SandForce compression, and software FDE. A bad combo?

[VirtualLarry]

Just wondering.

I mean, I know that SF achieves its superlative write amplification by using compression, and with software FDE, the sectors that are written to the drive are basically uncompressable, so that's a worst-case scenario for the drive.

What I'm curious about, is if it causes the data written to the drive to actually expand, larger than the source data? I know that this is true with many compression algorithms, they optimize for the main case of compressability, with the sacrifice that if the data is incompressable, it actually gets larger when output from the compression algorithm.

If so, doesn't that mean that I'm encroaching the reserved spare area with user data, that proved to be incompressable/expand larger than the user data sectors? Will this cause problems with SF's algorithms?

Anyone have any first-hand experience with this?

I did notice that drive speed was noticeably slower in CDM, after the FDE was finished.

 

[frostedflakes]

The engine is pretty complex I'm sure, I think Anand has mentioned that it uses numerous compression/dedupe techniques depending on the workload, and presumably no compression for workloads that wouldn't benefit from it. It's supposed to be able to analyze data on the fly and decide, OK this data will be compressed well with LZ77, use that algorithm, this chunk of data will work well with another algorithm, use that, this chunk of data doesn't benefit from any compression algorithms and will actually come out larger, so just store it straight to the NAND without any compression or dedupe, etc. That's my understanding.

 

[VirtualLarry]

I guess, in some ways, that it doesn't matter how the drive reacts to the effects of FDE.

The problem, if you can call it that, with performance and software FDE, is that the software FDE limits the performance more than whatever drive you happen to be using is, it seems like.

Edit: Some CDM Benchmarks:

Fresh: 196MB/sec random read, 146MB/sec random write
After FDE: 27MB/sec random read, 75MB/sec random write
Week later: 32MB/sec random read, 75MB/sec random write

Software FDE really does slow down performance quite substantially. And this is on a hex-core 3.51Ghz PC.

(I suppose an Intel chip, with AES-NI would speed things up a bit, but still, there is a slowdown.)