LSE Minutes for 24 Jan 2003

LSE Minutes for 24 Jan 2003

Post by Hanna Linde » Sun, 26 Jan 2003 01:30:18



LSE Con Call 24 Jan 2003

Michael Hohnbaum -  NUMA Update

Last call mbligh mentioned his mini numa schedular.
By late friday night Erich Focht had generated the last
piece of code needed (node rebalancing). He debugged it
and Linus picked it all up by Thursday!
Congrats to Martin, Erich, Michael, Andrew and everyone
else for getting it in.

As an added bonus Ingo Molnar responded to the
patches being accepted.  He made some additional
changes which has been debugged and it should all
work together well.

Andrew Theurer added code to help the numa scheduler
dispatch hyperthreaded processes. It was more of a
proof of concept than final solution for hyperthreading.
Ingo then made additional changes to the O(1) sched. and
he took some fixes that Andrea Arcangeli made to the O(1)
sceduler.

Michael did some basic runs and it appears there are some
nice performance boosts according to sched bench (on numaq).

There are still some issues that hbaum is looking at related
to poor load balancing.

Dave Hanson asked about lock contention on shared runqueues for
hyperthreading? Michael isnt using any hardware with hyperthreading
now.  Andrew is the one looking at that now.

If you are trying to support Hyperthreading then
Martin Bligh mentioned you should use ingo's patch
called D7 which provides one runqueue per pair of
hyperthreaded processors. You can use D7 without CONFIG_NUMA.

Could someone please write a list of patches or hacks that might be
helpful for working with hyperthreading? (Andrew? Martin?)

Pat Gaughen was interested in talking to Erich about
support for topology stuff. Hanna has figured out how to
call out to international numbers so Erich, I will really
be able to call you next time!

Hanna Linder- Move lse project of of sf?

        - leave lse-tech mailing list where it is.
        - sourceforge has no money to support any special requests
        - Consider Tigris, Savanagh.nongnu, OSDL umbrella project for scalability.
        - Discussed moving to osdl. Should talk to nathan and bryce.
        - Hanna will walk across the street and talk to them in person.
        - Hanna did talk to them and it is looking like a promising
          option.
        - Does anyone see any issue with moving lse to osdl?

The plan for now is just a place to easily put patches.
The osdl has other tools to enable automatic testing
of patches which we hope to utilize in the future.

        - Hanna will add offset from GMT for future announcements of lse calls.`
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in

More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

 
 
 

1. Minutes from LSE Con Call

On elm3b17 I found more 2.5.25 vs. NUMA-Q issues and got workarounds
for booting while doing work on stuffing pte_chains into highmem:

(1) cpu_init() heisenbug that goes away when adding printk's
        I just added the extra printk's and moved on
(2) in_interrupt() is racy and needs to disable preemption
        while checking per-cpu counters. The trivial fix.
(3) CONFIG_PREEMPT calls preempt_schedule() on secondaries
        before they're ready, and the exception is flagged as
        no vm86_info: BAD
        Didn't want to go to deep into it at 3AM and disabled preempt.

The usual ioremap()/smp_call_function() deadlock, irqbalance
plugging unreachable cpu's into ioapics, bio splitting to make
qlogicisp happy, and MAX_IO_APICS things are still needed.

So I guess the "real fixes" need to get found at some point and
propagated out to maintainers. hpa and mochel are on cpu_init(),
and rml should be on preempt vs. printk and in_interrupt(). axboe
has already got bio splitting ready and no idea about the MAX_IO_APICS
and getting irqbalance fixes past mingo doesn't look easy.

Cheers,
Bill
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in

More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

2. MPI for Linux?

3. Minutes from March 21 LSE Call

4. Can the function "connect()" & "listen() + accept()" be used with the same TCP port at the same time ?

5. Minutes from Feb 21 LSE Call - publishing performance data

6. how to open a program on a specific desktop

7. Minutes from LSE Call March 7

8. Can you echo from another tty??

9. Minutes from Feb 21 LSE Call

10. Minutes from LSE Call March 7

11. Server shipments [was Re: Minutes from Feb 21 LSE Call]

12. Minutes from Feb 21 LSE Call

13. COLA FAQ 3 of 6 04-Jan-2003