LTSP - NFS server not responding

[AdamK]

Hello!

Currently have a problem with my LTSP server when using a gigabit switch with a megabit thin client.
I am using K12LTSP with Fedora 9.

The thin clients get most of the way through booting up then I get error messages saying:
nfs: server 172.31.6.1 not responding, still trying.

I don't get this problem if I use a 100 meg switch though.

I have found several forum posts on the internet which all seem to point to the following webpage with a solution:
http://wiki.ltsp.org/twiki/bin/view/...not_responding

However, when I make the recommended changes It doesnt make a difference.

Below is an example of the default file located at /var/lib/tftpboot/ltsp/i386/pxelinux.cfg with its default settings:

prompt 0
 label linux
 kernel vmlinuz.ltsp
 append rw initrd=initrd.ltsp quiet selinux=0 rhgb

Below are the changes to append that I have tested

append rw initrd=initrd.ltsp selinux=0 rhgb MOPTS=nolock,ro,wsize=2048,rsize=2048

append rw initrd=initrd.ltsp selinux=0 rhgb NFSOPTS="-o etx"

append rw initrd=initrd.ltsp selinux=0 rhgb MOPTS=nolock,ro,wsize=2048,rsize=2048 NFSOPTS="-o etx"

Does anyone have any ideas of where im going wrong with this please?

Adam

[sej7278]

nope i get those errors with gigabit, not tried it over 100base-t, but the fixes didn't work for me when i found that page either.

i think its a redhat bug as i seem to get the same on fedora 7/9/10 and centos 5, one of these days i'll try an ubuntu or freebsd live cd.

[sej7278]

Just noticed that with my Fedora7 box as the server, my Fedora10 box can *read* files back without issue at 30MB/s so it appears to be *writes* that are the issue........

[AdamK]

hi,

Thanx for the replies.

Looks like im going to have to resort to using 100mb switches for our LTSP then, either that or replace the thin clients.

[sej7278]

i'm getting desperate now! its definitely looking like nfs is completely broken on redhat.

i've just tried nic bonding on the fedora7 box (2x gigabit) and iperf reports the speed increasing from about 580-680 to 915-940mbps (112MB/s) so not twice the speed, but not bad.

anyway, i ftp'd a 2.7gb iso image in under 70secs - in fact disk speed was limiting in some cases - i definitely saw a few seconds difference between and old sata1/200gb/8mb drive, a newer sata1/250gb/16mb drive my new uber-fast sata2/1tb/32mb spinpoint-f1 drive.

i copied the same file with nfs between the two fastest drive and got loads of "nfs: server not responding, still trying" messages and it took just under 8mins.

YES YOU READ THAT RIGHT - ftp in 1min, nfs in 8mins.

now ok, ftp has the lowest overhead of all protocols, but that nfs time is ridiculous - even scp with encryption only took 3mins!

EDIT: i've moved to NFSv4 and i'm not getting the "not responding" errors anymore, and and the transfer time is down to 3m30s, still a lot slower than ftp, but about the same as scp; its a lot more setting up though. now things are more stable, i might look at re-enabling autofs.

on macosx, which is still using NFSv3 (although it is said that v4 isn't backwards compatible, it seems you can still mount the fsid=0 "root" export but not the ones below it) the transfer takes 1m30, so much faster than fedora still.

[stevea]

Strange problems you are having here.

I'm getting better NFS client read performance with lesser hardware. So something is wrong w/ your configs. Please don't start badmouthing packages when you haven't diagnosed your own problems.

The "%: server %s not responding, still trying\n" message comes from the kernel implementation of the rpc library when the client call times out. This means your server is not configured correctly or your network is losing packets or the server is dead-slow. Most likely one of the necessary nfs server services is misconfigured or not running.

If you spent 2 minutes firing up wireshark you'd immediately see this problem in big red lines.

Is portmap (rpcbind) running & configured on the server ? Some systems will guess the port, but portmap is the standard. Do the rpc version match (I have no clue what LTSP uses). The Linux server implementation has changes d a lot in the past couple years so I can't give you specific advise except you should see he [nfsd] kernel processes and several user space support processes where misconfiguration can cause your troubles.

-S