by Chris Heyn

Murphy’s Law states that users of a web site service will receive the “server has timed out” message most often under the following conditions:

1. After spending 30 minutes or more loading a shopping cart with items you want to buy, you learn you need to start again.
   
2. You hit the “pay now” button and the screen goes blank before the “server has timed out” message appears. Have you been charged? Yes? No? Maybe?
   
3. When you need to recharge your prepaid mobile phone to make a vital call, the payment server goes down.
   
4. When you need to send an important report to your boss who does not like to be kept waiting.

As you bang your head on your keyboard in frustration, you will be able to take grim comfort in the fact that this situation is not caused by the web servers, or even necessarily by the applications that are being run on them. Rather, the network administration has failed and work will need to be done to improve network performance.

Who or what is responsible for the “server has timed out” message?
If the “server has timed out message” affects visitors to your site, then the damage to your image can be practically incalculable in terms of loss of reputation and business.

Smart network management teams will prevent this from happening by deploying load balancers to spread the load and ensure traffic is directed to servers that are not at risk of overloading. Smart network management teams who are hosting client services have realized that it is not enough to simply deploy a couple of load balancers in a single physical location. When hosting is your business, the “server has timed out” message is not an option.

A few years ago, I was talking to an ambitious company in Turkey who were launching a cloud-based, unified communications package that could be rented on a monthly basis by users. The concept and pricing were good; however the single data center location in Istanbul, in a location known for seismic activity, no doubt worried some potential subscribers who decided not to risk using the service. When I recently asked how things were going, the answer did not surprise me: “Ah, we have suspended the service.”

Another hosting company in Bari, Italy has taken a completely different approach, co-locating in the Netherlands, Italy and the U.K. and using Geographic Load Balancers to redirect traffic to a completely different server cluster should (and it has happened) one of their locations be attacked by denial-of-service (DoS) offensives.

How load balancers can make a major difference
Load balancing is critically important for heavy traffic web servers and web sites. Load balancing is also a great solution to keep cost control over the requirement to keep the web servers up and running. Overloaded and slow-running web servers will directly damage the business and associated profits. Well specified load balancers reduce the load on servers to achieve the highest level of efficiency with the lowest possible cost for your server requirements. Load balancers mean that servers that need to be taken out of service for essential maintenance to avoid future server failure can be taken out during working hours and save the staff cost for out-of-hours working. Below are the five key defenses that load balancers lend to protect users from the dreaded “server has timed out” message:

1. Incoming traffic is intercepted and shared among the servers to avoid overloading.
   
2. Any failed server will be isolated and bypassed by the load balancers, thus ensuring that web traffic continues to run smoothly.
   
3. Tasks are offloaded automatically, so users continue to navigate without being aware of the tasks that the load balancers are carrying out.
   
4. The load balancers treat the multiple servers as a single entity, which allows for greater productivity in regards to simultaneous tasks.
   
5. Response times are reduced, thanks to the redundancy load balancers can build in by duplicating critical functions.

What needs to be done?
Some load balancers are able to stop this problem that is called “drain” at both Layer 4 and Layer 7. What needs to be done is to disable Real Servers in a controlled manner. which will not result in the tearing down of established TCP/IP connections. With L7/application drain stopping, which means that the drain-stopping window can be lengthened beyond the TCP, it permits connected L7 sessions to stay active for a period that the load balancer administrator can define. This allows users the ability to have more time to complete their application activity before the real server is off-lined.

How to avoid the “server has timed out” problem most effectively?

1. Go to the global Real Servers panel in order to disable a server or an identified Virtual Server should you need to quarantine a specific problem.
   
2. Disable the appropriate Real Server(s).
   
3. Go to the statistics panel.
   
4. Use the “Virtual Services” or the “Real Server” button.
   
5. Monitor the connections’ real time counts until you are sure that you can safely offline the Real Server.

There you have it. By making load balancers a part of your server family, you will reduce the load on servers, thus providing a high level of efficiency and availability of application delivery and making the “server has timed out” notification a thing of the past.

Chris Heyn is the General Manager of KEMP Technologies Italy (Arcene, Italy) www.kemptechnologies.com