Actually my question was more general but you have some very good points.
Currently our clients are users over modem lines (we are an ISP) so I
guess you could say we have slow clients and there are many. Peak
connection rate is around 45,000 connections per hour, cache size
is around 10 gig. Hit ratio is around 0.35. Memory would be the
resource I want to make most efficient use of.
I made the block sizes 4K with out doing a dissertation on VM disk
cacheing or the internals of squid, but buy simply looking at the
read histogram via the cachemgr and noticing that %80 of the reads
are 4K or less.
Any explanations of when 4k is good and when 8k is good is welcome.
-
Barrett Richardson rabtter@orion.aye.net
On Mon, 20 Jul 1998, Alex Rousskov wrote:
> On Mon, 20 Jul 1998, B. Richardson wrote:
>
> > squid
> >
> > -
> > Barrett Richardson rabtter@orion.aye.net
>
> Squid request response time for slow clients? Squid request response time
> with fast clients? Squid disk response time for hits? Squid disk response
> time for cachable misses? Squid disk space usage? Squid memory usage?
>
> In other words, what is the metric you are optimizing and how will 4KB pages
> affect it? In most cases, there will be a tradeoff. It is not 100% obvious
> [to me] that 8KB disk page is worse in all respects... Please specify the
> metric or explain why 8KB pages are worse for everything we do.
>
> > On Mon, 20 Jul 1998, Alex Rousskov wrote:
> >
> > > On Mon, 20 Jul 1998, B. Richardson wrote:
> > >
> > > > I am theorizing that if a large proportion of files are 4k or less,
> > > > then 4k is a better choice for block size than 8k.
> > >
> > > What are you trying to optimize?
> > >
> > > Alex.
>
>
Received on Tue Jul 21 1998 - 10:29:22 MDT
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:41:12 MST