please remove from the list
----- Original Message -----
From: "Henrik Nordstrom" <hno@hem.passagen.se>
To: "Chemolli Francesco (USI)" <ChemolliF@GruppoCredit.it>
Cc: <squid-users@squid-cache.org>
Sent: Friday, May 11, 2001 12:42 PM
Subject: Re: [squid-users] Sizing a pretty Large Squid Install
> Chemolli Francesco (USI) wrote:
>
> > Yes.
> > The idea is to exploit (if possible) elevator optimization in moving
> > the drive heads, which requires having multiple outstanding
> > requests on the same drive. Diskd only has one active disk operation
> > at any given moment AFAIK.
>
> Not only elevator optimization, but also I/O queues to make sure there
> is another disk I/O request pending immediately when the first have
> completed. A single process cannot do this (on reads) as it has to wait
> for the first to complete before it can schedule the next I/O operation.
>
> > Probably mostly irrelevant under normal conditions, but it can help
> > decrease consistency checks and crash recovering.
>
> It should help a lot in decreasing total I/O latency times for cache
> hits.
>
> Does not help very much for consistency checks and/or crash recovery.
> Those are still limited by Squids single main process only capable of
> doing one small fragment at a time..
>
> --
> Henrik
Received on Mon May 14 2001 - 01:49:19 MDT
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:59:58 MST