On this blog I have primarily been writing about applications that are very visible to web users. With my next two posts, I will look a bit behind the scenes. Most web sites are supported by multiple servers, even small firms quickly graduate beyond a single server as traffic builds. It is important that server loads are distributed in a balanced way and that failed servers are detected and traffic routed to others in the network. Optimizing load balancing has usually meant an expensive hardware purchase with associated implementation and administration costs, limiting access to these resources to large enterprises with big IT.
I spoke recently with Chris Cook, the Director of sales and Marketing at TZO, a Boston-based firm that is offering load balancing as an on-demand software service at dramatically reduced price points. Like many Saas offerings it offers a reduction in ongoing costs, eliminates large upfront investments, and provides the flexibility to opt in or out as business conditions change. The firm started as a web domain name service and added AutoFailover in 2001. With AutoFailover, when TZO detects a server failure it automatically updates the DNS record for a domain so that the server requests are sent to the IP address of alternate servers or server clusters. The key being very short propagation times eliminating any realized down time.
Then in 2001, TZO added server load balancing as an on-demand service. With load balancing, you can specify that each server handles a certain percentage of server requests, routing more requests to your more robust servers. Servers may be in geographically dispersed areas so you can protect against regional disruptions yet continue to balance the load among the server locations.
Now they have just added geographic load balancing. In the past, a mathematical balance was achieved regardless of location. Now you can direct incoming traffic based on a location basis, as well as a load basis. This might allow for location based flash ads and potential reduction in traffic costs. Chris did a demo for me. I watched him set the desired service loads for servers in three locations. Then reports showed an imbalance in traffic handling. So we adjusted the loads and the reports showed a balance. It was easy enough for me to do it, always the acid test.
One investment fund brokerage house with locations in the US and UK was having a debate over which hardware solution to use for their web server load balancing. They decided to use the TZO service while they were sorting out the hardware alternatives. The TZO service was about a thousand a month. The hardware alternatives were close to a hundred thousand. To no surprise, they give up on the hardware idea and continue to use the TZO service. As the recent WebEx book covering on-demand services asked in its title, Why Buy the Cow?
One of the components of the new web is the increased use of on-demand service across a wide variety of applications. This is a great example, In my next post I will cover an on-demand, online backup data service designing for mobile computing.