On Mon, 17 May 2010 14:50:28 +0200, Georg Höllrigl <georg.hoellrigl@xxxxxxxxxx> wrote: > Hello Everyone, > > > How many HTTP requests can squid as reverse proxy do? This question is frequently asked. It is also somewhat akin to asking "how fast is the wind blowing?" The answer depends on a lot of environment factors, such as the CPU speed, RAM available, disk speeds, and size of objects in the website(s) being proxied. > > Im getting an average of 4500 per Minute which are aroun 75 requests per > second. > When do you think a squid will give up (given a load with many small > files)? 2.7 benchmarked around 800 req/sec a few years ago (regular setup with RAM caching). 3.0 lab-tested at around 500-600 req/sec a year ago (regular setup with RAM caching). 3.1 lab-tested at around 650-750 req/sec a year ago (regular setup with RAM caching). Fastest Squid seen so far is a reverse-proxy 2.7 setup benchmarked around 990 req/sec. Running the "extreme CARP frontend" example configuration with OS tuning. > > What would fail first, when I see around 80 - 90 % CPU idle time? disk or TCP stack. Tuning the TIME_WAIT status and persistent connections to clients will be important for very small files. They leave Squid very fast and can really churn through available sockets. > > Are there any numbers to compare anywhere? > Some old reports are at http://wiki.squid-cache.org/KnowledgeBase/Benchmarks Amos
