ANNOUNCEMENT: Squashfs released (a highly compressed filesystem)
Quoted 9 lines · Matthew Wilcox
Quoted 3 lines · Larry McVoy
Quoted 2 lines
A r/w compressed filesystem would be darned useful too :) mmap(2) is, err, hard. Not impossible, it means the file system has >to
support both compressed and uncompressed files, but it's interesting.
Easier than you think, perhaps. Depends how much compression you're after, of course, but here's how Acorn did it in RISCiX (a 4.3BSD derivative):
Strange you should mention this... I used to work at Acorn with the guy who did this (Mark Taunton). RISCiX was too early to be 4.3, I think it was 4.1, but may have been 4.2, certainly no later, this was 1989.
Quoted 2 lines
Pages were 32k (an interesting feature of the MMU...), and the >underlying filesystem was a fairly vanilla BSD FFS (probably 4k blocks
with 1k fragments; discs were around 50MB). Each page was written at a >32k boundary, but compressed. So there were holes in the file where >other files could store their data. Naturally you waste on average 512 >bytes per 32k page, but I think they managed to get 80MB of unix distro >onto a 50MB disc this way, so it's nothing to be sneezed at.
They only needed to compress the fs on the R140 (an ARM 2 based machine). The better R260 had a bigger disk, and so didn't need the compression. Interestingly, because of the slow disk I/O, compression made the filesystem access faster.
I had an R140, and space was so tight in the filesystem, you had to delete part of the distribution before you could even compile a program!
Phillip