Tsumufs Performance.

[GoogleCodeExporter]

Hello everyone,

I'have analysed Tsumufs performance on disconnected mode. It seems that
fuse slows down tsumufs application. For example, if I copy a file with a
size of 17Mo then the return of time command is this:
real    0m18.541s
user    0m0.003s
sys 0m0.070s

In fact, when fuse write on files, it works with blocks of only 4096 bytes,
hence we have many context switches.

I'have used your decorator from metrics.py and I got those results:
write: [4125, 13.473435640335083] 

There is a fuse option called direct_io which optimizes number of write.
here are the results with this option:

real    0m15.568s
user    0m0.009s
sys 0m0.044s

write: [2063, 7.5717041492462158] 

Debug fuse option displays that fuse works now with blocks of 8192 bytes.
I 'm actually working to extend this value. If you have any idea I would be
interseting.

See you, Bastien

Original issue reported on code.google.com by megab...@gmail.com on 6 Jul 2009 at 4:24

[GoogleCodeExporter]

Hi, I have found that we can use a fuseblk type mount which permit to change 
blocksize.
I wonder if you know where the fuse mounting is done? 

tsumufs on /tmp/tsumufs-test-nfs-dir type fuse.tsumufs 
(rw,nosuid,nodev,user=megabast)

I don't find it in fuse.py even in fuse library. To change block size we need 
to have
something like this

tsumufs on /tmp/tsumufs-test-nfs-dir type fuseblk.tsumufs
(rw,nosuid,nodev,user=megabast, blocksize=4096)

regards, Bastien

Original comment by megab...@gmail.com on 15 Jul 2009 at 2:21