SSD Write Endurance 25nm Vs 34nm

[bulanula]

Some pictures of that Mtron would be nice

[iceblade]

Christopher is it possible to somehow report these findings to Samsung, so they can fix this issue in a future firmware upgrade?

[Ao1]

Guys, TRIM seemed to be working fine with the normal ASU endurance test and as Anvil highlighted it does not work with the enterprise test as the files are not deleted. With the SF drive TRIM occurred in one hit and once it executed all other instructions were ignored. (i.e. the drive locked up until the TRIM operation had finished). With the 830 the TRIM executions were comparatively ultra-fast and in some cases I noticed small TRIM operations whilst the drive was still writing.

Since I disabled AV the 830 has shown consistent performance. If I recall correctly the performance issues that Christopher experienced only occurred once the 177 decimal attribute exceeded 5,000. (?) At this stage the theoretical P/E count would have expired. I haven’t had a chance to look, but maybe the performance drop is related to reallocated sectors, program error fails etc?

Without knowing the relationship between data retention longevity and P/E cycles past the theoretical limit set by the NAND manufacturers I believe vendors are negligent in letting the SSD continue to write past a certain point. In my view the SMART data should trigger a warning that the drive will become read only once the P/E cycles gets close to a threshold where data retention to meet JDEC standards is jeopardised. After a grace period the controller should prevent further write activity.

That might sound a bit draconian, but the number one task for a storage device has to be data integrity, any thing else is a bonus.

My 2 cents.

[Ao1]

Christopher is it possible to somehow report these findings to Samsung, so they can fix this issue in a future firmware upgrade?

I'm on the case with this and I’m hoping Samsung will provide an explanation on some of the issues that are being observed. The only issue I have seen is the failure of 177 to update properly without a power cycle. Hopefully Samsung can give an opinion on why Christopher saw performance problems past 5,000 on atribute 177. A lot of confusion is related to the poor documentation on the SMART data attributes, which really lets the drives down.

[sergiu]

Without knowing the relationship between data retention longevity and P/E cycles past the theoretical limit set by the NAND manufacturers I believe vendors are negligent in letting the SSD continue to write past a certain point. In my view the SMART data should trigger a warning that the drive will become read only once the P/E cycles gets close to a threshold where data retention to meet JDEC standards is jeopardised. After a grace period the controller should prevent further write activity.

That might sound a bit draconian, but the number one task for a storage device has to be data integrity, any thing else is a bonus.

My 2 cents.

A better strategy might be limiting the write speed to a "safe" value that would still allow usage but extend the lifetime with a few more months. If drive becomes suddenly readonly, I am sure it will conflict with Windows write requests and it might not even be readable. Seems Kingston is already doing that, but it remains to be seen how "safe" is their speed.

[Anvil]

Maybe LTT isn't such a bad idea afterall

For LTT to be sensible it would have to kick in much later than it currently does

[minpayne]

@Ao1, how did you figure out that the TRIM is working fine? Could you elaborate please? (It's easy to understand why TRIM doesn't work for full span or RAW writes as the OS cannot send the TRIM command to mark which LBAs are deleted, but how do you verify that TRIM is working when running the non-enterprise endurance testing in ASU?)

[Christopher]

I don't think the problems I've experienced have anything to do with throttling. Why would Samsung throttle reads? Why would it throttle writes in the way that it has?

Unless there is something going on, such as lots of NAND errors, and the drive is slowed down as it tries to sort everything out.

[Ao1]

@Ao1, how did you figure out that the TRIM is working fine? Could you elaborate please? (It's easy to understand why TRIM doesn't work for full span or RAW writes as the OS cannot send the TRIM command to mark which LBAs are deleted, but how do you verify that TRIM is working when running the non-enterprise endurance testing in ASU?)

http://www.hyperio.com/hIOmon/AddOns...GadgetHelp.htm

[Ao1]

I don't think the problems I've experienced have anything to do with throttling. Why would Samsung throttle reads? Why would it throttle writes in the way that it has?

Unless there is something going on, such as lots of NAND errors, and the drive is slowed down as it tries to sort everything out.

Everything was OK up until 5K P/E? I'd say it was a bug but it would need to be replicated to confirm. I'm still intending to stop at 3K P/E for the retention test, but I'm keeping a close eye on performance. It’s kind of weird if it didn't slow down until 5K, plus from what you have reported the chipset driver/ static data ratio and idle time have all made a difference suggesting there are at least variable factors involved.


EDIT: Btw…. your f/w version is CXM01B1Q? Is there an earlier version? Is it possible to “downgrade” to an earlier version?

