frode at nordahl.net
Wed Feb 9 14:54:46 PST 2005
I have been reading curiously about the ULE scheduler on the lists for
quite some time without ever getting any clear good or bad feeling
about it, so I thought it was about time to give it some real
I briefly tried it on my desktop about a year a go, and felt that
everything was a bit "choppy".
I currenty have several systems under testing, awaitng to go to
production once I have time to finish other projects, and can get focus
back to getting them ready.
I've been trashing a dual Xeon 3.06GHz box running Apache2 (default
port settings, default page) using two dual Xeon clients running
webbench -c 300 -t -1 (300 processes per box).
Running with 4BSD the server processed aprox. 400 requests at a time,
my ssh shell was very slow, and I often had to wait for up to a minute
for simple commands like "uptime" to return.
Switching over to ULE improved things drastically. The server now
serves aprox 580 pages at a time, and the ssh shell is as snappy as if
the system was completely idle (even though the load is at 590 :-) ).
Enabling DEVICE_POLLING as well, increases the number of concurrent
sessions handled to aprox 600.
I was surprised by the actual difference in system performance and
usability between 4BSD and ULE under such loads!
If you haven't tried it on your heavy trafic server yet, go and do it
right now! :-)
On Feb 8, 2005, at 13:07, Mipam wrote:
> I saw several changes to sched_ule.c in the 5 stable branch.
> Beneath is one of them:
> Is the ULE scheduler still far from stable in RELENG_5 or not?
> freebsd-stable at freebsd.org mailing list
> To unsubscribe, send any mail to
> "freebsd-stable-unsubscribe at freebsd.org"
More information about the freebsd-stable