On Wed, 21 Apr 1999, Matthew Dillon wrote: > : > :With a full duplex setup collisions don't exist. In a switched setup the > :latency should be very consistent and extremely low. Something else must > :be wrong here. > > I should explain this more: It isn't actually the ethernet latency > that is causing the problem, but instead it is the packet backlog that > is occuring on the interface. Even at 100BaseTX speeds. Under a heavy > NFS load, an NFS client might send several requests simultaniously or > the server might send several responses simultaniously. If each request > takes 0.5 mS to transit the network, the first request will have a > latency of 1 ms, the second will have a latency of 1.5 ms. The > third 2 ms. And so forth. When N requests are queued simultaniously, > the last request in the queue will have a much longer latency then > the first one. > > Full-duplex setups do not have collisions, but latency can occur due to > internal switch routing latencies. >
Perhaps the timeout should be slightly adjusted it seems that it can cause panics, panics by users, not the kernel. :) As in... "what the heck is this?" when in reality everything is fine except a miniscule amount of latency on the interface. Some error messages should be constrained to DIAGNOSTIC so new users or people not interested in the kernel don't get nervous about freebsd. "Everything seems fine, but why am i getting errors?" -Alfred To Unsubscribe: send mail to majord...@freebsd.org with "unsubscribe freebsd-current" in the body of the message
