What is the consensus on KeepAlive being on with a high traffic Magento site? It seems most people recommend having it on but then Magento states to leave it off in their Maximizing Performance and Scalability with Magento Enterprise Edition, they claim
"When a web server is under a heavy load, keeping persistent connections becomes disadvantageous, thus the KeepAlive directive should always be disabled"
Thoughts?
Great question!
Traditionally, KeepAlive was a good thing because it significantly reduces the TCP overhead of an overall page load, where many requests (like all the images, css, js) are being served from one server. If your page has 85 assets, that's 85 extra three-way TCP handshakes, and the latency adds up. Many years ago with slower Internet connections, this was much more important than it is now, although it is still quite relevant for mobile browsers or any slow/high latency connections.
The impact mentioned here though relates to the amount of time those TCP connections are kept open, and whether that might quickly exhaust your Apache child processes. Most of the defaults I've seen are:
KeepAliveTimeOut 15
MaxClients 256
This means that if I had 256 different browsers request content within the same 15 seconds, then the 257th client would have to wait for the connections to drop off. Not good - this isn't especially high traffic at all, so it explains that advice. It may also lead one to increase MaxClients to cope, which could eat a lot of memory. When I do use KeepAlives, I generally set KeepAliveTimeout to 2 or 3 seconds; it's the idle time between requests rather than the entire time for all requests.
If you are using KeepAlive, there's a balancing act between KeepAliveTimeout, MaxClients, and server resources. To help with that, a 'service httpd/apache2 fullstatus' will show you the number of connections being used by KeepAlives at any one time - indicated with the uppercase 'K'.
For Magento though, I don't think you need KeepAlives.
What you should be doing if you have a very high traffic Enterprise site, is using a CDN for your static content.
If you are selling in multiple countries, using a CDN will not only speed up overall page loads for your customers (which is good) but massively reduce the bandwidth coming to your server. The settings under System > Config > Web > [Un]secure make it really trivial to integrate any CDN for the Media, Skin, and JavaScript. This will be the bulk of the actual HTTP requests, and as a bonus you can use different DNS records to parallelise downloads across hostnames. If you're doing it right then those requests will barely touch your server, so there's no longer any real need for KeepAlive. In this case, you should disable KA; we don't want to keep the connectiona live when we know the rest of the content is being served from a somewhere else. If you want a standalone CDN recommendation: CloudFlare is fantastic, and you even get SSL thrown in with the free package.
When using a CDN like this, or if you're using any other kind of reverse proxy (like Varnish), your Apache HTTP KeepAlives are basically irrelevant.
To summarise, I agree that you should probably disable KeepAlive to avoid saturating your Apache processes under load, but definitely use a CDN or other reverse proxy for the assets, to keep your page loads as snappy as possible.
