reesesWeb

Digging out the orphan pages from the old blog, I came across this:

Interesting article on OpenTalk load balancing. Especially interesting is this quote:

Question from Len - which will distribute load better - a random round robin, or a sequential? Something for us to think about as he gets ready to run. The sequential distributed pretty evenly. There are some useful statistics available from the clients and servers - how long (aggregate) requests queued up, etc. Random round robin does worse, as the random assignment sometimes tosses "too much" load at a single server.

In addition, comments are made in re "least loaded" -- depending on the age of the samples, the "least loaded server" could be getting barraged with load even when it is no longer least loaded. It turns out a sequential round-robin is better. This has been consistent with my observations. A suggestion is to pick a "least-loaded pool", of half the servers or whatever, then distribute requests using another algorithm (random, linear) across those servers.

Bookmark/Search this post with:

delicious | digg | reddit | technorati