[Christopher]

Ao1, to my knowledge, there is no other FW available. Perhaps early PM830 drives for Dell and Lenovo had other FW, but not the consumer 830. I think I could replicate the issue with another identical drive, but if you're right about throttling, then it will happen to all drives.

Why would reads get throttled at 5000PE cycles? I think after the update today, I'll do some performance benches.

(Also, the drive isn't really using all that many fewer PE cycles per day -- surely true throttling would do something more... invasive to prevent rapid PE cycle consumption)

[Christopher]

Some pictures of that Mtron would be nice

Any preference? Your wish is my command.

[Ao1]

Ao1, to my knowledge, there is no other FW available. Perhaps early PM830 drives for Dell and Lenovo had other FW, but not the consumer 830. I think I could replicate the issue with another identical drive, but if you're right about throttling, then it will happen to all drives.

Why would reads get throttled at 5000PE cycles? I think after the update today, I'll do some performance benches.

(Also, the drive isn't really using all that many fewer PE cycles per day -- surely true throttling would do something more... invasive to prevent rapid PE cycle consumption)

I don’t think it is anything to do with throttling. Perhaps it is something to do with the huge jump that occurred in 177, which is clearly a bug. One minute the value was 129 and the next it was 5,444. That probably set off a few alarm bells. Who knows what internal housekeeping that might have triggered. Without extensive testing it might be hard to replicate and find the cause.The best bet is for Samsung to comment on the 177 update bug and give an opinion on the cause of the slow down.

[Christopher]

See, I'm not sure that it works like that. The SMART value didn't increase, but I think the drive "knows" what 177 really is. I thought along the same lines as well, but the performance degradation happened before the power cycle (I was trying various methods to get the performance back, and in the process power cycled the drive).

[Anvil]

Kingston SSDNow 40GB (X25-V)

618.83TB Host writes
Reallocated sectors : 05 21
Available Reserved Space : E8 99
POH 5591
MD5 OK

33.28MiB/s on avg (~142 hours)

--

Corsair Force 3 120GB

01 94/50 (Raw read error rate)
05 2 (Retired Block count)
B1 69 (Wear range delta)
E6 100 (Life curve status)
E7 10 (SSD Life left)
E9 671239 (Raw writes) ->656TiB
F1 893360 (Host writes) ->872TiB

MD5 OK

106.81MiB/s on avg (~190 hours)

power on hours : 2564

WRD has started decreasing.

--

edit

I re-booted the machine and filled the free space on the Force 3 with a "46%" and then deleted the file. (as a sort of cleaning )
Both drives are back running. (main cause was to update the SMART logging utility)

[Christopher]

Samsung 830 64GB Update, Day 35
FW:CXM01B1Q

GiB written:
216885.84

Avg MB/s
66.61

PE Cycles
13669, up from 13321 yesterday

Reallocated Sectors
40960
20 Blocks, up from 18

845 Hours




I took a few quick benches after power cycling the drive for the update, and also running the Samsung Performance Optimization in Magician 3.1



CDM is deceptive, as out of 5 runs it only displays the fastest result, so the best result is displayed. Generally, out of the 5 runs per category, 1 is decent, 3 are awful, and 1 is abysmal.

[Anvil]

CDM is deceptive, as out of 5 runs it only displays the fastest result...

Normally that's not an issue, the issue in this case is the drive that's not consistent.
(CDM could have used the average but that would not be representative in this case either)

[Christopher]

So I'm looking for suggestions for a new drive to test. I have acquired several drives recently that could be considered good test candidates to one degree or another, but I thought I'd put the question out there.

Any suggestions?

[B.A.T]

Todays update.
Kingston V+100

This one has dropped out again.

Intel X25-M G1 80GB
224,4992 TiB
20157 hours
Reallocated sectors : 00
MWI=140 to 149
MD5 =OK
49.15 MiB/s on avg


m4
212.3534 TiB
778 hours
Avg speed 80.54 MiB/s.
AD gone from 236 to 232.
P/E 3736.
MD5 OK.
Reallocated sectors : 00

[minpayne]

So I'm looking for suggestions for a new drive to test. I have acquired several drives recently that could be considered good test candidates to one degree or another, but I thought I'd put the question out there.

Any suggestions?

Which options have you got? I'm interested in these:

Intel X25-E (old skool )

Plextor M2P (Marvell + 32nm Toshiba Toggle)

[Ao1]

Average WA

60GiB (LBA) * P/E Cycles (690) = 41,400
Actual writes = 2,854
WA = 14.50

MWI 81
POH 89

Avg write speed = 4.51MiB/s

EDIT: I'll do a pure 4K only WA calculation later.

