The arithmetic of ADCs

Wide Area Networking Alert By Jim Metzler and Steve Taylor, Network World
March 15, 2010 12:02 AM ET

Insightful analysis by consultants Steve Taylor and Jim Metzler, plus links to the latest WAN news headlines

Previous newsletters have discussed the role of application delivery controllers (ADC) as well as the key functionality they provide. This newsletter will quantify the difference between the value of a server load balancer (SLB) and the value of an ADC.

The ABCs of ADCs

The deployment of an SLB enables an IT organization to get a linear benefit out of its servers. That means if an IT organization has implemented an SLB doubles the number of servers supported by that SLB, that it should be able to roughly double the number of transactions that it supports.

Unfortunately, the traffic at most Web sites is not growing at a linear rate, but at an exponential rate. To exemplify the type of problem this creates, assume that the traffic at a hypothetical company's (Acme) Web site doubles every year. If Acme's IT organization has deployed a linear solution, such as an SLB, after three years it will have to deploy eight times as many servers as it originally had in order to support the increased traffic.

Related Content

As mentioned in previous newsletters, an ADC offloads computationally-intensive, communications processing off of the servers and hence significantly increases the overall ability of a server to process transactions. If Acme's IT organization were to deploy an effective ADC then after three years it would still have to increase the number of servers it supports, but only by a factor of two or three – not a factor of eight.

The phrase effective ADC refers to the ability of an ADC to have all features turned on and still support the peak traffic load. Referring back to the Acme example, it was the ability of Acme's ADC to offload numerous computationally-intensive tasks from the servers and still process the peak traffic load that enabled Acme to not have to deploy eight times as many servers in three years in order to support the increase in the peak traffic load.

If an ADC cannot support the peak traffic load with all features turned on, in addition to not maximizing the performance of the servers, the ADC becomes a bottleneck. While it is always important to try to avoid having bottlenecks, that is particularly true in the data center as a bottleneck here negatively impacts the experience that a large number of users have with multiple business critical applications. As part of the evaluation process, IT organizations need to ensure that they will not deploy an ADC that quickly becomes a bottleneck. To do this, IT organizations should stress test the ADCs that they are evaluating at traffic loads that are well beyond their current traffic peak traffic loads.

While it is critical that an ADC can support the peak traffic load with all features turned on, this is not an easy task to accomplish. The key-determining factor relative to the ability of an ADC to support peak load with all features turned on is the internal architecture of the ADC. In particular, the ADC must perform the most computationally-intensive tasks, such as TCP offload, in hardware that is itself purpose-built for the task.

Steve Taylor is president of Distributed Networking Associates and publisher/editor-in-chief of Webtorials. Jim Metzler is vice president of Ashton, Metzler & Associates.