Re: 100-110 response/s Polygraph performance from Squid 2.2 using Solaris 2.7 and "RAMdisk" on 270 MHz Ultra5

> Henrik Nordstrom <hno@hem.passagen.se> replied:
> > Peter Jenny wrote:
>
> > I thought by eliminating hard disks Squid would perform much faster, but
> > I'm only seeing ~110 responses/second from my lab test configuration:
>
> If you want a baseline of what the CPU is capable of, try
>
> no_cache deny all

100 responses/sec with Squid Version 2.2.STABLE5, "no_cache deny all," 270 MHz
Sun Ultra 5, Solaris 2.7 and Polygraph "datacomm-1" workload. 100 13 KB
objects/sec is about 10 Mbps on the wire.

(neck sticking out:)
It seems that Squid is "CPU intensive" or as some say a "CPU hog" -- no
offense intended. Hard disk and the file system aren't it's only
performance-limiting factors.

I should have a loaner Sun Ultra 10 440 MHz this week and will try that.
Per SPECint95 benchmarks the 270 MHz UltraSPARC is about a 9.
Per SPECint95 benchmarks the 440 MHz UltraSPARC is about an 18.
Per SPECint95 benchmarks the Intel 650 MHz "coppermine" (full speed 256 KB L2
cache is about a 31 (I don't have one of those to try).

So if CPU speed is the limiting factor (via use of some RAM-based file system,
or many hard disk spindles), and it's more "integer intensive" than "floating
point intensive," Squid 2.2stable5 should be able to go to ~200 responses/sec
on the 440 MHz Ultra or ~300 responses/sec on a 650 MHz Pentium.

And consider rewriting section 3.1 of the FAQ (from "Your processor does not
need to be ultra-fast" to perhaps "For Squid 2.x, 300 MHz of CPU and 5 SCSI
disks are fairly well balanced. A 600 MHz CPU should have about 10 SCSI disks
for caching).

Peter
Squid newbie.

                   Peter Jenny
            Internet Engineering Services
         GTE Internetworking - powered by BBN
3 Van de Graaff Drive, PO Box 3073, Burlington MA, 01803
       pjenny@bbn.com http://www.bbn.com
       Tel: 781-262-6041 Fax: 978-428-7020
Received on Mon Dec 13 1999 - 16:07:12 MST