RELENG_6: Which scheduler for SMP?

[Michael Vince]

Mark Ovens wrote:

> I've never had any success with the ULE scheduler on my dual Athlon 
> box running RELENG_5; it was so unstable it made Windows 3.1 look 
> stable. In fact my current build, cvsup'd a couple of days ago, won't 
> even boot with ULE.
>
> From what I remember, ULE was intended to become the default scheduler 
> during the life of 5.0 but that hasn't happened.
>
> I've just cvsup'd the source for RELENG_6 and I'm surprised to find in 
> the GENERIC config file:
>
> #options    SCHED_ULE    # ULE scheduler
> options     SCHED_4BSD    # 4BSD scheduler
>
> so it seems 4BSD is still the default scheduler. Is ULE _still_ 
> considered to be in development/experimental? Even the SMP config file 
> doesn't use ULE.
>
> Also in GENERIC:
>
> options     PREEMPTION    # Enable kernel thread preemption
>
> [....]
>
> options     ADAPTIVE_GIANT  # Giant mutex is adaptive.
>
> Yet in src/sys/conf/NOTES, ADAPTIVE_GIANT is listed under SMP Options 
> and  PREEMPTION is listed under SMP Debugging Options which makes it 
> surprising that they are in the GENERIC config file.
>
> What I am trying to decide is whether there any point in making the 
> jump from a very stable RELENG_5 system to RELENG_6. AIUI the ULE 
> scheduler and it's associated options optimize the use of multiple 
> CPUs and by staying with 4BSD I'm not getting the best performance 
> from my system.
>
> Can anyone offer any advice on this please?
>
> The machine has dual Athlon MP2800s on a Asus A7M266-D mobo, 1Gbyte 
> ECC RAM and all SCSI disks and optical drives.
>
>
I have been benchmarking a Java servlet under ULE, and I couldn't get 
result scores as high under ULE as I could under the regular the 4BSD 
(although it wasn't far off) and when I left the machine benchmarking 
all night under ULE I came back in the morning to find the machine 
unresponsive and in need of a hard reboot.

Mike