SOMAXCONN woes (BSD 4.x and beyond)

[Christopher Seiwald]

Can anyone tell me why SOMAXCONN (the maximum queue length specifiable
by listen(2)) is so low?  I have a server which simultaneously handles
about NOFILE-3 clients (60 on a Sun), all of which would like to
connect at the same time.  BSD4.x and its decendants allow 5 clients to
line up on the listen queue, and then refuse the rest.  This leaves my
client application to retry (with timeouts - blech).

Is there any reason this non-configurable parameter hasn't been scaled up
to somewhere around NOFILE?

E-mail and I'll summarize.

Christopher
seiwald at rtech.com