Re: hot object memory patch

From: Lincoln Dale <ltd@dont-contact.us>
Date: Thu, 10 Aug 2000 18:05:39 -0400

At 08:38 10/08/00, Alex Rousskov wrote:
>On Thu, 10 Aug 2000, Lincoln Dale wrote:
> > in a real-world environment, it'd probably make sense (and add to
> > performance), provided the amount of physical ram is tuned
> > accordingly.
> >
> > in terms of testing such a change with polygraph, i doubt it'd make
> > any difference. we see <3% in-memory-hits with 2gb of in-memory
> > "hot" objects with polygraph, when in a customer deployment, this is
> > >55% of hits are served from.
>
>Polygraph workloads can be configured to have "hot set" of arbitrary
>size and temperature so you can simulate real-world environments with
>high/low memory hit ratio.

ok, let me rephrase what i'm saying then:
  - in a real-world deployed proxy, we see a large % of objects served
    from memory, when there are suitable algorithms for determining
    policy of storing in-core objects and there is sufficient ram to
    store them.
    when i say large %, it is typically >50% on a "warmed up" cache

  - in standard polymix-2/3 tests, we see <3% of hits from in-core
    objects.

i have no doubt that polygraph can be configured to reflect this more
accurately - i think we've both had this conversation before - but that in
"default parameters mode", it doesn't appear to be a big win when tested
with polygraph.

this merely reflects the rules that govern the polymix-2/3 workloads -
which aren't perfect, but are the best yet.

cheers,

lincoln.

--
   Lincoln Dale           Content Services Business Unit
   ltd@cisco.com          cisco Systems, Inc.       |         |
                                                    ||        ||
   +1 (408) 525-1274      bldg G, 170 West Tasman  ||||      ||||
   +61 (3) 9659-4294 <<   San Jose CA 95134    ..:||||||:..:||||||:.. 
Received on Thu Aug 10 2000 - 16:07:43 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:12:34 MST