On Friday 16 November 2001 01.05, Alex Rousskov wrote:
> Clients think that the requests are directed to a single "origin"
> server. That server is represented by a surrogate (a.k.a., reverse
> proxy or Web accelerator). The surrogate distributes the load among
> several true origin servers. Those origin servers are assumed to be
> identical as far as content is concerned.
Ok. Makes sense.
> You can configure Polygraph to test multiple different origin servers,
> accelerated by a single cache, but that would not be a WebAxe
> workload.
Fine.
> Working set "slides". The changes in working set are slow due to very
> high recurrence ratio of WebAxe. The set slides with the rate the new
> objects need to be introduced based on recurrence ratio.
Ok. So when a URL has fallen out of the WSS it is very unlikely to get
requested again. Sort of different than a real world situation but will do..
> > c) I read that the test used a 1GB set of origin data. Does the
> > "cache size" have any relation to the test at all?
>
> > Found the answer to the question on cache size (none, except for
> > the fill phase)
>
> The fill phase is not affected by the cache size for this test. You
> may be thinking of PolyMix where the duration of the fill is
> determined, in part, by the cache size.
>From your WebAxe-4 documentation:
4.3 Proxy cache size
Proxy cache size is the configured cache size plus the total amount of RAM
that the proxy box has. Configured cache size is whatever is specified in
proxy configuration file or the best approximation of that. High/low water
marks for garbage collection and other proxy-specific settings and algorithms
should not affect this parameter. Proxy cache size is used to determine the
duration of the fill phase and does not have direct effect on other phases
(though there may be performance side-effects, of course).
> It is a good idea to fill the cache first though. We just wanted to
> shorten WebAxe tests because we do not expect performance to depend
> much on the disk subsystem (as opposed to PolyMix). Most reasonable
> surrogate configurations will cache the majority of WebAxe objects in
> RAM.
Agreed.
> You can run 4 server agents inside one polysrv process if needed. Just
> make sure your server host is not overloaded.
Good.
> > Why is this requirement on at least four servers?
>
> Two reasons: many surrogates accelerate more than one physical server
> and to make sure the surrogate has an option of balancing the load
> across multiple servers.
>
> If your environment is different, and you are running private tests,
> you may want to violate this rule and use just one server.
I am perfectly happy with your answers. Will give this a try shortly.
Regards
Henrik Nordström
This archive was generated by hypermail 2b29 : Mon Feb 06 2006 - 12:00:20 MST