[Christopher]

minipayne,

I do have a another X25-E 32 on the way (on its way via UPS), but it would take... Years. As in at least a few - to test. The 40GB Intel drives have well exceeded their PE cycle ratings. Who is to say that an X25-E wouldn't last 2x the PE rating? And, I believe the 32GB actually has 40GB of SLC as well, meaning its overprovisioned a fair amount out of the box. So, I think testing an E would be a time commitment I'm just not willing to make.

I am trying to get a Plextor though. The first drive I tested was a SF2281 with Toshiba toggle. The Samsung is Toggle NAND, and the a 64GB Plextor would be interesting (Marvell + Toggle) as a counterpoint to the other two drives.

[minpayne]

minipayne,

I do have a another X25-E 32 on the way (on its way via UPS), but it would take... Years. As in at least a few - to test. The 40GB Intel drives have well exceeded their PE cycle ratings. Who is to say that an X25-E wouldn't last 2x the PE rating? And, I believe the 32GB actually has 40GB of SLC as well, meaning its overprovisioned a fair amount out of the box. So, I think testing an E would be a time commitment I'm just not willing to make.

I am trying to get a Plextor though. The first drive I tested was a SF2281 with Toshiba toggle. The Samsung is Toggle NAND, and the a 64GB Plextor would be interesting (Marvell + Toggle) as a counterpoint to the other two drives.

I can't believe that everyone is putting the X25-E as a legendary product Actually my X25-E 64GB is now at 99 (normalized value) MWI with around 10 TiB host writes, 0 reallocation. I guess it's pretty safe to leave it on a server and not having to worry about any swapping between physical memory and virtual memory, if lots of users always keep submitting memory-hungry jobs to make the server run out-of-memory

The Plextor M2P may only come in 128GB and 256GB (but I'm not sure on this). It would be expensive to test on it if there's no 64GB model...

[minpayne]

Christopher:

It seems you've got one more thing to play with the Samsung 830 There seems to be a new firmware released for it. Someone just mentioned that he updated his firmware version from CXM01B1Q to CXM02B1Q. Would this resolve the 177 updating issue and the performance drop?

[johnw]

I can't believe that everyone is putting the X25-E as a legendary product Actually my X25-E 64GB is now at 99 (normalized value) MWI with around 10 TiB host writes, 0 reallocation. I guess it's pretty safe to leave it on a server and not having to worry about any swapping between physical memory and virtual memory, if lots of users always keep submitting memory-hungry jobs to make the server run out-of-memory

The Plextor M2P may only come in 128GB and 256GB (but I'm not sure on this). It would be expensive to test on it if there's no 64GB model...

There is no 64GB model. Ditto for the Corsair Performance Pro (which is nearly identical to the Plextor M2P).

My 64GB X25-E is also still at 99 (see below). I was messing around with it the other day and I noticed that my locking SATA cable does not lock on the X25-E. But I couldn't see whether there was a broken plastic tab on its connector, or if it was just never designed to hold a locking SATA connector. Except for that, the X25-E is great. I've been using it for my boot drive for years now (check the power-on hours!).

ID# ATTRIBUTE_NAME          FLAG     VALUE WORST THRESH TYPE      UPDATED  WHEN_FAILED RAW_VALUE
  3 Spin_Up_Time            0x0000   100   000   000    Old_age   Offline      -       0
  4 Start_Stop_Count        0x0000   100   000   000    Old_age   Offline      -       0
  5 Reallocated_Sector_Ct   0x0002   100   100   000    Old_age   Always       -       0
  9 Power_On_Hours          0x0002   100   100   000    Old_age   Always       -       23874
 12 Power_Cycle_Count       0x0002   100   100   000    Old_age   Always       -       188
192 Unsafe_Shutdown_Count   0x0002   100   100   000    Old_age   Always       -       61
232 Available_Reservd_Space 0x0003   100   100   010    Pre-fail  Always       -       0
233 Media_Wearout_Indicator 0x0002   099   099   000    Old_age   Always       -       0
225 Host_Writes_32MiB       0x0000   200   200   000    Old_age   Offline      -       819652
226 Intel_Internal          0x0002   255   000   000    Old_age   Always       -       4294967295
227 Intel_Internal          0x0002   000   000   000    Old_age   Always       -       281474976710655
228 Intel_Internal          0x0002   000   000   000    Old_age   Always       -       4294967295