Interested in SSD longevity?

[VirtualRain]

The brain trust over on Xtremesystems.org have been running a variety of SSD's through a sustained write stress test for the last few months, and they just had their first casualty...

The Samsung 470 64GB drive was the first to wear out it's NAND rated for 3000 P/E cycles but that was largely due to it's amazing sequential write speed compared to others in the test and it's high write amplification (estimated at 5).

How much data could the drive take before wearing out?

So that 478.037 TiB comes to 525.607 TB (I prefer TB to TiB). That is equal to 8212 times writing the 64GB capacity of the drive. Which is equivalent to writing the entire 64GB capacity of the drive, every day, for 22 years. Or to writing 287GB every day for 5 years.

Was the data still in tact after the NAND was unable to write?

I was able to verify the MD5 checksum for the ~40GB static data file, and it was fine. Also, ASU never detected any MD5 errors. So it seems the data has been, and continues to be intact on the SSD.

The thread is over 1400 posts long as of this writing, but here are the interesting ones...

The first post (with periodically updated charts)

Start of the Samsung testing

Death of the Samsung

 

[Inconsequential]

Yay for the Intels doing very well

 

[clystron]

Quite interesting, thanks for sharing.

I have a more "conservative" test-case:

A machine with two 80GB X25-Ms in RAID1 running an application that modifies several hundred files up to 4GB (small changes) plus daily rotated logfiles. Its been running 24/7 for ~110 days now and according to SMART the SSDs write ~10GB per day. Both are still reporting 100 for media wearout so I guess the SSDs will outlive the reset of this machine even if the load picks up a bit

 

[Spacedust]

My current Intel SSD server drive working 24/7 in MySQL server since January 2011 with 1500 databases updated daily is having such SMART status:

=== START OF INFORMATION SECTION ===
Model Family:     Intel X18-M/X25-M/X25-V G2 SSDs
Device Model:     INTEL SSDSA2M080G2GC
LU WWN Device Id: 5 001517 95933a437
Firmware Version: 2CV102M3
User Capacity:    80,026,361,856 bytes [80.0 GB]
Sector Size:      512 bytes logical/physical
Device is:        In smartctl database [for details use: -P show]
ATA Version is:   7
ATA Standard is:  ATA/ATAPI-7 T13 1532D revision 1
Local Time is:    Tue Aug 23 16:06:35 2011 CEST
SMART support is: Available - device has SMART capability.
SMART support is: Enabled

Vendor Specific SMART Attributes with Thresholds:
ID# ATTRIBUTE_NAME          FLAG     VALUE WORST THRESH TYPE      UPDATED  WHEN_FAILED RAW_VALUE
  3 Spin_Up_Time            0x0020   100   100   000    Old_age   Offline      -       0
  4 Start_Stop_Count        0x0030   100   100   000    Old_age   Offline      -       0
  5 Reallocated_Sector_Ct   0x0032   100   100   000    Old_age   Always       -       11
  9 Power_On_Hours          0x0032   100   100   000    Old_age   Always       -       4666
 12 Power_Cycle_Count       0x0032   100   100   000    Old_age   Always       -       22
192 Unsafe_Shutdown_Count   0x0032   100   100   000    Old_age   Always       -       11
225 Host_Writes_32MiB       0x0030   100   100   000    Old_age   Offline      -       346798
226 Workld_Media_Wear_Indic 0x0032   100   100   000    Old_age   Always       -       17235884
227 Workld_Host_Reads_Perc  0x0032   100   100   000    Old_age   Always       -       1
228 Workload_Minutes        0x0032   100   100   000    Old_age   Always       -       3463975423
232 Available_Reservd_Space 0x0033   100   100   010    Pre-fail  Always       -       0
233 Media_Wearout_Indicator 0x0032   079   079   000    Old_age   Always       -       0
184 End-to-End_Error        0x0033   100   100   090    Pre-fail  Always       -       0

These drives are damn good

 

[VirtualRain]

I think the early conclusion that can be drawn from the testing is that there is little risk of any real-world desktop user ever wearing out their SSD, no matter what brand.

