Re: Problems with choosing faster parent cache!

Bartlomiej Solarz wrote:

> > > cache_peer cache_slow parent 3132 3134 proxy-only closest-only no-delay
> > > weight=1
> > > cache_peer cache_fast parent 8081 3131 proxy-only closest-only no-delay
> > > weight=1

> This one?

Yes.

> Cache Manager menu
>
> Parent : xxxx
> Flags : proxy-only no-delay
> Address[0] : xxxx
> Status : Up
> AVG RTT : 18 msec

Note. The RTT to XXXX is slighly less than to YYYY.

> PINGS SENT : 16429
> PINGS ACKED: 16429 100%

Fine. no lost packets.

> FETCHES : 55169 336%

Hmm.. it looks like most requests never went thru ICP pinging at all.
There was only 16429 ICP pings, but 55169 fetches to this peer.

Are you using prefer_direct off? If so, make sure to mark the preferred
parent as default, or have it listed first in squid.conf.

> IGNORED : 6571 40%
> Histogram of PINGS ACKED:
> ICP_HIT : 1450 9%
> ICP_MISS : 14979 91%
> keep-alive ratio: 66%
>
> Parent : yyyy
> AVG RTT : 19 msec

> PINGS SENT : 16429
> PINGS ACKED: 15857 97%
> FETCHES : 2482 16%
> IGNORED : 7956 50%

Looks more normal for an ICP peer, except that it didn't get selected
very often even when ICP pinged. Probable cause to this is if it is
slower when responding to ICP (most likely), or if the XXXX peer has a
much better hit ratio (not that likely, the difference is too large).

--
Henrik Nordstrom
Squid hacker