Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!ames!haven!aplcen!jhunix!ecf_hap
From: ecf_hap@jhunix.HCF.JHU.EDU (Andrew Poling)
Newsgroups: comp.sys.att
Subject: Re: sockets for SYSV (UNIXPC)
Summary: doesn't work right = panic
Keywords: select, timeout-table, panics
Message-ID: <1185@jhunix.HCF.JHU.EDU>
Date: 21 Mar 89 19:07:54 GMT
Expires: 21 Mar 89 00:00:00 GMT
References: <612@ivucsb.UUCP> <454@polyof.UUCP>
Reply-To: ecf_hap@jhunix.UUCP (Andrew Poling)
Followup-To: comp.sys.att
Organization: The Johns Hopkins University - HCF
Lines: 35

In article <454@polyof.UUCP> john@polyof.UUCP ( John Buck ) writes:
>If you get the UNIX PC WIN TCP package, you will get a BSD 4,2
>compatible library with it, including full sockets.  Beware, though,
>they don't work exactly correct (IE the implementation is flawed;
>select doesnt work right, for one thing).

	You mean it's not supposed to panic the system? :-) Actually, let me
	temper that remark with specifics.  I have run into problems with
	select(3W) on numerous occasions.  If you have too many long
	selects(3W) active at once, it will overflow your timeout table ==>
	PANIC.

	My solution (based on intuition more than fact) has been to limit my
	selects(3w) to one second.  This does hurt my performance, since it
	makes my machine impatient, but it allows me to run such goodies as
	BIND, Sendmail5.58, and other resolver-based routines that I have
	found SysV ports of here and there.

	On the positive side, 4.2 BSD socket code usually runs out of the box
	as soon as you fix all of the other SysV/BSD (ahem) problems.

	As a side point - I noted in a bug report release for the 3.* code
	for our 3b4000 warns of possile panics in (as I read it) much the
	same way.  I guess Wollongong only ports this stuff once, then
	re-uses the same buggy code. :-(


	Disclaimer:  These are my own opinions - don't blame my employer -
	they don't even know that I think. :-)


Andy
andy@gollum.hcf.jhu.edu
ecf_hap@jhunix.UUCP
ecf_hap@jhuvms.BITNET