Here are the results with a "dd if=/dev/zero of=/storage/20M bs=1m
count=20" running
$mount
/dev/wd0a on / type ffs (local)
/dev/wd0e on /home type ffs (local, nodev, nosuid)
/dev/wd0d on /usr type ffs (local, nodev)
<nfs_server_address>:/mnt/mfs_volume on /storage type nfs (v3, udp,
timeo=100, retrans=101)
$ vmstat
procs memory page disks traps cpu
r b w avm fre flt re pi po fr sr wd0 fd0 int sys cs us
sy id
0 13 0 298820 114832 1430 0 0 0 0 1 1 0 15 6926 746 2
8 89
$ top
load averages: 5.16, 2.83, 2.24
openbsd.my.domain
11:26:55
67 processes: 2 running, 63 idle, 1 zombie, 1 on processor
CPU states: 0.4% user, 0.0% nice, 3.0% system, 0.4% interrupt, 96.2%
idle
Memory: Real: 220M/374M act/tot Free: 111M Cache: 97M Swap: 73M/256M
(all processes are at 0% cpu and in idle/sleep state)
$ systat
http://dl.dropbox.com/u/1313066/mailing_list_openbsd/systat_openbsd.png
$ nfsstat
http://dl.dropbox.com/u/1313066/mailing_list_openbsd/nfsstat_openbsd.png
I can provide more data but I am not sure what commands to run, if you can
provide the commands I can run them :)
I agree that trying OpenBSD 5.1 is an obvious test but I prefer to keep
aside for now as I have no simple way to test it.
On 29 May 2012 12:04, Tomas Bodzar <[email protected]> wrote:
> On Tue, May 29, 2012 at 11:36 AM, Schmurfy <[email protected]> wrote:
> > Hi,
> > I have a moosefs cluster (http://www.moosefs.org/) running on linux
> servers
> > and one more linux server mounting the distributed filesystem and
> exporting
> > it using NFS (since OpenBSD does not have FUSE support).
> > Now here comes my problem, if I mount the nfs share on a linux host
> > everything is fine and I can read/write at full speed (100Mb network) but
> > if I mount the share on an OpenBSD host the write speed is capped around
> > 1Mb/s (!) but the read is around 80Mb/s.
> >
> > I found some old issues on this mailing lists (2004/2006) with users
> having
> > these kinds of problems which openbsd nfs client but I am not sure they
> are
> > related to mine, if I exports a nfs share on the linux gateway which is
> > local disk (not on the moosefs cluster) then the speed is fine on OpenBSD
> > as well as linux clients so it seems related indirectly to moosefs but
> the
> > linux client can still read/write at full speed.
> >
> >
> > I tested this with two OpenBSD 5.0 clients, one is a physical machine the
> > other is a VM running under kvm, both are using the em network driver,
> the
> > machines are all in the same local network directly connected by a 100Mb
> > switch, I tried many flags to mount the nfs share on OpenBSD but never
> > managed to get any improvement over a simple mount with default options.
> > I was using pf so I tried disabling it but it changes nothing, another
> > really strange thing for me is that if I copy the file using scp instead
> of
> > nfs from the OpenBSD client then the write speed is also 80Mb+
> >
> > On my linux gateway here is the /etc/exports file (I am using the
> > nfs-kernel-server package under unbuntu 11.10) :
> >
> > /mnt/mfs_volume *(rw,no_subtree_check,fsid=10,no_root_squash)
> > /tmp/data *(rw,no_subtree_check,fsid=5,no_root_squash)
> >
> >
> > Does anyone have an idea on what could happen here ?
> > is there an alternative to nfs I could try ?
>
> Obvious like trying 5.1 and/or current. Some vmstat, top,
> systat,nfsstat outputs will be fine as well to see your paging, speed
> of disk, interrupts, load, rpcstats and so on
> >
> > Thanks for any help this is driving me crazy...
