--- Pei Cao <cao@Theory.Stanford.EDU> wrote:
> oh gosh, the reason that I am putting input onto
> this is because Lee
> Breslau (Xerox PARC), Li Fan (U. Wisconsin), me,
> Graham Phillips (USC ISI)
> and Scott Shenker (Xerox PARC) wrote the INFOCOM
> paper arguing that the
> workload follows a Zipf-Like Distribution with alpha
> of 0.64-0.83. I didn't
> want my colleagues to say "why didn't you say
> something during the benchmarking
> discussion".
>
> And Cisco's Cache Engine handles the uniform
> workload perfectly fine :-)
That's good to know.
I don't know all the details about the zipf-like
distribution. But I can offer my view as a cache
buyer. Here are my points for favoring the
uniform distribution, again as a cache buyer.
(To make a simulation behaves like real-world events
is very very difficult, if not impossible, i.e.
virtual reality != reality period.)
1. I think to do disk I/O well is tougher than to
handle memory hits well.
2. If I know a cache can handle disk I/O fast, I know
it will perform well when the object space is
huge and diverse.
3. Polygraph uniform model is a very good test on how
well a cache's disk I/O performs.
That's all folks,
Amy
__________________________________________________
Do You Yahoo!?
Thousands of Stores. Millions of Products. All in one place.
Yahoo! Shopping: http://shopping.yahoo.com
This archive was generated by hypermail 2b29 : Tue Jul 10 2001 - 12:00:09 MDT