Re: finegrain locking in LDBM (relates to dissection of search latency)

[Mark Adamson <adamson@andrew.cmu.edu>]

> Interesting. Can you tell me on which hardware you tested, because it looks like
> my hardware produced some skewed results. I tested on a dual athlon 1.4ghz with 1gb of
> memory.


I ran it on a single 450MHz CPU Sun Blade with 500MB of memory and on a
Sun Ultra2 with 2 300MHz CPUs and 1GB of memory.

What sort of "skew" did you see?  Faster/Same/Slower?  Because there are
more locks and shared locks take a few more cycles to obtain, you might
not see any help on an untaxed machine. It's when there are a lot of threads
bustling about on the server that clients see slowdowns, and these fixes
are designed to alleviate that kind of slowdown.  So you don't get
benefits until the machine is busy.

-Mark Adamson