Subject: Re: stability of May 5 kernel
To: None <>
From: Mark P. Gooderum <>
List: current-users
Date: 05/06/1994 11:37:22
> I tried out the kernel from May 5 - after 10 hours of compiling X, a
> new kernel and several X applications it's still running. From the May
> 2nd kernel I was getting 2-3 hours between reboots under a lesser
> load.
> Is the current tree (esp. kvm) in a shape that I dare try out
> everything else?
> Thanks

I've seen similar.  I was getting dinked about 2-3 times a day by the
May 2 kernel (random SEGVs on ar's, etc).  I'm almost all the way through
building the May 5 tree under the May 5 kernel w/o a problem.

Another tidbit.  I have a simple dumb performance test I wrote.  It 
fork()s a child, and then the child sends several megs of data back to the
parent using domain sockets (both kinds), UDP, TCP, and a pipe().
It sends the data using increasing message sizes (16, 256, 1024, and 4K),
then prints out KB p/sec, msgs p/Sec, etc.

This is very non-VM intensive, zero disk I/O, pretty much pure CPU but 
it seriously beats on context switching, the IPC kernel code, etc.

Anyways, the May 2 kernel is about 2-3% slower than the Apr 12th and
28th kernel (which were about 20% faster than the March 15th one, the
locore changes made a difference).  However, I've never had the TCP
test succeed on NetBSD before.  It would run with 16 byte writes just
fine, but as soon as it went to 256 byte writes both procs hung.  They
were still killable, but the socket locked up.  Ironically, Solaris 2.0
and 2.1 had the same problem.  BTW, this code is a pretty portable set
of socket() calls.  It all compiles without source changes or any conditional
#defines except for which headers it needs (and the link has to change)
on SunOS 4, SunOS 5, HP/UX 8 and 9, AIX 3.2.5, and NetBSD.

But with the May 2 kernel, all tests now complete, so some of the fixes 
somewhere in the last two weeks made a difference.