Tech Tip #32: Maximizing Performance while Minimizing Load

As the SlipStream SP server becomes loaded with users, it begins to work harder: the server's load average increases, more memory gets used, and the processor usage increases. Assuming a constant increase in user load, the server becomes more and more taxed until the MaxLoadAverage is reached. When the MaxLoadAverage is reached, text compression and images are affected, and users will start getting redirected to other machines (if available). Finally, the whole system becomes loaded at an increased rate.

By properly tuning your SlipStream SP server, you can actually minimize the rate at which server load increases, thus delaying and even preventing the sudden transfer of users from a taxed machine to another; in effect, making your load balancing cluster work more efficiently.

What is MaxLoadAverage?

If the operating system load average for the past minute (as reported by 'uptime' and scaled down by the number of CPUs) exceeds this value, then:

1. Some image compression is disabled (the affected images are passed through to the user unchanged).
2. Some text compression proceeds with reduced gains.

As the load average increases past this value toward the 'MaxLoadAverage' setting, image and text objects fall into the previous two categories. Once the load average exceeds 'MaxLoadAverage', all text and image objects are affected.

What is ReduceCompressionLoadAverage?

If your server simply increases its load without correspondingly optimizing its performance, the increasing load will affect the server's Load Average, causing the server to reach its MaxLoadAverage setting a lot faster. Essentially the load produces a feedback loop - as the server becomes more loaded, it can't handle as much load; therefore it works harder, and more load is heaped on it.

With the ReduceCompressionLoadAverage setting, the server optimizes its performance by anticipating some of the extra load and reducing the work on some of the bigger compression tasks. As load increases, the server reduces the number of objects being compressed until it reaches the MaxLoadAverage setting, where it then operates as any other server that has reached its MaxLoadAverage limit.

The end result is a more gradual slope of the increase in load, making for a healthier server, and a longer period of useful service for users during peak time. Instead of a sudden drop-off in performance, you will see a moderate compression reduction to help the server through high-load periods, and users will be less impacted by busy servers.

Load Average and Load Balancing

MaxLoadAverage is one of the triggers used in redirecting users from an overloaded machine to another in a Peer-to-Peer (P2P) load balancing cluster. It's also the most commonly used trigger. By using ReduceCompressionLoadAverage, you can reduce the load on other servers by reducing the effects of user load on a loaded server. This has the net benefit of increasing scalability across your server cluster.