On Thu, 1 Mar 2001, Ian Holsman wrote:
> > The poor thing is spending most of its
> > time in mutex operations. :-(
I sort of expected as much. :-/
> > I'm surprised that it's doing so much malloc/free
> > work (rather than using a pool allocator). At least
> > on Solaris, all the malloc/free activity is going to
> > be a huge bottleneck, both because it uses a lot of
> > CPU and because of the mutexes in malloc and free.
Unfortunately, it can't really be avoided in the current scheme... all of
the buckets code uses malloc/free because allocating from a pool would be
a potentially large memory drain up until the time the pool gets cleared.
I've been trying to eliminate as many calls to malloc as possible in work
I've done over the past 48 hours or so... more work is yet to come if I/we
can figure out a clean way to do it (I still want to implement bucket
reuse lists, for example, which would be a large step toward eliminating
these mallocs).
Thanks for the data!
--Cliff
