What are the pros/cons of that?

Apache doesn't have to load php for every requests including static requests, and mpm_worker is threaded so uses less ram.

_________________
Paid support
How to ask for help
1. Give details of your problem
2. Post any errors
3. Post relevant logs.
4. Don't hide details i.e. your domain, it just makes things harder
5. Be polite or you'll be eaten by a grue

I'd be very careful going the FastCGI route. Apache's PHP FastCGI implementation is a mess. It sorta works if you configure it just right, but the setup is complicated, inelegant, and nowhere near as robust as lighttpd or nginx. Remember, lighty rewrote the book on FastCGI for a good reason.

I got burned pretty bad by Apache/mpm_worker/fastcgi a while ago, and nowaways I prefer to give all my static files to nginx.

I was able to get it down to about 40ms slower than Lighty with zip and a better wp-super-cache configuration (turns out it wasn't working properly cuz of permissions).

Apache Bench with 10000 connections and 10 concurrent gave a 40ms mean time lead to Lighty, but it's longest attempts were longer than any one attempt on Apache.

The big part seems to be the transfer rate. Lighttpd is transferring at 321 Kbytes/sec and Apache is transferring at 298 Kbytes/sec. Is this really a function of the web server? I would think the web servers would be so fast that transfer rate would be more capped by the link.

But they're both linodes in the same data center, so I would expect the transfer rate to be similar. Is that an incorrect assumption?

.

Are you running ab from a linode in the same datacenter, or are you running it from home? Your last sentence is a little ambiguous, and 300KB/s sounds suspiciously similar to what you'd get with a crappy "high-speed internet" service in the US.

Also, are you requesting a static resource, such as an image file, or a dynamically generated page, such as the home page? If the request goes through PHP at all, the transfer rate will drop dramatically because of the latency imposed by the PHP interpreter. The transfer rate is meaningless when benchmarking PHP scripts. Requests per second is a better indicator of performance. (The longest attempt takes much longer probably because Apache with mpm_prefork puts extra requests on the waiting list until it's done with previous requests.)

I wouldn't worry too much about 40ms latency differences. What you need to look out for is memory consumption of the server while you're hammering it with lots of concurrent requests. Try upping the concurrency to 50 or even 100. Run htop on the server in a different terminal and keep an eye out for memory consumption. If there's something wrong with your Apache configuration, memory will go up dramatically and the server might grind to a halt. As long as that doesn't happen, you're fine.

I was running ab from home. I started doing so from the datacenter.

In the end, I just couldn't tune Apache to my liking. Rather than learning nginx for this particular server since I was on a deadline, I just stood Lighttpd back up. It outperforms Apache in pretty much every discernible way. I'm sure there must be some reason to use Apache over Lighttpd, but I have convinced myself that I can't see it (especially for my needs).

Thanks for all the help. I certainly learned a lot if nothing else.

Next up on my list is to figure out why my php5-cgi processes are all using so much processing power. Under a typical "heavy load" scenario for my site, they're using 200% processor power in aggregate. Seems excessive just for some PHP.

What sort of scripts, and what sort of traffic? A bytecode cache like APC can help reduce the CPU usage of PHP, since PHP recompiles scripts every time they're run.

I've got APC running. I'm running Wordpress, Piwigo and Vanilla 2. I suspect that Vanilla 2 is the main culprit but I have nothing to back that up. Just history of problems with forum software being typically more demanding. I've got wp-super-cache on wordpress, so I think that's about as efficient as it can be.

And overall, the site actually runs like a dream. Everything is served up super fast and I couldn't be happier with the performance. I just don't want to upset my neighbors by stealing their processing time.

You can't steal others processing time, everyone gets an even share and the ability to use unused cpu.

_________________
Paid support
How to ask for help
1. Give details of your problem
2. Post any errors
3. Post relevant logs.
4. Don't hide details i.e. your domain, it just makes things harder
5. Be polite or you'll be eaten by a grue

Well that's a relief. I guess I should then worry about not having any unused processing time to borrow and my server getting slow. So far that doesn't seem like it's going to be a problem, but still seems like I should put in my due diligence.

If you haven't already install something like munin, it allows you to track resource usage over days/weeks/months/years so you can see if it looks like you're bottlenecking some where if/when things slow down.

_________________
Paid support
How to ask for help
1. Give details of your problem
2. Post any errors
3. Post relevant logs.
4. Don't hide details i.e. your domain, it just makes things harder
5. Be polite or you'll be eaten by a grue