SSD Write Endurance 25nm Vs 34nm

[Christopher]

My trip got cancelled, so here's an update:
MWI just hit 10

Mushkin Chronos Deluxe 60 Update, Day 23

05 2
Retired Block Count

B1 12
Wear Range Delta

F1 217697
Host Writes

E9 167875
NAND Writes

E6 100
Life Curve

E7 10
Life Left

Average 129.71MB/s ? Avg
RST drivers, Intel DP67BG P67

533 Hours Work (23 hrs since the last update)
Time 22 days 5 hours

10GiB Minimum Free Space


EDIT
It really seems to be getting faster over time.
Three weeks ago I was pulling down ~122MBs. Now it's averaging 129 - 130MBs

[One_Hertz]

500.71 TB. 1 wear leveling reserve left! 5478 reallocated sectors. Unfortunately I must go to bed now. When I get up, the SSD should be finished.

[Computurd]

thats IF it doesnt just miraculously keep on going...like lazarus that SSD and to think they are only 85 bucks at newegg now!

[Ao1]

My trip got cancelled, so here's an update:
MWI just hit 10

Mushkin Chronos Deluxe 60 Update, Day 23

05 2
Retired Block Count

B1 12
Wear Range Delta

F1 217697
Host Writes

E9 167875
NAND Writes

E6 100
Life Curve

E7 10
Life Left

Weird. So it looks like B1 does not tell the full story. SymbiosVyse’s drive has around the same WA factor. Considering your drive has 5K P/E cycles and is 60GB the host writes to get to MWI 10 does not compare that well against SymbioVyse’s drive. If your drive has 5K P/E cycles the wear levelling factor is 1.79, which compares to 0.81 for SymbioVyse with no static data.

B1 12 (103)
Wear Range Delta

F1 217,697 (195,264) – 89%
Host Writes

E9 167,875 (146,240) - 87%
NAND Writes

Assuming 5K P/E
5,000 = (217,697) * (0.77) * (1.79) / (60)

Assuming 3K P/E
3,000 = (217,697) * (0.77) * (1.07) / (60)

EDIT

Looking at the Intel’s drives they appear to have performed slightly better than spec’d. (Although the 320 uses 25nm I believe Intel NAND spec sheets quote 5K P/E).

X25-V

5,000 = (184,934) * (1.1) * (1.0) / (40)

320
5,000 = (194,560) * (1.1) * (0.95) / (40)

It looks like SF drives are not as efficient as Intel in static data rotation, although that is offset by WA due to compression.

EDIT 2:

SF drives use spare area for wear levelling, so if you calculate SymbiosVyse's drive as 48GB instead of 40GB the wear levelling index comes out as 1, (which makes more sense than 0.81).

[B.A.T]

New update
m4
676.2078 TiB
2442 hours
Avg speed 90.84 MiB/s.
AD gone from 227 to 221.
P/E 11746.
MD5 OK.
Still no reallocated sectors


SSDLife has decided that the m4 won't die so it predict another 8 years of service

Kingston V+100
163.5195. TiB
645 hours
Avg speed 75.70 MiB/s
AD gone from 39 to 33.
P/E?
MD5 OK.
Still no reallocated sectors

[Anvil]

The Kingston is back on the Z68 rig. (since last night)

Kingston SSDNow 40GB (X25-V)

395.14TB Host writes
Reallocated sectors : 11
MD5 OK

36.18MiB/s on avg (~12 hours)

--

Corsair Force 3 120GB

01 90/50 (Raw read error rate)
05 2 (Retired Block count)
B1 54 (Wear range delta)
E6 100 (Life curve status)
E7 55 (SSD Life left)
E9 173779 (Raw writes)
F1 231419 (Host writes)

104.16MiB/s on avg (~110 hours)

power on hours : 686

I guess I'm going to stop the Force 3 as it's now close to 5 days and re-test with C-States Enabled.

Corsair Force 3 SSD_120GB_1GB-20111015-1342.png

[Hopalong X]

SSDLife has decided that the m4 won't die so it predict another 8 years of service

Maybe that is next for the Intel 320? It could happen!

[Anvil]

It looks like SSDlife predicts 8 years lifetime or so on all drives on BAT's rig. (the Kingston as well)

I'm wondering what's next for the Intel 320, hopefully it stays readable.

[One_Hertz]

This 320 does not seem to care that it has no reserve space left. Still going strong...

502.35TB. 5588 reallocated sectors. Reserve space at 1.

[B.A.T]

It looks like SSDlife predicts 8 years lifetime or so on all drives on BAT's rig. (the Kingston as well)

I'm wondering what's next for the Intel 320, hopefully it stays readable.

It's been counting down to 1 before it went over the top and gave me 8 years.

[johnw]

This 320 does not seem to care that it has no reserve space left.

One might say it has no reservations about it.

[Anvil]

It's been counting down to 1 before it went over the top and gave me 8 years.

I know, all they need to do is to look at Wear Leveling Count, it tells the story.

--

Kingston SSDNow 40GB (X25-V)

396.19TB Host writes
Reallocated sectors : 11
MD5 OK

34.57MiB/s on avg (~23 hours)

--

Corsair Force 3 120GB

01 85/50 (Raw read error rate)
05 2 (Retired Block count)
B1 53 (Wear range delta)
E6 100 (Life curve status)
E7 55 (SSD Life left)
E9 176376 (Raw writes)
F1 234877 (Host writes)

104.16MiB/s on avg (~119 hours)

power on hours : 696

[Christopher]

One might say it has no reservations about it.

