We've been seeing intermittent network failures. The failure mode tends to be that long running connections tend to drop out. We see it most with ssh and samba connections, although some occasional http failures happen as well.
On the subnets that have these failures, we've been running a test that seems to correlate, and that's the time to form a ssh connection to the server. On affected subnets, where users also report problems, one in four attempts to open a ssh connection is exactly five or ten seconds longer than would be expected.
Is there a likely cause for why this type of failure would happen?