[H-GEN] Flash drives and wear levelling: user experiences?

[Joel Michael]

Brendon Higgins wrote:
> I was confused until I figured out you were actually talking about writing a 
> stream of data in fractional chunks of flash cell sizes. Very interesting.
> 
I've even still got the spreadsheet I used with all the raw data in it, 
if anyone is interested.  I also have performance numbers for a 4GB USB 
flash stick, a 512MB USB flash stick, and a 400GB SATA hard drive.

Interestingly enough, a normal SATA hard drive exhibits similar symptoms 
with many continuous small writes, but for a different reason.  I 
haven't tested real-world performance after optimising xfs on a single 
disk, though, so I can't say if it helps in the real world.

However, I can assure you that this tuning has made a world of 
difference to my MythTV system, which has its / partition on the 16GB CF 
card on XFS (and recordings on a couple of SATA disks).  Before the 
tuning, selecting the recorded program list may or may not have resulted 
in a timeout (and a "the backend has gone away, has it crashed?" 
message), now it's instant.