SSD Write Endurance 25nm Vs 34nm

[Christopher]

Does your 177 update properly, Ao1?

I took a look at the ASU pic you posted, and it seems... slow. Especially for the beginning of the loop.

[Ao1]

It’s too early to say if 177 is updating properly. So far ~470 GiB and no movement on the raw value. If it hasn’t changed by 2,000 GiB I’ll do a power cycle.

Avg write speed is ~64

[Christopher]

Yeah, you should be pulling down 110MB/s avg... that's why I say it's looking slow... it's looking like my 830 now.

The 830 didn't get down into the 64MBs range until after the performance problems. It was doing about 9000GB a day, averaging around 110MBs.

[Anvil]

Ao1

Did you disable AV scanning on the test folder?

Also, try running smartmontools, it woke up some of my attributes on the Kingston (X25-V), if that doesn't help a bit of idling may trigger it?
(worth a try as we need to figure out more on the Samsung)

[Ao1]

I’ve just disabled MS Essential on the drive and write speeds have improved. At the end of the first loop avg speed was ~83MiB/s. At the end of the second loop I was up to ~ 86 MiB/s. Nice

I also took the opportunity to reboot and there was no change in the 177 values. I’ll try a power cycle soon if it doesn’t budge sometime soon.


EDIT: ~ 700GiB. Power cycle. No change in 177. Avg write ~100 MiB/s just in from the start of the first loop.

[Christopher]

Wow, Ao1... you really should be getting closer to 160MBs at the beginning of the loop and stay above 100MBs for the whole loop.

Just to be sure, you turned the system off (or unplugged the drive) and there was no change in 177?

I promise I'm not crazy.

[Anvil]

It will most likely take 4-5TiB of writes for 177 to change, not sure if the raw value will change until value is depleted (1)

[Christopher]

I had movement in 177 for the first 10 days. The value increased from 0 to 134 over the first 10 days without a power cycle, and it went from 100 to 96 over that period as well. After the first power cycle, 177 jumped from 96/0/124 to 1/0/5444 I think.

Now I'm really intrigued. I'm thinking about getting another 830 as well. Or another Chronos Deluxe... which would go much more smoothly with 3.3.2FW.

[Christopher]

I had movement in 177 for the first 10 days. The value increased from 0 to 134 over the first 10 days without a power cycle, and it went from 100 to 96 over that period as well. After the first power cycle, 177 jumped from 96/0/124 to 1/0/5444 I think.

Now I'm really intrigued. I'm thinking about getting another 830 as well. Or another Chronos Deluxe... which would go much more smoothly with 3.3.2FW.

[Ao1]

I left the drive to idle for a while to see if 177 would change. It didn’t, but write speeds improved significantly when I restarted the ASU.

URL=http://imageshack.us/photo/my-images/16/withidle.png/][/URL]

Here is a shot at the same point in the second loop. Seems idle time enables the drive to clean itself up. I’ll check with hIOmon later to see if/ when the TRIM commands are being executed.



Christopher, have you tried letting your drive idle for couple of hours?

[Christopher]

Ao1,

I did let the 830 idle for about 8 hours. But if a SE won't help, then I don't think idling will as well.

See, the drive did start out fast, but running with the amount of static data + the min free space caused the drive to get slow after 48 hours. I changed it, and it was able to keep the speed up for the next 8 days -- until I messed with min free space again. This time, nothing would help. Not secure erasing, or full formats, or idle time. But I could try letting it idle again for a few hours. I'll let it idle for the next 5 hours, at which time I will post an update.

[Anvil]

Kingston SSDNow 40GB (X25-V)

610.82TB Host writes
Reallocated sectors : 05 21
Available Reserved Space : E8 99
POH 5521
MD5 OK

33.62MiB/s on avg (~72 hours)

--

Corsair Force 3 120GB

01 94/50 (Raw read error rate)
05 2 (Retired Block count)
B1 64 (Wear range delta)
E6 100 (Life curve status)
E7 10 (SSD Life left)
E9 651239 (Raw writes) ->636TiB
F1 866775 (Host writes) ->846TiB

MD5 OK

106.77MiB/s on avg (~120 hours)

power on hours : 2494

--

[Ao1]

OK, we have some movement (without a reboot or a power cycle).
64GiB * P/E Cycles (1) = 64
Actual writes = 106
WA = 0.60



EDIT:

As Anvil has pointed out in a PM, it looks like the LBA count was not updating correctly either. The jump occured after a re-boot and coincided with the increase in write speeds. Very strange.

