[H-GEN] Summary: Stable RH Linux Kernel with SMP support

[Laurence J. Rietberg]

Thankyou to all those who responded to my initial query regarding a stable
version of RH Linux Kernel with SMP
support.

Here is a summary of the responses I have received.

The general consensus was that 2.0.x kernels do in fact support SMP, but
since they only have 2 interrupts, they effectively disable one
processor--effectively rendering the machine to operate with just a single
processor.

This was an interesting response, and I guess I'm having trouble sorting
fact from opinion here, but I received this interesting response from Ian
Crakanthorp <ian at ansto.gov.au>
>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>.
The 2.0.35 kernel does have SMP support, but you have to manually add it
to the Makefile.

In /usr/src/linux<whatever>/Makefile

#
# For SMP kernels, set this. We don't want to have this in the config file
# because it makes re-config very ugly and too many fundamental files depend
# on "CONFIG_SMP"
#
# NOTE! SMP is experimental. See the file Documentation/SMP.txt
#
SMP = 1
#
# SMP profiling options
# SMP_PROF = 1

We are using it and works quite well.
>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>

Another point mentioned was that SMP is not supported by any stable
kernels, and one respondent remarked that even Linus himself doesn't claim
SMP support until at least release 5.2, and has even been providing sample
code for some SMP users to try out 

Development kernels do provide SMP support, but of course they don't
provide the stable assurance of stable kernels.  [Interestingly, it was the
use of a stable kernel which seemed to cause all the problems with
dual-processors in the first place, which a develpmental kernel seemed to
solve.]

In sum, the best approach if you have an SMP machine is to "keep on the
bleeding edge" of kernel development, trying new kernels until a kernel
which is stable on your machine has been found.

One reply noted another who had successfully obtained a reliable kernel
from the bleeding edge:

"In short, if you can grab 2.1.125, or (if you're brave) 2.1.126 with
all the 2.1.127 prepatches you can find, you should have a fairly
stable SMP system."

So far I've gotten 2.1.117 (DMA request time out errors initially, then
debugging output from the kernel prior to crashes) to last about 2 weeks.
I tried 2.1.125 which lasted 2 days before the server locked up (eth0 not
found errors from the kernel were being produced).

Now I'm giving 2.1.118 a try.  So far it is going well, except for some
more kernel logs which had occasionally appeared with 2.1.117: TcP4: bad
checksum...

One administrator was confident that SMP support will gradually improve
with release of the next level of stable kernels ..." 2.2 will be out there
soon, so the 2.1.x tree will firm up."

Laurence
-
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
Laurence J. Rietberg, BA.  Systems Programmer (Part-time)
Language Laboratory, (CLTR-Arts)
The University of Queensland,	
St. Lucia, QLD 4072 AUSTRALIA		
Work hours: Mon8-12:00; Tue8-4:30; Thur & Fri 12-4
Email: lr at cltr.uq.edu.au
Phone:  +61 7 3365 6916
Fax: +61  7  3365 6919; Phone (A/H): +61 7 3841 2185
N.B. the statements in this email are my own,
and do not necessarily reflect those of my employer.

Skepticism: "If q is incompatible with p, and x claims
to know that p, then x's reason for believing p must
exclude the truth of q, if x really does know p."
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=