Re: Cache traffic - odd?

From: James R Grinter <jrg@dont-contact.us>
Date: Thu, 4 Sep 1997 15:30:00 +0100

On Thu 4 Sep, 1997, Simon Rainey <srainey@rmplc.net> wrote:
>served. However the MRTG traffic analysis shows about equal traffic flow,
>and sometimes the inbound traffic greatly exceeds the outbound. Some of
>this can be explained by client HTTP requests, sibling UDP queries and DNS
>traffic, but I would have thought that these would be small compared to the
>actual data served.

Really you need to break down this into 'from/to server and from/to
clients' (which would be easy if you had a different route to your
clients as against your default route to servers. We don't either).

However, taking a snapshot of a Netcache system we have here (we've
currently got a bit of everything), because it counts plenty of
things:

  Total bytes sent to servers : 1,085.582 MB
  Total bytes read from servers : 17,968.373 MB
  Total bytes sent to clients : 26,442.826 MB
  Total bytes read from clients : 1,330.103 MB
  Byte hit rate : 25.05%
  Connection cache hit rate : 12.00%

I'm not quite sure what that Byte hit rate value is calculated from,
I've not played around with the numbers themselves. This particular
host reports 0.56 in the Utilization section, and whilst it isn't
Squid I know it's calculating that the same Harvest-heritage
way. (Processing the logs produced a 51% hit rate for TCP HIT
yesterday, using Calamaris.)

I also don't know where siblings come into this - I expect they're
bytes read from servers, and bytes sent to clients. We get about 20%
by bytes of what we didn't have from our neighbour servers.

Bytes sent to clients, on another screen, breaks down into:

  Bytes Sent To Clients As A Hit = 7,485.138 MB
  Bytes Sent To Clients As A Miss = 22,612.594 MB
  Bytes Sent To Clients That Were Generated By The Cache = 83.200 MB

(they don't quite add up to the same, because it was a few seconds after).

in otherwords, benefit is down to how you measure it. (Any saved
traffic is good, to my mind, though it would be quite nice if we could
do something about many mirrors of the same information.)

-- jrg.
Received on Thu Sep 04 1997 - 07:32:20 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:36:55 MST