> > The above ratio looks too high, even for solaris. On my system,
> > page faults is slightly (15-20%) less then TCP connections. (some TCP
> > connections result in IMS hits, so no physical I/O, assuming it
> > wasn't swapped out by the OS).
>
> Not exaclty. All IMS hits (mostly 304 and 200 replies) require at least one
> disk I/O to read HTTP headers from disk. 200 IMS hits actually fetch the
> whole object from disk (Squid still considers it a "hit" since nothing was
> fetched over the outbound network).
>
> There are so called "memory hits" when no disk I/O has been made. However,
> Squid does not tag memory hits for non-internal objects.
>
> Some negative hits and non-cachable misses may not require a disk I/O, but,
> again, it is hard to isolate those cases using just "fgrep" and "access.log".
>
Oopps, I was getting IMS hits mixed up, and was thinking they were
hits from in-memory (cache_mem) objects...
---------------------------------------------------------------------------
John Lauro email: lauro_j@msb.flint.umich.edu
University of Michigan - Flint jlauro@umich.edu
Information Technology Services
303 E. Kearsley St. phone: (810) 762-3123
Flint, MI 48502 fax: (810) 766-6805
Received on Mon Apr 27 1998 - 14:15:48 MDT
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:39:58 MST