Slides for tomorrow night
Maged Michael
magedm at us.ibm.com
Mon Oct 18 16:05:42 BST 2004
Andrei.
The slides look good. If you feel time is tight you might consider moving
the slides about lock-free structures as backup. The simple and compelling
case is made by the flag example and user-level locking.
A possibly helpful anecdote, the contention-free latency of pthread_mutex
is about 75% (170-180 ns vs 95-100 ns) more than a user-level
test-and-test-and-set mutex implemented on new 1.7 GHz Power5 (the mutexes
are protecting a shared counter). Furthermore, if you are using the lock
to protect a single variable read-modify-write operation, and replace that
with a user-level implementation using CAS or LL/SC, the contention-free
latency (42-45 ns) drops to about 25% that of pthread_mutex and 45% that
of the user-level mutex.
Maged
Andrei Alexandrescu <andrei at metalanguage.com>
10/18/2004 03:26 AM
To: Kevlin Henney <kevlin at curbralan.com>, Maged
Michael/Watson/IBM at IBMUS
cc: Doug Lea <dl at cs.oswego.edu>, Doug Lea
<dl at altair.cs.oswego.edu>, "Boehm, Hans" <hans.boehm at hp.com>, Ben
Hutchings <ben at decadentplace.org.uk>, pugh at cs.umd.edu
Subject: Slides for tomorrow night
Hello,
Tomorrow is the presentation of our paper. I have slapped together
some slides, for which quality I need to apologize in advance. I am
seriously under stress here with a number of tasks that need be done
in negative time, and at a point I almost canceled the presentation.
But somehow I miraculously found one hour to write these slides, and
what I have left to do is pray for a silver tongue.
So if any of you has time, let me know of any feedback. The
presentation will be 30 mins plus 15 mins q&a. I think there are too
many slides for 30 mins, and that the flow could be seriously
improved. Actually everything can be seriously improved :o). If you
have time, I'd be happy to hear from you.
The presentation is tomorrow 19:30 PST.
Thanks,
Andrei
#### N1680.ppt has been removed from this note on October 18, 2004 by
Maged Michael
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://shadbolt.decadentplace.org.uk/pipermail/cpp-threads/attachments/20041018/5f6cc6d1/attachment.htm
More information about the cpp-threads
mailing list