Re: SNMP stats & Cachemgr.cgi

Marc Lucke wrote:

> Considering that the figures are almost identical for
> cacheProtoAggregateStats 4 & client_http.kbytes_in &
> cacheProtoAggregateStats 5 & client-http.kbytes_out, I make the assumption
> that these are the same.

They are.

> What exactly is cacheHttpOutKb & cacheHttpInKb?

Amount of data sent and received to/from HTTP clients (on http_port).

> I have assumed that client_http.hit_kbytes_out divided by
> client-http.kbytes_out would give me the ratio of cache hits to the total
> number of kb served which would give me a kb hit ratio of about 22%.

It will give you one kind of hit rate. A more appropriate hit rate is
perhaps delta(client_http.hit_kbytes_out) / delta(server.all.kbytes_in)
Where delta is the difference over the time period you which to measure.
This measures all overhead injured by the cache in terms of ICP or
run-away requests into the hit rate calculation.

> What am I missing? Or what have I got wrong/misunderstood? Can I have a
> hit ratio of 45% & a hitkb ration of just 5% in a normal Squid setup?

See above.

No 45% request hit ratio with 5% byte hit ratio does not make sense in a
normal setup, and as you see above 22% is closer to the truth for your
cache.

---
Henrik Nordstrom
Spare time Squid hacker