I spent way too much time over the last few days chasing bugs which turned out to have nothing to do with Linux.
I bought a SATA controller which arrived just before the weekend. It seems there is a fundamental flaw with the Silicon Image 3114 chips. Or to be more precise, with the firmware on some of the boards using this chip.

This thread is a summary of all manner of problems with it, but in short, it corrupts data past a certain block number. This took a lot of tracking down. (And badblocks takes forever to run when in destructive mode).

There is mention in that thread that a firmware update fixes the problem. Unfortunately, the DOS based flasher program seems completely unable to even write to my card.

I guess I’ll only use this controller for smaller disks, unless someone comes up with a workaround.

I just got two 64GB SSDs. In particular the gskill 25S2S’s. Initial impressions: Yes, faster than a hard disk, but not *that* fast.

The blurb claims sequential reads of up to 155MB/sec. Maybe the ICH7 I plugged it into sucks for some reason, but I could only hit about half that. Running them both in RAID0 didn’t give me a significant increase either. I’ll do some more conclusive experimenting on newer SATA chipsets when I go into the office next week.

One thing that I did find really strange though, was the hdperf results.

Interesting how the performance isn’t uniform across the SSD, and also how the patterns shape up as we get into the higher sectors.

With SSD prices continuing to tumble, it’s great that this kind of performance is getting closer to becoming mainstream. Whilst the results above don’t blow me away, it’s still a decent improvement over rotating media.
As a final example: untarring the 2.6.28 kernel to 2x 7200rpm disks in RAID0 = 27 seconds. The same procedure to my raid0′d SSDs = 11 seconds. Not too shabby.