SF-2281 question regarding how it calculates storage space

[Hulk]

I realize that SSD's which use the Sandforce 2281 controller compress data to increase transfer speeds. My question is when data is written to the drive does the the drive indicate a reduction in capacity based on the compressed or uncompressed size of the data written?

For example. Assume you write 500MB of data to the drive, which the drive compresses to 300MB. Will explorer or whatever disk management tool you use report that the capacity of the drive has reduced by 500MB or 300MB?

I assume it is 500MB and the drive still has that 200MB of space free. But I'm not sure. This would make sense as to why these controllers don't perform well when they are filled with in-compressible data because when they are filled they are really filled! But if they are "filled" with compressible data then they really aren't filled and there is still room for the drive to operate efficiently.

And now my second question is why do other controllers, ones which do not compress data handle the full scenario after a TRIM pass so much better than the Sandforce controller?

Not sure about all this and now that SSD's are here I'd like to get my facts straight.

 

[hhhd1]

The compression is totally transparent form the OS or the user, you won't see it in action except when you analyze SMART data.

When you write to SSD 500mb, the controller compress it to 300mb and write that and leave the rest unwritten, which could lead to write amplification of less than 1 in some cases, this would result in less wear out to the cells, and also, faster writing data to cells since writing 300 would be done in shorter time that writing 500mb.

 

[greenhawk]

by reporting the drive has used 500MB, it removes any issues caused by compressing of data.

Compressing data on a drive is not new, but it always had drawbacks in some way. Personally it is something best avoided.

As to the TRIM question, no idea. Maybe the chipset is just crap at deal with it regardless of why.

 

[johny12]

I realize that SSD's which use the Sandforce 2281 controller compress data to increase transfer speeds. My question is when data is written to the drive does the the drive indicate a reduction in capacity based on the compressed or uncompressed size of the data written?

For example. Assume you write 500MB of data to the drive, which the drive compresses to 300MB. Will explorer or whatever disk management tool you use report that the capacity of the drive has reduced by 500MB or 300MB?

I assume it is 500MB and the drive still has that 200MB of space free. But I'm not sure. This would make sense as to why these controllers don't perform well when they are filled with in-compressible data because when they are filled they are really filled! But if they are "filled" with compressible data then they really aren't filled and there is still room for the drive to operate efficiently.

And now my second question is why do other controllers, ones which do not compress data handle the full scenario after a TRIM pass so much better than the Sandforce controller?

Not sure about all this and now that SSD's are here I'd like to get my facts straight.

The data compression from Sandforce improves the speed because there is less to write. It also takes less physical space on the flash, but the OS does not know about that. You get a larger over provisioning space which also further improves performance, endurance and lowers write amplification. Check out this article http://thessdreview.com/daily-news/...resentation-on-over-provisioning-at-fms-2012/

 

[Hulk]

Thanks for the link. Very informative. I didn't know that over-provisioning has so much impact on random write and hardly any on sequential writes. Of course that's a big deal since most disk activity is small random writes.

 

[razel]

My question is when data is written to the drive does the the drive indicate a reduction in capacity based on the compressed or uncompressed size of the data written?

Reports uncompressed to the file system.

And now my second question is why do other controllers, ones which do not compress data handle the full scenario after a TRIM pass so much better than the Sandforce controller?

It's been fixed now. But I assume the issues before the fix was simply because writing was more complicated. However, honestly I think Sandforce just didn't have the smarts to fix it in time. It took Intel a year to deliver a product based on it. In fact, Intel still hasn't released a fixed TRIM firmware yet either. But let's not rush them. I'm pretty danned sure, it's in QA. Take your time Intel, you guys know storage that isn't a toy.