Here's a report, that you may ignore if needed...

I'm building news machines with two partitions for OSen, to allow
me to boot into my choice, where my choice has been FreeBSD-STABLE
or FreeBSD-CURRENT to see how the two compare, and if there are any
significant improvements in -CURRENT.

I know, ``don't do that'' but hey...

Anyway, using the performance with -STABLE as a reference on this
system with currently a single CPU, I built a freshly cvsup'ed
-CURRENT just under 24 hours ago and then ran it in production for
about ten hours before reverting back to -STABLE.

First of all, after building a custom kernel and mounting several
disks with softupdates, I then gave a command to cp -pR /news/dir
to /news/FreeBSD-STABLE-dir , where the /news disk is mounted with
both softupdates and noatime.

Quickly I got a  panic: ffs_valloc: dup alloc  and everything froze
solid.  I didn't attempt to repeat this to see if it is repeatable.
I disabled the softupdates, remounted the disk (just noatime) and
again gave the cp -pR command, which succeesed.  In fact, for the
next ten hours, I attempted to pump a full newsfeed through this
machine with no problems and stable operation.

A few other drives are mounted with both noatime and softupdates,
but without the file creation activity one gets with the command
I gave.  Also, I was sort of running low on inodes, although I never
actually ran out, if that would make any difference.

Now, as far as performance goes, after running for ten hours and
getting a feel for how well it was doing, I rebooted back into
-STABLE and restarted things.  However, I see a huge performance
increase with -STABLE compared to -CURRENT.  That is, I'm able to
take in many more times the number of articles with -STABLE than
the machine running -CURRENT could handle.  Like by a factor of ten.
Basically, apart from the current/stable switch, the machine is
identical in both OSen, and there shoulr be no difference in the
news software proper.  The kernel configs should be comparable too.

Yeah, I know, -current is in a state of transition, but I didn't
expect its performance to be quite *this* bad...

These are just some observations, in the hope they might be useful.

barry bouwsma, putting hardware to waste since 1997
(use reply-to header if this message is worthy of comment best kept
off the list)

To Unsubscribe: send mail to [EMAIL PROTECTED]
with "unsubscribe freebsd-current" in the body of the message

Reply via email to