I was running two pkg_comp builds today (-current sources as of a few
days ago)  when (unexpectedly of course) it seemed like there was some
problem with some I/O subsystem; the two pkg_comp processes seemed
like they were blocked waiting for some filesystem request to be
fulfilled.  When I exited them the 'make' processes and those beneath
them could not be killed; other processes were fine.

When I went to reboot the machine, I never got past "umounting
filesystems..." though the disks synched fine.  My machine is
overclocked, but I haven't had any problems from it heretofore (ran
OpenBSD on here as well with no problems like that, though I never
used null mounts because IIRC they don't have them).  Note that I had
an async mount null mounted into the chroot jail for the obj
directory, and then the chroot jails themselves had their respective
null mounts.

I am using BUFQ_PRIOCSCAN with the constant in bufq_priocscan.c
(forgot what it was called, but IIRC it is the only one) set to [ 512
128 32 ] as per a discussion I read a few days ago on how to get the
maximum throughput.

I can provide kernel config/dmesg if need be, or anything else.

Thanks,

--Blair

--
"What is the practical application of a million galaxies?"

--Alan W. Watts