cak@purdue.ARPA (Christopher A. Kent) (10/22/85)
I'm developing a system which, among other things, uses the network to log data to another machine. I noticed that, at times, the log file grew incredibly quickly when there was little or no actual activity. The machine running the system also often crashed (with panic: sbflush 2) shortly thereafter. If it didn't crash, I never got new data in the log stream, just the same block or part of it repeated forever. I dug into the log files, and discovered that blocks of data were repeated; these blocks were always an even multiple of 2k (usually 4k or 8k.) I cast suspicion on my code, on stdio (which I tried to use on the sockets, but it doesn't really work right), and finally on the TCP. I discovered that at times I do very large (>30K) writes in one system call, and wondered if this might be the problem; sure enough, when I recoded to never write more than 1K/syscall, the problem disappeared. Has anyone else seen this? Is there a fix somewhere? I don't have the time or inclination to track it down, but it should be repaired... Cheers, chris ----------