SSD Write Endurance 25nm Vs 34nm

[bulanula]

It seems like the M4 and the 320 are close to cracking the 200TB barrier. Wonder how the C300 will perform.

[Anvil]

The 320 should be > 200TB, not sure of how you could have misread the TB for the m4? (it's currently at ~84TiB)

--

154.72TB Host writes
MWI 15
Reallocated sectors still at 6

MD5, no errors.

It looks like speed is a bit up for this session, 33.8MB/s on avg. (~35 hours)

[devsk]

@B.A.T. : it seems like your avg write speed is going up! That's a good feature to have: more you write on me, faster I become...

[B.A.T]

It's only going up because the test was haltet for a couple of hours. From my next update I'll be using the latest v of anvils app and there the avg speed is calculated for me. At the moment the avg speed is 88.38 MiB/s

[Vapor]

C300 Update from yesterday when XS was down:

36.89TiB, 61.8MiB/sec since switching to the latest Anvil App, 88MWI, 620 P/E cycles (SMART raw wear).



Going to change the "normalized writes" charts to Anvil's suggestion (just a change of units and an addition of a new "Writes So Far" chart).

[Vapor]

Updated charts


Host Writes So Far




(bars with a border = testing stopped/completed)


Raw data graphs

Writes vs. Wear:


MWI Exhaustion:


Writes vs. NAND Cycles:



Normalized data graphs
The SSDs are not all the same size, these charts normalize for total NAND capacity.

Writes vs. Wear:


MWI Exhaustion:



Write-days data graphs
Not all SSDs write at the same speed, these charts factor out write speeds and look at endurance as a function of time.

Writes vs. Wear:


MWI Exhaustion:

[bulanula]

The 320 should be > 200TB, not sure of how you could have misread the TB for the m4? (it's currently at ~84TiB)

Well it is currently at about 50 MWI and has done almost 100TB so I expect it to be around 200TB ( or even more ) when it reaches MWI 0 point. Maybe that is not a good assumption ???

[Vapor]

Well it is currently at about 50 MWI and has done almost 100TB so I expect it to be around 200TB ( or even more ) when it reaches MWI 0 point. Maybe that is not a good assumption ???

You want this chart:

[Anvil]

Well it is currently at about 50 MWI and has done almost 100TB so I expect it to be around 200TB ( or even more ) when it reaches MWI 0 point. Maybe that is not a good assumption ???

I didn't get what you were referring to (you didn't mention MWI), I expect all drives to go past 200TiB.

Going to change the "normalized writes" charts to Anvil's suggestion (just a change of units and an addition of a new "Writes So Far" chart).

It's easier to relate to imho

[B.A.T]

Evening update.
I updated to anvils latest ver. and have started using md5 check.

296 hours, 87,1098 TiB, Wear Leveling Count and Percentage of the rated lifetime used has gone from 53 to 50.
Avg speed reported from anvils app is 88.38 MiB/s
MD5, no errors.

[johnw]

I didn't get what you were referring to (you didn't mention MWI), I expect all drives to go past 200TiB

I'll be impressed if the Samsung 470 makes it past 200TiB. I'm far from certain, but my best guess is that the write amplification of the Samsung is around 5 -- much higher than the other SSDs in the test. If the Samsung makes it to 200TiB, I guess that would be about 16,000 erase cycles on the flash.

[johnw]

118.719 TiB, 332 hours, sa177: 1/1/9774

127.012 TiB, 355 hours, sa177: 1/1/10535

Average speed reported by Anvil's app has been steady at about 112MB/s.

The other two unknown SMART attributes, 178 and 235, are still at 72/72/276 and 99/99/2, just as they were when the SSD was fresh out of the box.

[Anvil]

We'll just have to wait, I'm already impressed by the Samsung.

The Samsung is a bit of a dark horse as we don't know much about the drive at all.

[Vapor]

C300 evening update, plus a new chart to add to the mix.

41.787TiB, 86MWI, 702 P/E cycles (raw wear indicator), all MD5 okay, 81.7MiB/sec




New chart!

