Startup Finds New Ways to Free Up Application Bottlenecks

Crescendo Networks says its new ALP software takes app accelerators to the next level by speeding up the processing side of the equation.

To date application accelerators have addressed performance bottlenecks in front of the Web server. Startup Crescendo Networks on May 1 at Interop in Las Vegas introduced new technology that deals with the bottlenecks behind the Web server.

The company believes it is taking application accelerators, or Application Front Ends, to the next level by making it possible to speed the processing side of the equation.

Crescendos new ALP (application-layer processing) software "allows us to look into the application flow—the HTTP or request flow—and optimize and accelerate throughout all the tiers of the application," said Mike Long, vice president of marketing, at the companys U.S. headquarters in Tenafly, N.J.

ALP addresses the Web, application and database tiers of multitiered Web applications, optimizing the request flow from users all the way into the back end.

For one existing Crescendo user, the new focus on the performance bottlenecks in the application infrastructure comes at the right time, according to Ian Rae, founder and CEO of Syn10ic, an e-commerce performance consultant and managed services provider. "This is the next area that needs to be addressed. Until Crescendo told me they were working on this, I was trying to figure out how to solve these problems," said Rae, in Montreal. "I knew it was exactly what we needed to provide very big improvements for some of our customers application performance problems," he added.

ALP is implemented as a software module for Crescendo Networks Maestro application acceleration appliances. It is based on patent-pending application-layer processing algorithms that use Crescendos rules engine to classify any request that comes through.

It includes an admission control capability that can learn the load generated by each request on a per-tier basis and can queue requests once the application has reached its capacity.

/zimages/5/28571.gifThe application acceleration market hit $1.2 billion in 2005. Click here to read more.

"ALP is built on the fact that any two requests can have significantly different processing weights. Something can take three seconds versus 20 milliseconds for something else," said Hooman Beheshti, vice president of technology at Crescendo.

Admission control can be manually configured to classify which requests are heavy and which are light, or it can watch and automatically learn and classify.

In addition, admission control never lets a tier reach its maximum limit in the number of concurrent requests it can handle by putting subsequent requests in a queue. In that function, a mechanism called request scheduling controls which requests are sent for processing.

Request scheduling can be used to put lighter requests first in the queue and heavier ones in the back, rather than processing each one as it comes in. "Some requests take 15 seconds and then youll have a bunch of 10-millisecond requests, so we put those first," said Beheshti. "There is a penalty the heavy request has to pay, but weve added an aging mechanism so that we prevent that request from staying in the queue indefinitely. We make sure nobody gets starved," he added.

As a result of its flow-control mechanism, ALP also offers visibility into the application flow for monitoring and reporting. It allows administrators to watch on a per-tier basis how long each takes for each set of requests to process server time, client time and end-to-end time.

In theory the techniques have a lot of promise, but with years of data center experience under his belt, Rae is waiting eagerly to test out their application. "Ill believe the performance improvements when I see it contribute to my customers applications. Ive not been able to quantify the improvement, but Im looking forward to doing that," he said.

Crescendo believes it is uniquely positioned to provide such optimization techniques because of the hardware architecture it uses in its application accelerator. The purpose-built appliance uses field-programmable gate arrays with dedicated memory to execute each feature. No one feature relies on the performance of another, so that turning on one function such as TCP termination wont affect the performance of others. "It requires a lot of horsepower to tear [an application request] apart, make decisions about it and then do something," said Long.

ALP, implemented as a licensable software module for the Maestro application accelerator, is due in the third quarter from Crescendo Networks, which is based in Tel Aviv, Israel.

/zimages/5/28571.gifCheck out eWEEK.coms for the latest news, views and analysis on servers, switches and networking protocols for the enterprise and small businesses.