[Ao1]

And after a power cycle:

64GiB * P/E Cycles (40) = 2,560
Actual writes = 684
WA = 3.74

[Ao1]

And let the punishment begin. The 177 decimal value has increased by 1 since starting with the 4K, so it seems that it does not update properly. (At least not when the drive is being subjected to a continuous load).

[Anvil]

I'll get you an updated version sometime this weekend with more options for write duration.

Lets hope it survives the testing

[B.A.T]

Todays update. Nice to have 24h without problems
Kingston V+100
325.6350 TiB
1778 hours
Avg speed 25.05 MiB/s
AD still 1.
168= 1 (SATA PHY Error Count)
P/E?
MD5 OK.
Reallocated sectors : 00


Intel X25-M G1 80GB
212,6894 TiB
20087 hours
Reallocated sectors : 00
MWI=143 to 142
MD5 =OK
50.69 MiB/s on avg


m4
192.9329 TiB
707 hours
Avg speed 80.03 MiB/s.
AD gone from 248 to 243.
P/E 3392.
MD5 OK.
Reallocated sectors : 00

[Christopher]

While I wait for the 830 to finish it's idle session, I'm reflecting on the performance of the 40GB Intels. In a way, the fact that they're continuing with no sign of slowing is pretty damn impressive. The 320 should hit 700TB before too long, which is just plain awesome for a 40GB drive. I still think there is something to the slow and steady approach.

[Ao1]

So far the 830 is coping quite well with the 4K random work load. 177 does not update properly unless you power cycle, but at least the LBA count seems to be working OK now. Everything else seems to update as well, so it’s just 177.





64GiB * P/E Cycles (115) = 7,360
Actual writes = 868
WA = 8.47 (Ouch)

[johnw]

64GiB * P/E Cycles (115) = 7,360
Actual writes = 868
WA = 8.47 (Ouch)

I am surprised it is not higher since you are doing nothing but full-span 4KiB random writes.

By the way, did you verify that 512 * LBA count is equal to the bytes written according to ASU?

[Ao1]

^ Yes. In post # 3189 the total LBA count was 1433682636. (Just before starting the 4K workload). In post #3194 the LBA count was 1819470188. The difference = 385787552 or 184 GiB.

Anvils app is showing 126,454.69 MiB plus 59.51 GiB for the test file = 183 GiB.


EDIT:

64GiB * P/E Cycles (189) = 12,096
Actual writes = 1,056
WA = 11.45
MWI (?) 94

[Christopher]

My 830's Sector written count is about 900GB more than ASU, but most of that is writes from other sources over the past month. I believe it is pretty accurate.

[johnw]

64GiB * P/E Cycles (189) = 12,096
Actual writes = 1,056
WA = 11.45
MWI (?) 94

Aha, WA going up. I wonder what it will stabilize at. It would be interesting to try this:

1) power-cycle the SSD and then note the raw values of attributes 177 and 241, call them r177a and r241a

2) run ASU endurance writes for a while, a day or two, note the ASU bytes written at the beginning and end of the period

3) power-cycle the SSD and check attributes 177 and 241 again, call it r177b and r241b

4) verify that the ASU bytes written = 512 x (r241b - r241a)

5) compute the WA = 64 x 1024^3 x (r177b - r177a) / ( 512 x ( r241b - r241a ) )

By the way, what QD does ASU Enterprise endurance benchmark use for its full-span 4KiB random writes? I suppose QD does not matter much for the 4KiB random write speed on the Samsung 830, but for some SSDs (like the m4 or Performance Pro), the write speed increases quite a bit with higher QD.

[Ao1]

I am surprised it is not higher since you are doing nothing but full-span 4KiB random writes.

If I count only the 4K writes the WA is currently coming out at 29.47 (What I reported before included all write activity)

From the graph below the temp of the SSD has jumped by around 20 oC once I started with the 4K workload. The red lines coincide exactly with power off cycles.



EDIT, the SSD is noticeably warmer to touch. I’d say that the temp readings are quite accurate.

[Ao1]

By the way, what QD does ASU Enterprise endurance benchmark use for its full-span 4KiB random writes? I suppose QD does not matter much for the 4KiB random write speed on the Samsung 830, but for some SSDs (like the m4 or Performance Pro), the write speed increases quite a bit with higher QD.

According to the resource monitor the QD is 1.