Approximate Write Amplification



Using the normalized writes (total GiB written / total NAND capacity), raw wear indicators (Samsung and Crucials), and raw wear inferences for Intel ( (100-MWI) x 50 / normalized writes ), I came up with a chart for approximate write amplification during testing. Not sure how useful it is, ultimately, but it seems interesting

Only value I'm unsure of is the X25-V, it may have 10k cycle NAND, in which case it's WA would be 2x what is shown as I calculated with the assumption of 5k cycle NAND.

[johnw]

Only value I'm unsure of is the X25-V, it may have 10k cycle NAND, in which case it's WA would be 2x what is shown as I calculated with the assumption of 5k cycle NAND.

Also, the Samsung is a guess, since we are not certain that SMART attribute 177 is actually erase cycles on the flash. I think it probably is, but if it gets to 20,000 and the SSD is still going fine, I will start to have more doubts....

[bulanula]

Very interesting data, especially the Samsung as they are probably using different methods of production compared to onfi people like Intel micron but numbers like 16 000 seem pretty insane!

[Eggcake]

The V should have 5k cycle NAND (34nm).

[B.A.T]

Morning update:
310 hours, 91.2361 TiB, Wear Leveling Count and Percentage of the rated lifetime used has gone from 50 to 48.
Avg speed reported from anvils app is 88.37 MiB/s
MD5, no errors.

[Anvil]

Only value I'm unsure of is the X25-V, it may have 10k cycle NAND, in which case it's WA would be 2x what is shown as I calculated with the assumption of 5k cycle NAND.

I've got to find some smaller tools, looks to be torx, really small stuff. (I expect it is 5K)

--

157.63TB Host writes
MWI 14
Reallocated sector count : 6

MD5 -> OK

[stevecs]

Is anyone testing endurance of the intel 520 series? Or do we expect that to match one of th profiles we're currently testing due to same nand tech being used or similar? Trying to find relatively in-expensive drives for a write cache for the array which I would expect >200TiB of writes per drive in say 1 year. I know the pliant ones can do this but they're 10x+ the cost of everything else.

[Anvil]

@stevcs

I expect you are thinking of the Intel 510 Series SSD (the 520 Series is due Q4?)

There are currently no-one testing the 510 series, being 34nm it should do well vs the m4, you should also consider the 320 Series 80-120GB as 25nm looks to be great!

[Ao1]

The Marvell 9174 Controller is used by the Crucial M4, Intel 510 & Corsair P3, however both Intel & Micron produce their own unique firmware. There may also be other significant differences. NAND configuration, DRAM, Spare Area etc.

Intel do however specify workload parameters for Client and Enterprise applications, so those specs could be used to determine suitability for a particular workload. As can be seen in this thread those workload estimates appear to be quite conservative.

A small increase in reserve area would be enough to significantly increase work load capability, if the workload was a problem.

[stevecs]

@anvil- you're correct, sorry (always thinking a generation ahead) but I meant the 510/elmcrest not the 520/cherryville. Problem with the 3xx is that they're all 3Gbps, not a problem in itself (as I don't except a single drive to do more than 100MB/s) but I'm looking to put them all into external chassis so want to make sure that all are running at 6gbps for the expander(s). (I've run into problems before mixing/matching, it's better to have everything the same signaling).

@Ao1, yes, that's what I thought as well. I was actually planning on 28% or so over-provisioning for whatever SSD. Just trying to pick the right ones for the workload. Since this is a 'write cache' drive (it's main purpose is to cache all random writes to the back-end datastore which is ~200TB) with about 100GB/day I don't want to have the replacement issues at work (don't know the actual brand, but EMC&Oracle use SSD's for their Tier 0 in the sans, for heavy database functions they don't last a year. Not a big deal as the client's are paying for the speed so the thousands $$/drive is not an issue. A 'little' different for a home system however.

[deathman20]

While it can't be shown in this test here is a test someone ran on an Intel X25-V measured durability

http://translate.google.com/translat...ogle.com&twu=1

[Ao1]

Don't know if this will help but as a guide the Intel 320 is spec'd for around 15TB with a 4K 100% random workload over the full span. (TB varies depending on drive capacity). Reducing the span and adding over provisioning will increase write capacity significantly, but 200TB of random 4K writes is going to be a tall order for any non enterprise SSD any way you try to cut it.