by Lenny Liebmann

About Lenny Liebmann


Previous Columns

MSPs make sense...probably
June 2000

DSL-to-frame:
an object lesson in industry economics

May 2000

The W2K Problem
April 2000

Keeping an eye on IM
March 2000

Load balancers ascendant
February 2000

A world of trade-offs
January 2000

 

Load balancers ascendant

Resource management, not network QoS, provides the key to sustaining the quality of e-commerce services.

It’s amazing how quickly the technological landscape changes in this industry. Not that long ago, networkers were deeply engaged in a debate over how to deliver end-to-end quality of service (QoS) over IP networks. QoS, everyone agreed, was the key to successful e-business. After all, if we couldn’t prioritize mission-critical application traffic over lowly game-playing consumers, what good was the Internet?

Well, there is still no standardized mechanism for QoS over the Net—although the process of hammering out such a mechanism continues. But the amount of business being transacted on the Web is exploding anyway. Those lowly game-playing and site-browsing consumers have ignited the hot, hot market for broadband access. And e-commerce service providers aren’t wasting much time talking about network QoS anymore.

Instead, they’ve shifted their attention to the far more critical issues of back-end resource management—which is very good news for the load-balancer (LB) market.

You see, it turns out that the worst bottlenecks in your typical on-line interaction have nothing to do with the Net. They have to do with the site’s back-end infrastructure. Database connections get saturated. Content servers get overloaded. People click on a link, and nothing happens.

And who has stepped up to the plate to help address all these issues? The load-balancer guys.

HYPER-EVOLUTION

What’s really amazing is how quickly this little market segment has evolved. Eighteen months ago, when everyone was furrowing their brows over QoS, load balancers were dumb little devices that simply let you spread HTTP requests across multiple Web servers. Not really the kind of world-shattering functionality that brings the venture money swarming in, wouldja think?

But, as sites got more complex and multitiered, the LB guys realized they had to do a better job of assessing the health of all the processes in the service-delivery chain. So, they began engineering monitoring tools and load redirection algorithms that let site managers get the most out of their back-end systems’ resources.

This traffic redirection led to another set of problems. As people were getting bounced around, their shopping carts were being lost. Encryption servers were being forced to crank up new sessions for the same visitor over and over. LB vendors responded with workarounds that helped to maintain a customer’s “state” throughout any given site visit.

Instead of just redirecting traffic based on the health of back-end resources, LBs were also now examining incoming packets to determine the identity of the visitor. This was done through IP addresses, HTTP “cookies,” and other mechanisms. Once you’re looking at users’ cookies, you can start doing all kinds of neat things. You can put someone with a shopping cart filled with $1,000 worth of stuff on a faster server than someone with one $9.95 item. You can make sure that loyal customers get priority over first-time visitors. You can give someone with a DSL connection access to a faster streaming-media feed than someone with a 28.8-kbps modem.

Hmmm. Sounds like QoS to me. In fact, it sounds more sophisticated than network-based QoS could ever be—because, instead of a few arbitrary prioritization levels—this new load-balancing paradigm lets you apply real business rules to your real resource mix as it exists at any given moment.

It hardly makes sense to call it “load balancing” anymore. Mark Hoover of Acuitive, Inc., a market research firm that specializes in this area, prefers the term “virtual resource management.” I think he’s right. The LB vendors have come very far, very fast—and their old moniker no longer does them justice.

KEEPING YOUR EARS AND MIND OPEN

There are many lessons to be learned from the rapid rise of load balancers as high-value e-commerce infrastructure solutions. For vendors, the LB story is a great example of why you should listen to your customers. Cisco, for example, had an early lead with Local Director and has now been left in the dust when it comes to features and functionality. Of course, Cisco can just buy one of these upstarts when it wants to, but that’s a bit more expensive than doing the market research and product development yourself.

For Communications News readers, there’s a more subtle lesson. When you’re in the networking business, you tend to want to solve every problem with a networking solution. That’s what QoS is about. And, in these heady early days of the Internet, it’s easy for us networkers to believe that the world is inexorably heading our way.

It isn’t. E-commerce isn’t about networks. It’s not about systems. It’s not even about applications. It’s about the end-to-end delivery of services over a hybrid infrastructure that includes networks, systems, applications, virtual business partners, thin air, human beings, and—yes—load balancers. Load balancers happen to occupy a key intersection point between the public network and back-end resources, so they’re turning out to be very crucial devices. Don’t worry, though. Things will change. Will you be ready?