I have a friend that had a really weird problem with a program that used RPC. Every once and a while the program that used RPC would open a port and tell the client to contact on the port it just opened. The client would try to do that and fail. Problem was this only happened randomly. Random problems are a pain to track down. The problem was eventually tracked down. The following is a description of what it was and the fix for it.
RPC uses a range of ports on Linux to allow clients to contact the server. By default these are 650-1023. RPC will pick a port at random from the range of ports. That is as long as port in the range is not already being used. Their servers have built in Baseband Management Controllers (BMC). BMC communicates with a BMC management utility (BMU) on a remote client using IPMI protocols. The BMC for the servers are in band and use the network controller on the motherboard to send and recieve data. The ports the BMC uses are 623 and 664.The thing is the controllers don't run any software in Linux to do this. So how do they work you ask? Well, they intercept the traffic on those ports before it gets into the OS's network stack. That means any data you send to those ports on those machines the OS will never see. You won't see it on any network sniffer on that machine at all. So any machine that RPC told to connect to it on 664 was failing. But it was random because that port would only come up from the pool of ports randomly.
Now that you know what the problem is how do you solve it? You could just run a dummy process out of xinetd to listen on port 664 (and 623 if your RPC uses ports that low) and be done with it.
You could also try to bump up your port range to start at 665. That can be done on some 2.6 linux kernels in proc.
echo 665 > /proc/sys/sunrpc/min_resvport
Restart your daemons that use RPC. If that works then you can put line sunrpc.min_resvport=665 in /etc/sysctl.conf so it sets on reboot. I have heard that some of the latest Linux kernels are going to up the port range to start after 665 to avoid this problem in the future.
RPC is most notably used by NFS. This can cause NFS mounts to fail randomly or just any other program that uses RPC.
