Just Launched: You can now import projects and releases from Google Code onto SourceForge
We are excited to release new functionality to enable a 1-click import from Google Code onto the Allura platform on SourceForge. You can import tickets, wikis, source, releases, and more with a few simple steps. Read More
From: Hanna Linder <hannal@us...> - 2002-08-23 17:31:01
LSE Tech Con Call Minutes from 08-23-02
1. Peter Wong -
2.4.17 vs 2.5.26 both patched and run under a very heavy
db load. Showed a performance benefit which is promising.
Cliff White asked what did you change in execv? Badari Pulavarty
made the changes. Basically to support the db. OSDL would
like to talk about what was changed because they might want
them. He will contact Badari.
2. Jenny -
we did 2.4.18 vs 2.5.29 and did not see much difference performance
wise except the memory pages behavior is different. tried on 2.5.31
got file.c fcntl() not available errors.
Cliff said the OSDL now has automated stp testing of the latest 2.5
kernels. Gerrit asked if they have tried running with oprofile which
they havent but think they should start. Some IBM 8-ways have a
problem with NMI's running oprofile. But the OSDL 8-ways are NEC
so they might not have the same NMI problems with oprofile. It might
be nice to verify if that is true or not.
They would really like some help tuning the db. so feel free to
send them email if you would like to help.
3. Paul Larson -
Doing nightly ltp tests of the most recent 2.5 bitkeeper kernels.
Every morning get an email with the results of what changed and
what broke. Comes in very handy because they are usually
related but not always also very useful to the developers
whose code changes are still fresh in their mind when the
results come in.
We are also doing code coverage analysis thanks to the
kernel gcov patches available on the lse site. to see
the code coverage results go to http://ltp.sf.net or
http://ltp.org. However, there appears to be a problem
with the gcov tools regarding block or line counts.
Paul will contact Janis Johnson to get help from the
4. Mala Anand -
skb init patch sent out last week and ben lahaise wanted
more performance results with all the cycles spent in the whole
area. He is happy with the new data provided and gave her more
suggestions for things to look at. Still needs to get
performance results from those suggestions.
Dave Hanson provided read profile results showing no benefit
from the skb patch. However, read profile only works off
sampling and Mala is not concerned with those results due
Next step is to continue working on the suggestions Ben
gave her and getting out the performance results.
Martin said she is changing the nature of the code
with her measurements so it is not a valid way to
prove the patch has a benefit. He agreed read profile
is not a good profiler for this type of patch but
dave also showed a reduction in throughput of the
benchmark with the patch applied (regardless of
Rick recommended she cc the netdev alias to get
feedback from more people.
5. Matt Dobson -
Submitted against 2.5.31 again yesterday. There are two
pieces he is trying to get in. The first one is the topology
which will be in driverfs now. second part is a memory binding
api to allow programs to bind their memory to particular nodes
or memory block. (request allocation of memory from a particular
area == bind).
He has been submitting many times and will submit again
with the driverfs changes. Will not use the system call or prctrl
solution anymore and will change it to use driverfs instead.
Hopefully more people will like this solution better.
Will resubmit today.
Matt requested patches from other architectures with the
macro changes be sent to him at colpatch@... and
he will include them.
Minutes compiled by hannal@...