Web server clustering for Windows-centric sites

By Thomas Powell, Network World |  Software

We used a New Cluster Wizard to create a general Web cluster, a COM+ application cluster and COM+ routing cluster. We focused primarily on the Web cluster for our tests. After choosing and naming the cluster type, you can specify the form of load balancing. The product primarily relies on Network Load Balancing using two network interface cards in each server in the cluster. However, it lets you use a third-party hardware load balancer. Of course, if you had such a product, the value of Application Center 2000 diminishes, but some feeatures (particularly content replication and server health monitoring) might be useful. We focused solely on the standard Network Load Balancing without such outside assistance.

Application Center 2000 uses a basic round-robin style of load balancing. To adjust this, you can modify a slider indicating the amount of traffic a server should handle. However, adjustments beyond this gross distribution -- such as URL-based, content-based or any form of global load balancing strategies -- are not directly supported. Clever administrators could figure out how a couple of clusters could be built (via redirection rules for Internet Information Server [IIS]) and implement a URL-based form of load balancing, but in short the product only provides the most basic form of load balancing.

We liked one feature of Application Center 2000 that puts it ahead of overly simplistic schemes such as DNS round-robin. The feature applies to state management issues using an idea called request forwarding. Basically, if a state has to be preserved, the initial server visited becomes the "sticky" server and subsequent requests are forwarded to the server holding the user's state information by any other server in the cluster that receives another request from the end user. Of course, if the sticky server crashes, problems will ensue. In contrast, centralized state management -- either via hardware or software -- would alleviate this problem, but at the expense of providing a single point of failure for the whole cluster.

Join us:
Facebook

Twitter

Pinterest

Tumblr

LinkedIn

Google+

Answers - Powered by ITworld

ITworld Answers helps you solve problems and share expertise. Ask a question or take a crack at answering the new questions below.

Join us:
Facebook

Twitter

Pinterest

Tumblr

LinkedIn

Google+

Ask a Question
randomness