W3C home > Mailing lists > Public > http-caching-historical@w3.org > February 1996

Re: Ordered 'opqque' validators

From: Luigi Rizzo <luigi@labinfo.iet.unipi.it>
Date: Wed, 7 Feb 1996 09:06:06 +0100 (MET)
Message-Id: <199602070806.JAA24953@labinfo.iet.unipi.it>
To: fielding@avron.ICS.UCI.EDU (Roy T. Fielding)
Cc: http-caching@pa.dec.com
> > We know that there are some very popular servers on the internet,
> > so that caching traffic to these servers might look effective. But
> > the objects are single files/documents. What percentage of the
> > millions of files of these servers gets really accessed let's say
> > 10 or more times (within a reasonable time frame, say 1 week .. 1
> > month), so that caching really turns out to be a significant saving ?
> Like many things, it depends on the nature and size of the user base
> and the breadth of subject area being accessed.  In practice, all
> caches are effective -- if they were not, they wouldn't be used.

There is a theorem in statistics which says that the the combination of
large numbers of random variables yields to a gaussian distribution,
independently of the distribution of each single variable. Similarly,
when we have a national or continental cache, I expect the number of users
to be so large to make differences disappear and make the aggregate
result almost the same everywhere.

> > And, especially, how many of these files get accessed just once,
> > so that no caching policy is going to help ? I suspect getting
> > these info from some very large server would really help in
> > determining what we could expect from our cache, the depth of the
> > hierarchy, etc.
> I'm sure it would.  However, those concerns are irrelevant to our task.

Yes, I agree, they are not directly related to our task.

However, I expect people on this list to be interested not only in
designing the protocol, but also in designing/tuning applications/systems
which use it. To them it is important to know what is the best hit
rate they can expect from their cache, so as to avoid using, say,
100GB of disk space to get some 1% improvement over a 10GB
configuration, etc.

Luigi Rizzo                     Dip. di Ingegneria dell'Informazione
email: luigi@iet.unipi.it       Universita' di Pisa
tel: +39-50-568533              via Diotisalvi 2, 56126 PISA (Italy)
fax: +39-50-568522              http://www.iet.unipi.it/~luigi/
Received on Wednesday, 7 February 1996 10:55:18 UTC

This archive was generated by hypermail 2.3.1 : Tuesday, 6 January 2015 19:55:57 UTC