Re: Looking for pointers

On Tue, 18 Apr 1995, Daniel O'callaghan wrote:
> On Thu, 13 Apr 1995, Al Hartshorn wrote:
> > We are about to put into effect a very large proxy/cashe server. This one will
> > not be just for one campus, but for all campuses in the country (no joke and 
> > this is not being sent on the 1st).
> > 
> > I am looking for anyone that may have some expertice with handling a very large
> > area. At this time, I do know that I will have 7 to 8Gb of disk space for cashe.
> > I may have to cut it down just a little, a Gig or so, for the local non 
> > proxy/cashe server.
> > 
> > The only thing that we have not picked as of this time, is the machine and op
> > for it. We will be running the Cern V3.0 server. 
> > 
> > Looking for any ideas or pit falls. No pointers to the warning.gif please.
> Don't run CERN httpd as your proxy server.  Have a look at 
> <>
> I'd go for a regional server and institutional servers which query the
> regional server.  I think the Harvest Cache is the right software to use.
> Danny

Does anyone have experience using the Harvest cache to do this sort of thing?

I am particularly interested in hearing from someone who has moved from Cern
3 to Harvest (or anything else) and how it went.

For the record, we are currently using the Cern 3 server on a dedicated
Sparc 5 with 32MB of RAM and a 9GB caching partition (just over 2GB
currently used).  This machine only accept requests from a known set of
"regional/institutional" caches for documents from overseas
(non-Australian) sites.  We have just started getting some decent load
from 3 Australian Universities and some CSIRO sites, and I think we are
seeing the start of some stress on the machine. 

Our load stats for most of yesterday (tuesday Apr 18) were:

		      Docs     Bytes	

Total requests      : 21385   (274.50 MB)     

External Fetches    : 14994   (186.56 MB)      
Fetches from cache  :  6391    (87.94 MB)     

Cache hit rate      :  29.9%  ( 32.0%)       

>From stats reported by iostat (60 sec samples), the Sparc 5 was commonly
50% non-idle (mostly system and wait, very little user time), the 9GB disk 
was doing 20 read or write ops a session and reporting 25% busy with 
"service" times around 60 - 80 millisec.

Kent Fitch                           Ph: +61 6 276 6711
ITSB   CSIRO  Canberra  Australia
"Unusual travel arrangements are dancing lessons from the gods."
				- Kurt Vonnegut

Received on Tuesday, 18 April 1995 16:41:03 UTC