Re: Seeing context switch storm with 10/13 snapshot
Remember the suggestion I made that PostgreSQL add the capability
to define named caches and bind specific objects to those caches?
One of the reasons Sybase recommends using such named caches
(per their performance tuning documentation) is to reduce spinlock
contention.
I don't know whether PostgreSQL would get a similar benefit, but
this discussion rang a bell, so I thought I should throw it out there
on the chance that it might be useful.
-Kevin
Tom Lane <tgl@sss.pgh.pa.us> >>>
The problem right now is not lack of ability to reproduce the
problem, it is lack of ideas how to fix it.
Try getting lucky in google with "spinlock contention is greater"
Bruce Momjian <pgman@candle.pha.pa.us> >>>
How do other databases deal with this? I can't imagine we are the only
ones. Are we doing something different than them?
Import Notes
Resolved by subject fallback
On Mon, 2005-10-24 at 13:08 -0500, Kevin Grittner wrote:
Try getting lucky in google with "spinlock contention is greater"
Kevin's reference is actually to a Sybase manual page; very interesting.
This refers to the idea of subdividing the lock into may partitions.
That's already been discussed, but that is the next step beyond where we
are right now with a single cache but a much improved cache algorithm.
That technique is an algorithmic improvement rather than a hardware
specific improvment, which is what is required in the case we are
dealing with here.
Cool reference.
The idea of named caches is more familiar to me. Tuning for that is just
too hard and massively inflexible, so I don't think its the way anybody
really wants to go, but I hold the door open for anybody travelling that
way.
Bruce Momjian <pgman@candle.pha.pa.us> >>>
How do other databases deal with this? I can't imagine we are the only
ones. Are we doing something different than them?
I'm not sure the people qualified to answer that are able to do so. What
do other OSS projects do about this is more likely an answerable
question.
Best Regards, Simon Riggs
"Simon Riggs" <simon@2ndquadrant.com> wrote
How do other databases deal with this? I can't imagine we are the only
ones. Are we doing something different than them?I'm not sure the people qualified to answer that are able to do so. What
do other OSS projects do about this is more likely an answerable
question.
I happened to know that a similar problem which is called "convoy
phenomenon" is discussed in 1979 (Wow! :-)) But not exactly the same case
maybe.
I copied the paper and put it here:
http://www.cs.toronto.edu/~zhouqq/convoy.pdf
Regards,
Qingqing