I have an interesting problem with NFS that someone may be able to help with.

Sunday I came home from a vacation. Whenever I am gone for more than a couple of days I shut down all my systems, including my myth setup (1 combined frontend/backend and 1 frontend). When I returned I switched everything back on.

For some reason now, I constantly get NFS timeouts on my frontend. Watching live tv does not work either. Nothing has changed on my server for almost 5 months when I installed another tuner. My frontend has been up since January with no changes. I have made no changes either place and both boxes have been rebooted a few times since changes were made so there are no settings that I may have forgotten to save. The server works fine and I have no issues with it, it is my frontend box that has the issues.

The only thing worth noting that has changed is that I ripped 2 movies from the main server prior to restarting my frontend. It was not until Monday when I restarted my frontend that I noticed the issue. I still have 30 gigs of free space (of 820) on my LVM /myth partition so I don't think I am starved for space just yet.

Both my network cards come up properly (from dmesg):
server:
eth1: link now 1000 mbps, full duplex and up.
frontend
eth0: link up. (no indication is given for speed or mode)

ifconfig on each box says the interfaces are up. My switch indicates (dlink - by the status lights) that both are connected at 1000mbps.

When I ping I never get any packet loss and my response times are between 30 and 60ms in either direction. I know a small single packet ping is a far cry from streaming video so this is probably not a good indication of network stability.

I get no errors of any kind on the server. Not in any log files I have looked at. That includes all of the logs in /var/log/mythtv and everything in /var/log

I do get errors on the front end but it only amounts to the following:
nfs: server 192.168.254.1 not responding, still trying

eventuall it will display the message:
nfs: server 192.168.254.1 OK

and data will be transfered.

The logs when trying live tv are as follows:
server
2007-05-01 11:14:32.264 MainServer::HandleAnnounce Playback
2007-05-01 11:14:32.274 adding: mythfrontend1 as a client (events: 0)
2007-05-01 11:14:32.283 TVRec(1): Changing from None to WatchingLiveTV
2007-05-01 11:15:25.985 TVRec(1): Changing from WatchingLiveTV to None
2007-05-01 11:15:26.105 Finished recording Thomas & Friends "Responsibility": channel 1015

frontend:
May 1 11:12:41 mythfrontend1 kernel: nfs: server 192.168.254.1 not responding, timed out
May 1 11:13:36 mythfrontend1 last message repeated 14 times
May 1 11:17:12 mythfrontend1 kernel: nfs: server 192.168.254.1 not responding, timed out
May 1 11:18:14 mythfrontend1 last message repeated 8 times
May 1 11:19:20 mythfrontend1 last message repeated 16 times
May 1 11:19:28 mythfrontend1 last message repeated 2 times

When I do a df it comes back with the correct filesystem information from the server but sometimes it takes several seconds (up to 30 sometimes) and sometimes it comes back immediately.

the server is not loaded down as when this is happening I have never seen the cpu usage higher than 36% (64%idle) and having 2 gig ram I have never seen the server spill over onto the swap.

I have gone through all the troubleshooting on the linux nfs-howto site to no avail as the server never throws exceptions, I have just been fiddling with different settings in the /etc/exports and fstab to try to remedy the problem.

Everything pertaining to the network seems fine as ssh/sftp/etc. all work fine.

Running a stock R5D1 setup on both.

Andre

I have gone through the performance tuning section of the linux NFS-howto and found that I am in fact dropping huge numbers of packets.

a 'ping -f'' reports that I have almost a 5% packet loss.

a 'ping -s 8192' gives me a 100% failure rate. I slowly reduced the packet size down to 1024 and still get a 100% packet loss.

The 8192 block size is what I had been set up previously and my network was rock solid since January when I originally put up my front end until I went on vacation.

My question now is- Why all of a sudden am I dropping packets? Nothing physically changed, nothing virtual changed.

ethtool reports that both my interfaces are in fact set up for gigabit, full duplex and there are only 2 boxes plugged into the switch so there is no extraneous traffic that could be causing the dropped packets.

I am going to try to set up NFS using TCP instead of UDP tonight but I don't believe that is the correct solution as I had enjoyed the last 4 months of perfect performance. My only other thought is that either my switch or one of my network cards is flaky and needs to be replaced.

Andre