You could auction the drive on Ebay -- Intel 320 40GB NO RESERVE!


Still, I'm a little confused... if the drive would theoretically become RO at Reserve Space = 0, couldn't you keep over provisioning the drive until nothing is left? Like a popsicle melting?

You could OP by a few GB, then when that gets trashed, you could OP again. Lather, rinse, repeat. Because of the wear on every cell, each time you'd get less and less additional time, but you should still be able to extend the drive's life considerably.


QUESTION?
Why does Crystal Disk Info report an Intel drive's life as being 100% when the MWI is at 99 or 98 (or possible lower)?

[Anvil]

It reports E8 (Available Reserved Space)
E9 is based on the NAND P/E count and as we know the P/E count is worst case.

CDI_2011-10-15-23-36.PNG

It is an interesting question though, extra OP should in theory have a positive effect on endurance and increased random write performance (IOps) on some controllers.

[Christopher]

The extra space in over-provisioning should increase Available Reserved Space as well, right? If the ARS = 1, you should be able to OP by 4GB and get it back to 100...

I've heard that some controllers don't use the extra space in over-provisioning in this manner. With the Intel's, you should get some performance benefit just from not using all the available space for your partition, though I'm unclear on whether the drive makes some sort of distinction between real spare area and unallocated space on the drive.

Since I don't really need much space, every drive I'm using gets OP'd now (even though I surely don't need to, but it's probably not a bad practice). I could OP my 128GB drives down to 40GB and still have plenty of space for most of my system images.

[Christopher]

Mushkin Chronos Deluxe 60 Update, Day 24

05 2
Retired Block Count

B1 13
Wear Range Delta

F1 225336
Host Writes

E9 173772
NAND Writes

E6 100
Life Curve

E7 10
Life Left

Average 126.02MB/s Avg
Intel RST drivers, Asus M4g-z

551 Hours Work (24hrs since the last update)
Time 22 days 23hours

12 GiB Minimum Free Space

[B.A.T]

New update
m4
684.1035 TiB
2467 hours
Avg speed 90.71 MiB/s.
AD gone from 221 to 217.
P/E 11876.
MD5 OK.
Still no reallocated sectors


Kingston V+100
The Kingston dropped out during the last 24 hour so I'll just have to repeat the process of restoring the log, deconnect the drive and restart the test, but it will have to wait until thursday.

[Anvil]

Kingston SSDNow 40GB (X25-V)

397.91TB Host writes
Reallocated sectors : 11
MD5 OK

33.64MiB/s on avg (~37 hours)

--

Corsair Force 3 120GB

01 87/50 (Raw read error rate)
05 2 (Retired Block count)
B1 55 (Wear range delta)
E6 100 (Life curve status)
E7 53 (SSD Life left)
E9 180625 (Raw writes)
F1 240532 (Host writes)

104.16MiB/s on avg (~134 hours)

power on hours : 712

[Ao1]

This 320 does not seem to care that it has no reserve space left. Still going strong...

502.35TB. 5588 reallocated sectors. Reserve space at 1.

Assuming 1.1 WA & 1.1 WL the P/E cycle count works out @ 15,560

15,560 = (514,406) * (1.1) * (1.1) / (40)

[Ao1]

I found this formula for calculating life time.

SSD Life (in seconds) =
(P/E Cycle) x (Capacity)
(Write Speed) x (Duty Cycle) x (Read/ Write Ratio) x (Write Amplification)

There is a field test that Emphase have put together that seems to utilise elements of that formula. It requires DiskMon & Iometer to calculate the workload variables, but it’s quite easy to then plug numbers into the Xcel worksheet that is provided.

I only ran Anvils app for one cycle, which included generating the test file. Using data obtained from DiskMon and Iometer the Xcel file calculated a life time of 0.08 months. I’m not sure why the % random come out so high, perhaps it was because I only ran one cycle. I’d guess that is why the calculated lifetime is so low. I’m not sure how WA & WL are factored, but it should be possible to tweak the formulas based on the results of the testing in this thread. (A bit beyond me though )

[Anvil]

Nice finding Ao1.

ASU vs IOmeter are two different things in this regard.

ASU creates new files at different sizes and can be seen as random but it is not.

IOmeter creates a single "large" testfile and all IO goes into that file. (both random and sequential)

ASU is by far the most realistic scenario for desktop and most server usage (file sharing), if the scenario was set to a database the IOmeter test would be the more realistic.

The Benchmark in ASU is much closer to how IOmeter works.
(as it works on fixed large files)

What happens if you reverse the seq/random distribution.

[Ao1]

The stats come from data that DiskMon collects whilst ASU is running. DiskMon looks at the sectors that data was written to and this determines random vs sequential. Iometer is only used to determine IOPS, which is based on the Avg file size and % random to sequential that DiskMon creates.

Perhaps DiskMon is not reading the data correctly?

[One_Hertz]

Clearly, the reserve space attribute does not mean what it says it means.

[sergiu]

The number of reallocated sectors is still increasing or is now more stable? Maybe it started to recycle bad blocks, like using only pages that can be successfully programmed. Or maybe it started to drop parity data and use the area as reserve.

[Christopher]

The X25-V 40 has 40GB on board, while the 320 has 40GB plus some unknown amount extra, but no one knows how much, right?

I remember the discussions about the 320's error protection scheme, but I don't recall if the actual amount of extra NAND on the 320 was known.