Normison….

Xanax….

Viagra. Viagra dosage. Order viagra. Viagra buy. Generic viagra. Free viagra. Best price on viagra….

Enzo on Fri, 18 Apr 2008 at 12:46 said:

The odd thing when I benchmarked Lighttpd, Lighttpd + Squid, and Lighttpd + Varnish, Lighttpd all by itself turned out to be faster than either reverse proxy plus Lighttpd. I was benchmarking with ab, not sure if that made any difference.

I wasn’t expecting lighttpd by itself to be faster though.

Really? Never heard before.
It’s appreciate if you could give your benchmark results(maybe a link?). Although most benchmark results is just for *reference*, which is not from real life.

BTW: I just have a *feel* that Nginx is faster than Lighttpd in production, according to my experience.

I wasn’t expecting lighttpd by itself to be faster though.

willy tarreau on Sat, 05 Apr 2008 at 03:46 said:

Nice update! This benchmark is quite better now. As you see in vmstat, with varnish, only system CPU is used and there is a lot of idle time remaining. That means that varnish is optimized to induce a very low overhead. On the other side, you see both squid versions. They use half user and half system time, and the CPU is saturated (50% idle on a dual proc means 1 proc saturated). So squid is limited here by its architecture and design. If it consumed less user time, this time would be better spent in system which would simply double the
performance, leading to approximately what you see in varnish. Maybe there are things that can be done with squid to optimize it, honnestly I don’t know. 25% user seems a lot to me on such a component, but since it’s a very flexible proxy I don’t know if there is an entry cost even if everything is disabled.

I think I will soon have to take a look at varnish, it looks promising :-)

Willy

Yeah, this time the result is much better :-)
We now have one project run varnish, I think it’s a better choice.
IMHO, there’s no doubt that varnish is more powerful by reading its architecture notes.
Thanks again for your time, Willy.

I think I will soon have to take a look at varnish, it looks promising :-)

Willy

About two weeks ago, I did a simple benchmark on varnish and squid, and wrote this Benchmark Caching of Varnish and Squid post.
Today Willy Tarreau gave me some advise. So I took a short time re-ran this benchmark. Here follows the detail.
platform
Eve…

Willy Tarreau on Tue, 01 Apr 2008 at 05:16 said:

Hey, you have saturated your 100 Mbps network, your files were too
big. You should retry either with Gigabit NICs, or with smaller files (far
below 10 kB) to see how each proxy performs. I suspect that they did
not eat much CPU either during the test.

Willy

Yeah, you’re right, that’s a point. The ethernet controller of the proxy server is a bottleneck.
I’m not satisfied with this benchmark result. I’ll re-run when I’ve got time.
Thanks for your advise, and your great HAProxy. :-)

Willy