.
E8 - Available Reserved Space
This attribute reports the number of reserve blocks remaining. The attribute value begins at 100 (64h), which indicates that the reserved space is 100 percent available. The threshold value for this attribute is 10 percent availability, which indicates that the drive is close to its end of life. Use the Normalized value for this attribute.
E9 Media Wearout Indicator
This attribute reports the number of cycles the NAND media has experienced.
The normalized value declines linearly from 100 to 1 as the average erase cycle count increases from 0 to the maximum rated cycles.
Once the normalized value reaches 1, the number will not decrease, although it is likely that significant additional wear can be put on the device. Use the Normalized value for this attribute.

[Computurd]

someone should make a nice calculator out of that, so that all you have to do is input your numbers and it figures it out. it would be one hell of a popular app and for math retards like me it would be frickin awesome

[Ao1]

someone should make a nice calculator out of that, so that all you have to do is input your numbers and it figures it out. it would be one hell of a popular app and for math retards like me it would be frickin awesome

Done

This replicates the formulas as posted by azzie on the OCZ forum. Add SMART data to the red cells only and the rest will occur automatically.

If you have an Intel G2 drive there is no need to do this as the information is available via the Toolbox tool. (The Toolbox tool has been withdrawn but an upgraded version should be available before the year end.)

[alfaunits]

Is it possible to get the SMART values from a RAID controller in any way? (other than the few things in the administrator tool)

[Computurd]

@audienceofone--thank you, that is awesome!! they should give this a sticky just for that!

[Ao1]

^ No problem, it’s only a replication of azzie’s work.

How are your Vertex drives standing up to all the benchmarking you have done?

[Computurd]

getting weird readings like it says power on only 32 hours, so it only counts from last flash of firmware? also, would that render all numbers inaccurate?
also my numbers have letters in them as well?? gah i feel retarded!!

[Computurd]

ok guess i had to switch to decimal data fields...doh! first time with drive sentinel i like it! says i have average of 50 years on my drives and like 350,000 hours of use left. i think they may be coming out well because i have eight members of the raid, which means less erase on each, etc. spreading my use around....also i stay away from write testing as a rule!!

[alfaunits]

JFYI for everyone - SpeedFan can read SSD SMART attributes of SSDs in Areca arrays.

[Computurd]

wow that is awesome, i had read that areca had done some compatibility things with speedfan and their controllers, but i dint realize that they were going that in depth. i figured it was for fan control and pcie bus etc...better monitoring of the device etc.
HMMMM so if speedfan can pull SSD SMART from the individual devices whilst they are in a RAID config, why in the hell can areca not give us raid TRIM!! if they can pass smart attrib, why not trim?

[alfaunits]

Is that mathematically possible, though?
If the erase block size is 20KB (i.e. not a divident of the any stripe size), the TRIM command would overlap multiple blocks on multiple contriollers (actually it would trim only a part of one block not entire block sometimes). Right?

[Ao1]

@ Computurd.

Did you clock the huge difference in wear on Marios's drives in raid 0 in this post? (Disk1 17,077,632 vs Disk 2 7,078,019 or around 240% more writes on disk 1)

Marios told me that those drives were both new when he set up the array so the huge difference in writes on each drive occurred due to way the raid 0 array split the writes.

It would be really interesting to see how the writes get distributed when you use as many drives as you do.

I'm kind of puzzled why the writes on each drive vary so much. If one drive is writing a lot more than the other drive how can a raid 0 array scale so well?

[Computurd]

there is a difference in usage. i noticed that as well, however i did have a number of the disks on a ich10r for a period of time. three of them. however, it wasnt long. i did not write down the numbers of erases, however there were variances of disk life left, i saved the results to a file, i will post them when i am at that comp shortly. i had much the same thoughts when i noticed that. curious. however, it was definiterly not such a pronounced difference, his is surprising.

@alfaunits- well that is a good question. one i wish i had, however the main basis of my whining about trim and arrrays is that it has been promised that trim functionality will 'eventually' work with raid and trim passthrough. of course i have no idea about the mechanics of it, and the questions you pose are probably the same problem they are having. areca has said that with the 18xx series they will have improved ssd support and will also enact that support retroactively with older devices via software/firmware updates...truth be told though i havent heard areca directly say trim, i guess i assumed that trim would be 'enhanced ssd support' will be interesting the next few months . ms has said for a long period of time however that trim would be implemented into raid systems.

