BarneyC wrote:
The 800 RPS is on a 100Mb/s network? The 5000 RPS? That was my question. I'm
not really concerned with what "other" people get on their systems, as every
system will be different. I want to know how many RPS I can expect to
encounter on a busy 100Mb/s residential network.
I have a fairly special case, in that the majority of my customers are
on the far end of a satellite link (each client also has a Squid server
on premise), but perhaps you will find the combined numbers from my
central servers useful as a data point to extrapolate from.
Client side traffic: 58 Mbit/sec
Internet side traffic: 44 Mbit/sec
Req/sec: 215
Number of unique clients: 150 (remember, this is the number of child
Squid servers. Real client numbers are north of 10,000).
Each server runs Squid 2.7STABLE6 with the following hardware.
2 x Xeon 3110
8GB RAM
3 x 45 GB COSS
3 x 150 GB aufs
My servers are not remotely taxed. Squid's CPU usage hovers around 10%
and the load average (5 or 15 minute) rarely rises above 2.
Chris
