On Thu, 8 Apr 1999, Admin-Eng wrote:
> So,is there any great performance boost between 2.0.36 and 2.2.x for
> beoowulf ?
I have to defer this question. The system I'm upgrading is my personal
system and while I'm building a beowulf at home that will be based on
5.2+2.2, it isn't finished as I'm also building a tool to build the
system and IT isn't finished. And my kids just got "Diablo" and talked
me into playing (rats!) and I have to sleep sometime...;-). Our beowulf
at Duke (brahma) will be upgraded to 5.2 (currently Slackware) and 2.2.x
at the same time, once we get all the kinks out of RH+2.2.x -- our
regular Systems person is working on prototyping the RH stuff but I'm
responsible for making sure that we can run 2.2.x on it.
>From what I've read on this list and the linux-smp list, though, there
are certainly great performance boosts to be found in the 2.2.x upgrade,
but not necessarily in raw CPU. 2.2.x is MUCH better at SMP, especially
at managing spinlocks, which can result in much better network
performance, disk performance and so forth. It is much more intelligent
at handling IRQ's (it uses the IO-APIC, for example). It can deal with
PnP. It's SMP load averages are reputed to be correct (so broken in
2.0.x that it was decided not to bother to try to fix them). Overall, I
think that there is no question that it is a much "better" kernel than
2.0.36 and almost certainly will yield better performance on various
beowulf bottlenecks.
Soon I should have data. Soon. Of course, on this list there is bound
to be somebody with actual numbers either at hand or readily
determinable.
rgb
Robert G. Brown http://www.phy.duke.edu/~rgb/
Duke University Dept. of Physics, Box 90305
Durham, N.C. 27708-0305
Phone: 1-919-660-2567 Fax: 919-660-2525 email:[EMAIL PROTECTED]
-
To unsubscribe from this list: send the line "unsubscribe linux-net" in
the body of a message to [EMAIL PROTECTED]
