Hi! On Mon, 26 Mar 2012 21:31:45 +0300, Maksym Planeta <mcsim.plan...@gmail.com> wrote: > I'm currently working on getting tmpfs work on the Hurd. I already asked > on IRC to make or suggest some test for tmpfs. Last time Thomas Schwinge > suggested me to compile packages and thanks to that a found some bugs. > Now they are seems to be fixed, [...]
Great! > I also made some performance tests. First was measuring time of working > command "apt-get source elinks". Approximately half time of working of > this command takes extracting files from archive. Second test was > measuring time for command "dpkg-buildpackage -b -nc". I've tested > tmpfs, ext2fs that mounted on ramdisk and usual ext2fs. Results are in > following table: Did you reboot the machine after every test? <http://www.bddebian.com:8888/~hurd-web/open_issues/performance/degradation/> > apt-get dpkg-buildpackage > ramfs+ext2fs 22s 50m > ext2fs 32s 46m > tmpfs 16s 47m > > So, although tmpfs has advantage in speed, it doesn't give much gain in > task, like compiling packages. OK, the apt-get task shows how the results get better if a) the disk I/O layer is removed (ext2fs -> ramfs+ext2fs), and b) one indirection (and RPC) layer is removed (ramfs+ext2fs -> tmpfs). As for the dpkg-buildpackage: my understanding is that once the data is cached in RAM, there's not much difference anymore. No idea off-hand why ext2fs is the fastest, though. > 1. > http://git.savannah.gnu.org/cgit/hurd/hurd.git/log/?h=mplaneta/tmpfs/master Someone to review the patches... :-/ Grüße, Thomas
pgpxjqAm2f5yA.pgp
Description: PGP signature
