Further to my trip down to CommunicAsia 2007, this is the first of several reports I am working on pertaining to products that caught my attention. I have learned much, and I certainly hope you will be able to gain something from my write-ups as well.
One of the booths that I stopped by was Crescendo Networks. They sell networking products, though networking solutions might have been a more appropriate term, given the specialty of their products. The niche that Crecendo’s flagship Maestro product addresses lies in the “still-being-defined” product category of network and application acceleration.
Now, the traditional approach when the number of visitors exceeds one’s Web server capacity is to simply increase the number of servers. Extremely popular sites, such as Slashdot, runs off a pool of Web servers. In such a situation, there is a need to “distribute” visitors among the various servers using a technology termed “server-load-balancing,” or SLB.
The mechanics for the various SLB techniques is an entire subject by itself, and there are a number of good books you can read on this relatively esoteric topic. Google, for example, is said to have hundreds of thousands of servers in its Web farm. Suffice it to say that with a properly implemented SLB, the number of visitors that a single site can serve is hence increased exponentially.
Yet, as the number of Web sites requiring SLB becomes more common, a new paradigm is starting to emerge. This is fueled by the cost savings inherent to the question: if one can reduce the 20 Web servers currently running by half (to just 10), the monthly datacenter (and potential bandwidth) savings can be phenomenal.
This is where Crecendo comes in. Crecendo’s Maestro Product range is able to address both what it calls the Application Layer Processing (ALP) and the Application Front End (AFE). ALP works directly on the application-facing end to enable less application hardware to do more. AFE, on the other hand, deals primarily with the front-end Web clients, off-loading most of the overheads involved with dealing with Web clients so that lesser Web servers is needed.
What I found fascinating was that they have architected the entire system using custom-engineered ASIC, squeezing everything into a single 2U device that scales up to handle one million simultaneous total connections.
A whole slew of techniques is incorporated into a Maestro. It is broadly able to do on-the-fly content compression, off-loading of the resource-intensive TCP connection creation and tear-down, multiplexing, and even SSL off-loading. And because the features are incorporated into different ASIC chips, you can define your own set of features and pay for only what you need.
Implementing the Maestro can be as simple as putting your Maestro between your Web server and your firewall. If all the options are purchased, you can then proceed to switch off all the processor-intensive features from your Web server, letting the Maestro do the grunt work instead. HTTP compression and SSL can then be disabled, freeing precious processor cycles to handle more Web connections.
Benefits will definitely vary depending on actual site and features selected of course.
Mr. Or Yaacov, Manager, Customer Support
I spoke to Mr. Or Yaacov, the manager of customer support whom gamely fielded all my questions. In response to my queries about real world performance, he told me that he has personally deployed the Maestro (one unit!) onto a social-networking site with up to 70 million page views per day. Serving the same number of users, the site saw an immediate bandwidth reduction of 25% in outgoing traffic!
In case you're saving up to get one for your Web site for your next slashdotting, the cost of each beau lies between USD $20,000 and $50,000.