SandForce Using Intel 25nm MLC Flash On SSDs

[zoob]

If Intel decides to add compression to their G3 (or G4) SSD what will you think then?

Innovation from smaller companies drives the larger companies back to competition. Intel got comfortable with the performance of the X25-M first gen, and look what happened to the G2 - they had a firmware update increase the artificially capped write speeds.

[Ejizz]

Snip...

Don't be so brand sensitive, I'm a happy G2 owner, but it's obvious sandforce 'currently' has a far superior controller than Intel.

I hope G3 get's a serious boost because upgrading from a G1 to G2 was a waste of time, I wil certainly think twice before listening to Anands Intel hype again.

[Levish]

^ not sure what you expected, latency was the same, read / write speeds were nearly the same. If something isn't at least 20% faster in the things you are most affected by (IMO) its not worth upgrading to.

Even if a drive like the C300 is faster at reading and writing large sequential files, where it matters is at the 4-512kb file size and latency/IOPs unless you regularly work with large files in which case you'd likely be better served by a conventional disk :p

The C300 does happen to be significantly faster at the 4k file sizes than the intel drives though and I would like to swap out my 2x vertex 120's for a pair and retire these to another role

[zoob]

http://www.anandtech.com/show/2899/13

The Sandforce (Vertex 2) results don't look "UGLY" to me.

I don't like the sandforce compression crap. It's smart, but it kills performance if you need to read/write incompressible data (IE compressed RARs). All current benches on them are misleading because 99% of them are done with highly compressible data (IE OS files or synthetic bench data).

Anandtech has an article up a while ago where they used an incompressible data file for IOmeter and the results were UGLY. The lack of a buffer really shows.

[Fuji]

so since we're talking about sandforce compression.

I read that the sandforce controllers are able to achieve a write amplification of as high as 0.5.

Does this have to do with the compression?

[alfaunits]

The SF results are artificial, since IOMeter data is compressible and thus IOMeter gives higher results.
That does not mean it's bad with non-compressible data, just not impressive like C300.

http://www.anandtech.com/show/2899/13

The Sandforce (Vertex 2) results don't look "UGLY" to me.

[danielkza]

The SF results are artificial, since IOMeter data is compressible and thus IOMeter gives higher results.
That does not mean it's bad with non-compressible data, just not impressive like C300.

http://www.anandtech.com/show/3681/o...200-reviewed/4

[Ao1]

Something I don’t get with the Sandforce drives. They compress data so that less data is written to the drive. Although less data is written the OS records the uncompressed data size. So if a drive is showing up as full on the OS in reality it isn’t. So why could this unseen capacity not be used for over provisioning to cut down on the amount of nand being used?

[Silentvois]

Perhaps that's their way of cutting down on wear.

[alfaunits]

http://www.anandtech.com/show/3681/o...200-reviewed/4

So?

[zoob]

So?

Did you read the article?

In our Agility 2 review I did what you all asked: used a newer build of Iometer to not only write data in a random pattern, but write data comprised of truly random bits in an effort to defeat SandForce’s data deduplication/compression algorithms.

[zads]

Something I don’t get with the Sandforce drives. They compress data so that less data is written to the drive. Although less data is written the OS records the uncompressed data size. So if a drive is showing up as full on the OS in reality it isn’t. So why could this unseen capacity not be used for over provisioning to cut down on the amount of nand being used?

Okay, what happens if you write 100% incompressible data to the drive?
SSDs require overprovisioning to function properly and efficiently...
Since Sandforce often has "extra overprovisioning" due to the data compression,
this simply allows it to function that much more efficiently.

so since we're talking about sandforce compression.
I read that the sandforce controllers are able to achieve a write amplification of as high as 0.5.
Does this have to do with the compression?

Well, the write amplification on all other drives is >1.0,
the Sandforce gets as LOW as 0.5..
that means some data written of 1GB uncompressed size is being compressed to (on average) 0.5GB.
But Sandforce says they are seeing an average WA of 0.7 in most user cases.

[Ao1]

Okay, what happens if you write 100% incompressible data to the drive? SSDs require overprovisioning to function properly and efficiently...
Since Sandforce often has "extra overprovisioning" due to the data compression, this simply allows it to function that much more efficiently.

But the chance of that happening would be remote and if the drive was full you would only be reading from it anyway. Let’s say your drive is 80% full. How much of that 80% is actually written data? 40%? 60%?

Did you read the article?

You forgot to include the sentence that followed that quote.
“What we saw was a dramatic reduction in performance”.

“………..The takeaway? While SandForce’s controllers aren’t immune to performance degradation over time, we’re still talking about speeds over 100MB/s even in the worst case scenario and with TRIM the drive bounces back immediately.”

Still good performance even in a worst case scenario, I guess the issue is that mileage can vary.

[zoob]

You forgot to include the sentence that followed that quote.
“What we saw was a dramatic reduction in performance”.

“………..The takeaway? While SandForce’s controllers aren’t immune to performance degradation over time, we’re still talking about speeds over 100MB/s even in the worst case scenario and with TRIM the drive bounces back immediately.”

Still good performance even in a worst case scenario, I guess the issue is that mileage can vary.

The bolded text is exactly what I'm trying to get at - Even in the worst case scenario it can still outperform the Intel SSD in 4K Random writes. Even with an Agility 2 on random data on 4K random, it can outperform an Intel X25-M G2.

[Ao1]

No doubt; compressed writes, both sequential and random are faster. Compressed reads are slower going by AS SSD Benchmarks; ~ 210 vs 250. In some instances the Sandforce is faster and in others it’s slower. Either way no difference to the majority of users. Both drives are fast.

[Silentvois]

Regardless, they'll definitely serve their purpose as an OS drive. Longevity of these SandForce drives is what I'd worry about.

[alfaunits]

So I said... it's not bad, just not as impressive as C300.
It would be a hazard if it could not outperform an intentionally gimped >1yr old X25-M

The bolded text is exactly what I'm trying to get at - Even in the worst case scenario it can still outperform the Intel SSD in 4K Random writes. Even with an Agility 2 on random data on 4K random, it can outperform an Intel X25-M G2.