Does OpenSSH employ some sort of Nagle's algorithm at the SSH protocol level? I just experienced a rather strange situation: xterm running on host A. From there, interactive ssh to host B. I run a program on B that generates lots of output to stdout. Host A is equipped with a hifn(4) crypto accelerator used by ssh. Traffic from B to A is on the order of 250,000 bytes/s and 200 packets/s. I see on the order of 10,000 (!) interrupts/s from the hifn card. That seems very bizarre, unless there are lots of tiny SSH packets in each IP packet.
(OpenBSD 4.2, ssh is all defaults.) -- Christian "naddy" Weisgerber [EMAIL PROTECTED]
