On Fri, 7 Jul 2000, Alan Smithee wrote:
> I've recently noticed that the working set size (WSS) with Polymix-3
> seems much larger than WSS with Polymix-2.
We've recently noticed that as well. :-/
> In particular, there is a fundamental difference in the working set
> calculation, and I want to know if it is intentional. In Poly-2,
> the working set time was chosen to be equal to the time of the
> plateau phase (4 hours). The "fill rate" in that time was 25%
> (since there were 55% offered hits and 20% uncacheable).
True.
> The current calculation says
> working_set_length(platDur/FillFactor); // XXX: this needs more work!
>
> This gives the same number of total client-side requests as in
> Polymix-2. However, the fill rate now is 75% (since the working set
> freeze occurs during the "fill" phase, which has 5% offered hits and 20%
> uncacheable). So, this means roughly a 3X bigger working set than
> Polymix-2.
That is the reason behind the "needs more work" comment. We want to have
WSS in PolyMix-3 similar to the one in PolyMix-2, but could not come up
with a general formula. The problem is with the fill phase duration. The
fill phase duration depends on the proxy cache size and can be very
short or very long. On the other hand, to preserve fair conditions WSS
should not depend on the proxy cache size. Thus, we need to write
something like this:
if fill phase duration is long enough
set working_set_length assuming phFill's fill rate
else
set working_set_length assuming phFill's fill rate in
the beginning and phTop1's fill rate at the end
but PGL does not have conditional statements (and the above formula is
still missing some minor details).
> I hope to hear back from you soon, as we'd really need to know the
> proper working set lengths before we can fire off the reports that we
> should be sending you this Monday.
The next Polygraph release will have a PGL function to compute working
set length given the fill and peak rates. For now, please use a
pre-computed constant that gives the same working set size as 4 hours of
peak request rate given your workload parameters. If you make a mistake
in calculations, it is not a big deal! The "soft registration" report is
not meant for reporting the exact product performance. We just want to
be sure that you can complete a PolyMix-3 test and we know what bugs to
fix.
Once again, you can expect WSS to be close to 4 hours of the fill
traffic during a "top" phase unless we hear sufficient number of
arguments to increase that value.
Thanks,
Alex.
This archive was generated by hypermail 2b29 : Tue Jul 10 2001 - 12:00:14 MDT