This page is dedicated to Squid caching proxy of the Australian Defence Force Academy. The patched version of Squid (v1.1.17) was run on Wednesday, November 05, 1997. We also have statistics for other days.

Note that on this proxy normally uses a sync kludge. However, for comparison purposes, the kludge was not active on Wednesday, November 05, 1997.

A detailed description of the environment extracted from cache owner e-mail follows.

The proxy this comes from is about 5.3G effective size (3x2G spindles) on a dual CPU sparc 20 with upgraded CPUs (150MHz hypersparc). It has 284M of RAM, and normally runs squid and a number of glimpseservers - during most of the day it would have had about 20MB of RAM free. The squid version was 1.1.17 with your patches and the double read bug fixed. The machine is a stratum 3 ntp server so timestamps should be within 10msec of UTC. We were having some DNS problems on our main campus nameserver until about 0945 local (+1100 currently) so that lookups were slow till then, though this was not particularly busy. The machine has 3 SCSI chains with logging/most unix on one string, one 2G drive on a 10M synchronous chain and 2 drives on a 20M synchronous chain.

The neighbour/network topology is roughly as follows. The local cache is on an ethernet switch which also interfaces onto an FDDI ring. The ring has two 7000s forming a collapsed backbone with 30 main segments (10Mbit/s) hanging off it. One of the ciscos has a 45Mbit/s ATM link to a switch which interfaces via a similar link to the University of Canberra and its proxy. It also goes into the regional router which has ethernet links to the ANU and CSIRO, as well as the regional parent on its own segment (currently 10Mbit/s).

We split our experiments into 6 categories.

1. General
   • Transfer Sizes
   • File Sizes
   • Proxy Traffic Intensity
   • Proxy Response Time
   • Repsonse Time vs. Size
   • Concurrent Requests
   • Hit Rates
   • Request Response Time Components (Misses)
   • Request Response Time Components (200 Hits)

2. Disk
   • Disk Traffic Intensity
   • Concurrent Disk Requests
   • Disk Utilization
   • Disk Response Time
   • Disk Response Time Anatomy
   • Atomic Disk Requests

3. Memory
   • Memory Hit Analysis
   • Hit Classification

4. Outbound Network
   • Concurrent Outgoing Connections
   • Proxy Connect Time
   • Distribution of Proxy Connect Time (Peak)
   • Server Reply Time
   • Server Reply Time vs. File Size
   • Server Response Time

5. Inbound Network
   • Client Connect Time
   • Proxy Reply Time
   • Proxy Reply Time Anatomy
   • Proxy Reply Time Versus File Size
   • Proxy Reply Time Versus File Size (Detailed)

6. Hits
   • Hit Classification
   • Memory Hit Analysis