Best effort and PolyFill-2

From: Alex Rousskov (rousskov@ircache.net)
Date: Thu Dec 16 1999 - 11:59:19 MST


On Wed, 15 Dec 1999, Alan Smithee wrote:

> With PolyFill-2 workload, there are approximately 1500 concurrent
> requests for 1000 robots. I thought "best-effort" meant one
> outstanding request per robot. What is going on?

        The PolyFill-2 mimics PolyMix-2 in terms of robot configuration
(except for request rate which is absent in PolyFill-2). Just as in
PolyMix-2, each robot is allowed to open at most 4 HTTP connections.
Thus, you can have up to 4000 open connections and somewhere between
1000 and 4000 outstanding requests at any given time. The actual numbers
for both measurements will depend on the speed of your cache and the
algorithm it uses to maintain persistent connections.

The fill workload is still best effort in a sense that robot request
arrivals are not Poisson distributed but arrive immediately after
replies. The only difference is that now robot has more (4) connections
to handle things like embedded objects.

In other words, you can think of current best-effort fill as best-effort
in terms of "pages" rather than "objects" and with at most 4 connections
to fetch a page. A single robot will not request pages concurrently, but
may request [embedded] objects concurrently. A page can be defined as a
"container" together with all embedded objects. Current workloads have
only one type of a container that can actually contain more than zero
objects: HTML "content" (cntHTML).

Alex.



This archive was generated by hypermail 2b29 : Tue Jul 10 2001 - 12:00:10 MDT