Amos Jeffries wrote:
Can someone please say how well Squid 3.1/tproxy scales? Would it have
problems servicing more than 10k simultaneous HTTP requests, and pushing
as much as 300 mbit/s of traffic? 500 mbit/s? 1 gbit/s?
Planned hardware & setup:
Dell Poweredge 6850 server QUAD Dual Core 3.4GHz 8GB
Hard Drives:
cache_dir will be split across
5x 73GB SAS 15K hard drives
All will run on Ubuntu 9.04/testing mix w/ WCCPv1.
Thanks in advance.
As for overall scaling. Less than 10K requests though for those on regular
proxy duty. We've seen 50Mbps pipes being flooded by Squid-3 no problems.
Very conservative! Let me mind you: a somewhat less powerful Dell
PowerEdge 2950 w E5410 Quad-Core Xeon and 4 gigs RAM under FreeBSD 7.1
currently allows us to use ng_netflow to pass the traffic locally to a
flow-capture process (flow-tools-0.68), which analyzes it and dumps
Netflow files once a minute. At 300 mbit/s, the box looks almost idle
(cpu time mostly spent in handling interrupts). Is Squid's main loop
hungrier?
> Above that you need to start tuning stuff like disks, reducing ACLs,
> multiple Squid etc to fill the pipe.
I'm not sure I know how to further optimize 15K SAS disks :)
But I'll take your advice and optimize ACL usage, thanks.
Multiple side-by-side Squids is also an option we might consider in the
future, even without bothering to set up a cache hierarchy, and thus,
allowing duplicated resources.