[Ao1]

^ trim is comming for raid, it's just a question of when.

[alfaunits]

The question is will it comme too late

[Computurd]

well on that one i agree alfa, there is alot of headway being made with the garbage collection features using idillinx onboard ARM processor. nice thing about this built in programming set:
1. OS independent...works with anything
2. does not create performance issues as it is not another command set being issued.
3. headless....no thought required (like TRIM)

OCZ and idillinx are almost done with a new version of this (Firmware 4.5) that will run in real time. it is auto trim basically. the version that i am currently using only works when your array or disk has been idle for a certain amount of time.
The TRIM spec has many drawbacks, such as I/O overhead. there is much talk of this new GC actually being used in preference of trim. The next gen of devices, sandforce and jetstream, are going to rely heavily on these features. it will be interesting if Intel follows suit.

[Computurd]

@ audience of one.. when i originally tested these devices using that calculator i kept track of remaining life/usage. did not think to track the erase count, etc...definitely will the next time i take the array down, i am working on something with it right now, though.
Life remaining on ocz vertex
519948 hours 59.3 years
577726 hours 65.9 years
519948 hours 59.3 years
433281 hours 49.4 years
305830 hours 34.9 years
433281 hours 49.4 years
472675 hours 53.9 years
399948 hours 45.6 years
there is a disparity in usage there. a big one. however it would be hard to quantify as i have used three on a ich10r array on one computer and the other four on another controller. also it would be hard to figure out which controller had what usage patterns. this array has seen the ich10r, highpoint 3520, highpoint 4320 (total piece of ing crap dont ever buy that POS @#$@!!) the areca 1680-ix the lsi 9211 and lsi 9260.....

------------------------sidenote on my total disgust and hate of the highpoint....the 4320 had issues with the OCZ ssd's. i had found five other guys with OCZ arrays and we were all bomabarding there tech support for a fix. the end result was they say 'we would need the drives themselves to make a fix, but it is possible"
SO>> I talked with the OCZ rep Tony oveer at their forums, and he contacted there people and ended up giving them eight vertex SSD's so that a fix could be issued.
guess what? said fix is still not made. so highpoint with a rubber . i will nefver even look at another of their products. too bad cause the 3520 was a awesome controller. *end rant*

[Ao1]

G1 Intel drives have GC. G2 Intel drives have trim and if you don't use system restore the over head is zero as far as I can see. A manual trim with the toolbox takes 1 to 2 seconds without system restore points and with auto trim you don't notice it one bit. Benchmark scores have actually gone up following the trim f/w update and Intel encourage TRIM to reduce wear. Seems that the OCZ trim is achieving the opposite

[Ao1]

@ audience of one.. when i originally tested these devices .....

Thanks Are you sure your "power on" time got reset after a f/w update? If it did that invalidates the calculation....but I did not think it possible to change SMART data with f/w. I know SMART data can be changed but not by end users afaik.

[Computurd]

oh ocz has trim as well. the GC versions of their firmware is for those who use raid or dont have win7. i was not aware of intel having GC.
not sure about the F/W resetting it. as far as i know it cant...

[alfaunits]

I wonder how GC works? I mean it surely doesn't recognize the underlying FS and free blocks... so how does it know which block it CAN erase?? Seriously.. I am puzzled!
The only thing that comes to my mind is that since SSDs don't use the same blocks to write the same LBA always, a rewrite of the same LBA would eventually (via GC) free the previously used block that was used for that LBA.

If this were true, the blokes at all fronts could instead implement bloody MFT instead

[Computurd]

your ssd RUL calculator is still getting good usage audienceofone! thanks!

[Mastakilla]

where are SMART attributes stored on an SSD?
Will a low level format (or HDD Erase or something) reset SMART attributes?

[Computurd]

nope, low level format wont wipe them, certain "destructive" firmware flashes will though.

[Mastakilla]

Thanks for the answer...

The purpose was NOT to reset the SMART attributes btw
I am preparing my SSD for RMA and the only thing "wrong" with it, is a SMART attribute not being 0 anymore.

I havent found any other "test" that shows it is defective and Intel support advised me to RMA it anyway.

So if I reset SMART, there is no "proof" anymore that the SSD is defective...

I'll now try to wipe the SSD then...