NFS VPN network hangs

longtex · 10-11-2005, 06:04 PM

I have a really strange problem.

Here are three computers, HQ, S1, and S2.

All three are linked via the 'net using SonicWall firewall/VPN.

S1 and S2 have each mounted a shared directory on HQ:

mount -t nfs 192.168.0.24:/stores /wanserver

Periodically (once a minute) S1 and S2 look at their own subdirectory on /wanserver to see if there is any new data for them. If there is, they copy the new files to their own HDD and process it.

For some reason, the process of copying files occasionally hangs. It just stops, and the program that's directing the transfer (it kicks off a simple "cp -f /wanserver/mydir/myfile.xxx /here/inbox" and waits for it to complete). Most of the time, everything's great, and the file is transferred in a couple of seconds (most of them are 5-10K), but that every once in a while... hung... than it may take 15 or 20 minutes to continue.

I don't know whether to blame nfs, or the SonicWall, or the ISP's access points, or what.

ANY ideas???

Agrouf · 10-13-2005, 02:27 PM

try using ftp and see if it makes a difference (in which case you can blame nfs).

longtex · 10-20-2005, 07:58 PM

Quote:

Originally posted by Agrouf
try using ftp and see if it makes a difference (in which case you can blame nfs).

Eli at Flying Pig Systems (light shows extraordinaire - Rolling Stones, U2, Olympics, Super Bowl... etc etc) runs hundreds of computers to control 30,000+ lights, pyrotechnics, etc with super-critical latency tolerances - he tells me that nfs is VERY sensitive to problems in the network commections...

this explains the problem, i think. the particular ISP's connection is wireless through several antennas and access points, and when the thing gets busy (someone on one of the ap's apparently is in the habit of downloading loads of bit-torrents), nfs chokes and has a hard time coming back. with a hard connect, it just locks up; a soft connect would kill the using program, which is almost as bad as a lockup.

our decision was to change the transmission from online to ftp.

i don't think i "blame nfs" for this, it's really the ISP in this case. i suspect that a set of non-shared ap's would probably work, but "probably" isn't good enough... ftp is less convenient, but it works.

thanks for the response