View Full Version : PBI's cuse system crash with smp complied kernel
08-16-2005, 05:13 AM
Particularly noticed with Firefox, screen goes blank, reboot follows, eventually killing the system.
I recomplied the generic kernel config with the added option SMP and ULE scheduler - no other changes.
I dont suppose you can fix this, so I have built from ports and not installed any pbi's on this build - so far so good.
Just a heads up for new users of SMP machines.
08-16-2005, 06:27 PM
The ports built firefox fell over in the same way on a page that requires flash - could be at the root of the problem as I hadn't installed the linux plugin wrapper.
Even so - I'll stuick with the KDE default browser as the sysent crashes are damaging.
DON'T USE ULE WITH SMP. Trust me on this. ULE works with SMP once again in 6.0, but not in 5.4.
You don't mention if it is for a multiprocessor, dual core or a hyperthreaded CPU, but my dual Athlon was terribly slow and unstable with ULE until I upgraded to the 6.0-Beta (now Beta2). 5.4 was fine with 4BSD. I'll leave to others whether this affects hyperthreaded CPUs or not, but ULE is known to be broken until 6.0 with SMP.
08-16-2005, 08:34 PM
its a dual processor system, i'll bear that in mind, thanks - if it crashes again i'll recomple with the std scheduler
I would leave out your qualifier "if it crashes again." Please recompile your kernel with 4BSD. You will save yourself a lot of trouble. ULE can give a bit snappier system, but the cost on 5.4 is enormous. I lost about a factor of 5 in throughput on 5.4 using ULE. Note -- not 50%, but a factor of five. In my mind it is simply not worth it, if you are tracking -stable.
I did upgrade to 6.0 to get a stable ULE, and by and large it has been worth it. But not on 5.4 or -stable.
08-17-2005, 11:22 AM
Done, thanks for the advice :)
vBulletin® v3.8.5, Copyright ©2000-2013, Jelsoft Enterprises Ltd.