nfs and getattr
A NFS client through 100Mb Ether takes a loooooooong time to read a database from the server, which the server itself reads in less than 10 seconds. The database is a scene description for render that basically 'includes' the same small set of files (around 10) many times, instanced all over the place...
Mount in client:
10.0.0.1:/home on /home type nfs (rw,nfsvers=2,noac,addr=10.0.0.1)
Looking at nfsstats in the server:
Server rpc stats:
calls badcalls badauth badclnt xdrcall 491811 0 0 0 0
Server nfs v2:
null getattr setattr root lookup readlink
1 0% 429338 87% 0 0% 0 0% 61474 12% 26 0%
read wrcache write create remove rename
963 0% 0 0% 1 0% 1 0% 0 0% 1 0%
link symlink mkdir rmdir readdir fsstat
0 0% 0 0% 3 0% 0 0% 2 0% 0 0%
Server nfs v3:
null getattr setattr lookup access readlink
1 100% 0 0% 0 0% 0 0% 0 0% 0 0%
read write create mkdir symlink mknod
0 0% 0 0% 0 0% 0 0% 0 0% 0 0%
remove rmdir rename link readdir readdirplus
0 0% 0 0% 0 0% 0 0% 0 0% 0 0%
fsstat fsinfo pathconf commit
0 0% 0 0% 0 0% 0 0%
What is that ton of getattr ? Do they come from nfs itself or must be done by the reader via stat()s (perhaps it checks for file presence before opening) ?
Is there any way to speedup that ?
Will nfs-v3 perform better ?
TIA
(Kernel is a patched (only bugfixes and bproc) 2.4.21-pre5. No -aa included.)
Signature · 4 lines
J.A. Magallon <[email protected]> \ Software is like sex: werewolf.able.es \ It's better when it's free Mandrake Linux release 9.1 (Cooker) for i586 Linux 2.4.21-pre5-jam0 (gcc 3.2.2 (Mandrake Linux 9.1 3.2.2-3mdk))
Quoted 9 lines · J.A. Magallon
A NFS client through 100Mb Ether takes a loooooooong time to read a database from the server, which the server itself reads in less than 10 seconds. The database is a scene description for render that basically 'includes' the same small set of files (around 10) many times, instanced all over the place...
Mount in client:
10.0.0.1:/home on /home type nfs (rw,nfsvers=2,noac,addr=10.0.0.1)
More with a silly test:
#!/bin/bash
for i in $(seq 1000)
do
cat $0 > /dev/null
done
annwn:~> time nfst
0.69user 1.49system 0:02.04elapsed 106%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (126416major+42508minor)pagefaults 0swaps annwn:~> time bpsh 0 nfst
1.02user 1.86system 0:33.64elapsed 8%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (280major+30minor)pagefaults 0swaps
Signature · 4 lines
J.A. Magallon <[email protected]> \ Software is like sex: werewolf.able.es \ It's better when it's free Mandrake Linux release 9.1 (Cooker) for i586 Linux 2.4.21-pre5-jam0 (gcc 3.2.2 (Mandrake Linux 9.1 3.2.2-3mdk))
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
" " == J A Magallon <J.A.> writes:
> What is that ton of getattr ? Do they come from nfs itself or > must be done by the reader via stat()s (perhaps it checks for > file presence before opening) ?
See the fine Linux Kernel archives.
They come mainly from open(). Use the "nocto" option if you think you can do without them, but *do not* do so if you think there might be any chance whatsoever of 1 machine having to access a file that another machine has written.
Cheers,
Trond