As an aside, you may have noticed an absence of the latest generation of Sandforce drives in the test... One interesting fact that surfaced is that the latest Sandforce controllers/firmware have a throttling mechanism built in to prevent premature NAND wear-out. It essentially ensures the drive will survive the warranty period by throttling writes if they begin to exceed a predefined curve. While most users would never hit this threshold and trigger the throttling, it made it impossible for the latest Sandforce drives to participate in this accelerated stress test.

 

[Inconsequential]

I think the early conclusion that can be drawn from the testing is that there is little risk of any real-world desktop user ever wearing out their SSD, no matter what brand.

As an aside, you may have noticed an absence of the latest generation of Sandforce drives in the test... One interesting fact that surfaced is that the latest Sandforce controllers/firmware have a throttling mechanism built in to prevent premature NAND wear-out. It essentially ensures the drive will survive the warranty period by throttling writes if they begin to exceed a predefined curve. While most users would never hit this threshold and trigger the throttling, it made it impossible for the latest Sandforce drives to participate in this accelerated stress test.

That is cheating!

Glad I've gone and will be staying, Intel!

 

[derbothaus]

That is cheating!

Glad I've gone and will be staying, Intel!

Really? Thinking ahead and attempting to fix issues before they are a problem is cheating? Usually we call that evolving or in Apples case, innovation. No real world would ever use the SSD's in the manner they are testing in.

 

[Inconsequential]

Really? Thinking ahead and attempting to fix issues before they are a problem is cheating? Usually we call that evolving or in Apples case, innovation. No real world would ever use the SSD's in the manner they are testing in.

Throttling performance to ensure the drives last past the warranty is cheating in my book.

Reason being others don't do it, why should sandforce drives do it?

If I bought an SSD I want it to run at 100MB/sec until it runs out of life.

I don't mind it running out of write cycles, because that is what happens with this tech.

Really makes no difference to most people so why even bother with it in the first place?

 

[derbothaus]

I guess in the long run I would rather have it last and work slightly slower so I can get good backups or prep for replacement. Whether or not you'd notice it before total failure is a mystery though.
You make good points.
I'd personally love it if they tested all these without TRIM as well. You know, to level the playing field. We'll see then who's controllers flake out first. Or last forever in a 75% degraded state.

 

[fat jez]

You don't get a total failure though. The spec says that when a drive stops being able to be written to, it should still be able to be read from, so you shouldn't lose any data.

 

[derbothaus]

You don't get a total failure though. The spec says that when a drive stops being able to be written to, it should still be able to be read from, so you shouldn't lose any data.

Oh, in that case, *** SF-2200. Kidding, no really, thanks for the heads up. In 5 years when I make a new purchase it will come in handy

 

[toxic]

throttling an SSD because it's being used too much would be like automatically shutting down a desktop hard drive after 12 hours use in one day because it wasn't built for server use. they're effectively forcing you to buy enterprise hardware.

 

[xper]

throttling an SSD because it's being used too much would be like automatically shutting down a desktop hard drive after 12 hours use in one day because it wasn't built for server use. they're effectively forcing you to buy enterprise hardware.

1) The throttling will never come into play in normal use, sure, if you write the same amount of data as the size of the SSD each day you would hit the throttling in a couple of month, but seriously, who uses their 128gb SSD to write 128GB of data each day.

2) Even when the SF SSD hit the throttling it still would be much faster than an Intel drive so the argument that "i am glad i use an Intel SSD" is pretty weird

 

[Hellhammer]

Considering that NANDs lose their charge after about 10 years, that should be a lot bigger concern than wearing out the NANDs

As an aside, you may have noticed an absence of the latest generation of Sandforce drives in the test... One interesting fact that surfaced is that the latest Sandforce controllers/firmware have a throttling mechanism built in to prevent premature NAND wear-out. It essentially ensures the drive will survive the warranty period by throttling writes if they begin to exceed a predefined curve. While most users would never hit this threshold and trigger the throttling, it made it impossible for the latest Sandforce drives to participate in this accelerated stress test.

IIRC SandForce SSDs also have realtime compression so 1GB of data won't be 1GB after compression (I don't know how much the compression is but sounds like this is another way to avoid wear out). One die is also used as a backup (i.e. you cannot write to it. In 25nm SSDs, this is ~8GB, 32nm ~4GB)

You don't get a total failure though. The spec says that when a drive stops being able to be written to, it should still be able to be read from, so you shouldn't lose any data.

MLC NANDs keep the data for 12 months.