Second Generation Load Balancers
The rapid growth and evolution of the Internet are changing the requirements for load-balancing technology. A few short years ago, content providers grappled with the problem of putting together a technology mix that would guarantee the highest degree of access to their sites. By combining small clustered computers -- on which content was identically replicated -- and adding first-generation load-balancing technology, they assembled a tandem configuration that evenly distributed access requests across the content farms. In the era when traffic counts and visitor loyalty were the measures of success, this winning combination more than met the needs of the most advanced and sophisticated content providers.
The Web and Internet market, however, is now moving into its next phase of development, thereby changing the metrics by which success is measured. With the advent of electronic commerce, a failed visitation now amounts to a missed commercial transaction opportunity that can be measured in terms of financial loss. Simply put, whenever a Web surfer cannot get to an electronic commerce site, a customer -- and his or her money -- is lost forever. This new metric for success is the driving force behind second-generation load-balancing technology. Some features of this new generation include wide-area load balancing, fault tolerance and heterogeneous system support.
Today's advanced -- and tomorrow's typical -- Internet site will likely be distributed over a wide geographic area. By adopting this approach, companies will enjoy the benefit of having one www.company.com identity, but will be able to satisfy access requests from many different geographic sites. In effect, a company will present a single Internet address to the world, but will satisfy Internet, intranet and extranet accesses from locally, regionally, nationally and even globally distributed sites.
Wide-area load balancers have the ability to determine the current load on the content servers at each geographic site and, on the basis of this determination, transparently shuttle Web, database or IP requests to the least-burdened site. Within the site, traditional first-generation load-balancing technology takes over, passing the access request off to the least-burdened server at the site.
Because of this two-stage process, wide-area load balancers fulfill access requests in the least amount of time, first by sending the access request to the least-burdened geographic site and then by forwarding it to the least-burdened server within the site.
Anyone who has faced a multimegabyte file download -- but first checked their watch to find a download site on a continent where everyone is asleep -- will appreciate the benefits of wide-area load balancing. With second-generation load balancers in place, this act of exploiting time zone differences, and its impact on content server load, is done automatically
Wide-area load balancers also have fault-tolerance implications. If a geographic site goes out of service, a second-generation load balancer detects this failure and logs the site as being out of service. By doing so, it no longer sends access requests to this particular site. If and when the geographic site comes back online, the wide-area load balancer detects this change and once again includes it in its complement of sites to which it sends access requests. Fault tolerance is, in effect, distributed over a wide geographic area.
Second-generation load balancers also support failover redundancy within a site. Administrators can set up configurations in which dual load-balancing units work in a failover relationship. If the primary unit should fail, the backup unit takes over the load-balancing chores. This failover process can take place in milliseconds. When the primary unit is repaired and once again online, the secondary unit automatically goes back into standby mode.
Heterogeneous system support is also an important feature of second-generation load balancers. Companies will not accept the idea that they must standardize on one computing platform and operating system in order to get second-generation load-balancing capability. Companies want the flexibility to mix and match systems of differing capacities into one composite whole that can service all access requests. This occurs, for example, when a company has a stock of powerful UNIX systems that it wants to combine with smaller NT systems under the control of a load-balancing system. Second-generation load balancers support such heterogeneous configurations.
The more advanced load-balancing companies are already moving forward to fill the need for second-generation products. F5 Labs (Seattle, www.f5.com), for example, has put together a product combination that delivers many of the features described in this column. Its 3DNS product, combined with its BIG/ip family of load balancers, delivers wide-area support, fault tolerance and heterogeneous systems support. IBM Corp. is another company with advanced, second-generation features. Its eNetwork Dispatcher 2.0 delivers wide-area load balancing support. As the Web and Internet market continues to move forward, more load-balancing companies will release products with second-generation features.
Sam Alunni is vice president of networking at Sterling Research (Sterling, Mass.). Contact him at firstname.lastname@example.org.