John Peacock <[EMAIL PROTECTED]> writes: > OK I checked my VMS account and I have Bytelm of 64000 (which strikes > me as probably high rather than low). I do have Pbytelm of 300; is that > the one that needs to be upped? I'm pretty sure it's the Bytlm that is relevant. >> >> What do people think about production environments here? There are >> MBX deadlock situations with small mailbox sizes, right? >> >> Should we be setting BYTLM much higher for accounts that might be >> runnng perl now? Even if you're not running the piping code, in >> preparation for it? In testing the piping code, I tried to hit the mailboxes with as high an i/o rate as possible to get it to deadlock. If you go to http://www.crinoid.com/perl560.htmlx and scroll down to "Pipe Torture tests", the test_pipe.pl is the one to try. It does stuff like filling up mailboxes before subprocesses have a chance to read them, closing them prematurely, etc. Even with 256 byte mailboxes, the piping code passes the torture tests. When you try fill a mailbox, the code is either checking to make sure the mailbox is empty prior to writing or immediately (via AST) removing messages as fast as they come in. You *can* get i/o problems (of the 'extra record separator' type) if you have mailboxes that are smaller than the records you are writing. But the usual reason for having large mailboxes to try and avoid full-mailbox deadlocks stochastically just doesn't apply anymore. The "default" mailbox size was left as it was prior to the piping fix (i.e., mailboxes of maximal size) out of concern for deadlocks. But, as we've seen in the past week or so, running out of quota is a more likely source of problems.
