Continental Airlines has high-flying plans to expand its e-business infrastructure to Europe and Asia this year, and global load balancers are playing a pivotal role.
In fact, says Andre Gold, director of Internet engineering at Continental, "If the [global] load-balancing technology fails, we fail. Our entire e-business strategy is based on this technology."
Global load balancers use a number of different techniques to direct Web traffic to the most appropriate server farm or data center. Unlike server load balancers, which manage traffic within a company, global load balancers manage traffic between geographically distributed sites.
The Houston airline uses two LinkProof traffic management appliances from Radware Ltd. to juggle traffic between its two ISPs D WorldCom Inc. and Savvis Communications Corp. The ISPs support the airline's busiest Web sites: those that provide frequent flyer information and handle online booking. The global load balancers monitor the performance of both ISPs' networks and shuttle Web traffic from Continental data centers back to the customer through the fastest route.
This improves the efficiency of Continental's e-commerce site, which means customers can obtain information or book a ticket faster than before. Gold estimates that the performance improvements have let Continental shave US$20 off the price of a ticket.
This year's e-commerce data center expansion will hinge, in large part, on an upgrade to Radware's Web Server Director-Network Proximity appliances, which will give the airline the ability to redirect customers to the "closest" server using a customizable algorithm.
Global load balancers don't solve all problemsIf you need to deliver cacheable data more efficiently and cost-effectively to customers, then a content delivery network (CDN) might be your best bet. A CDN will reduce the amount of traffic hitting a server by moving frequently requested content to the network's edge, says Zeus Kerravala, an analyst with the Yankee Group.
But if you need to handle frequent requests for transaction applications, it's more useful to balance and direct traffic between data centers, because that type of data is not easily cached.
Cendent Mortgage in Mt. Laurel, N.J., has 100 different Web sites across its divisions that are used by individuals seeking mortgage and financial information. Jim New, an advanced technical services specialist at Cendent, says even though the price tag on Cendent's load balancing rollout is about $240,000, the cost is well worth it.
"We can handle about 10 percent more requests than we otherwise might," New says. "We can also take down servers if necessary to update them without affecting availability."
The company has two global load-balancing appliances from F5 Networks, which both sit outside the firewall. The appliances translate external IP addresses to an F5 Big-IP controller, which routes IP traffic to the best server. Cendent has data centers in Mt. Laurel, Moorestown and Garden City, N.J.
It's not always easy to nail down specific monetary figures to show where global load balancing delivers value. But many corporations are happy knowing they have more control over what's already in their network.
The Web servers at Landstar Systems, a transportation services company, are the most important means of communicating with more than 10,000 employees who work for about 1,000 Landstar agents. These are owners, operators and contractors who log on to Landstar's site each day to conduct business.
The company distributes network traffic to 15 servers in data centers in Jacksonville, Fla., and Rockville, Ill., using two Radware switches -- a WSD-Distributed Solution to ship traffic to either data center, and a WSD NP, to direct traffic to the closest client.
Patrick Wise, vice president of e-commerce at Landstar in Jacksonville, says he has no statistics to prove he is saving money, but by redirecting traffic to an underused Web server, he avoids having to add an extra server.
More important to Landstar is that it can remove a failed server from a cluster without customers ever knowing. Another benefit is it lets Landstar see potential bottlenecks. "We can manage our site better and control our load better," Wise says.
Global load balance bits
-- Aside from F5 and Radware, other players are Cisco, through its purchase of ArrowPoint Communications; Nortel, through its acquisition of Alteon WebSystems; and Foundry Networks.
-- Vendors differ on how they measure the capacity of global load balancers. Some use throughput, others talk about connections per second. The important thing is having a device that can scale up to handle peak numbers of requests coming in to the data centers, Yankee's Kerravala says.
-- Global load-balancing appliances generally sit outside a network and catch requests before they hit the firewall, although there is flexibility in how they can be installed, and in some cases they can sit between the firewall and the servers.
-- Most vendors sell more than one box to handle the actual load balancing and traffic distribution, although there are products that combine the functions.
It's not so important how a company chooses to handle the task as long as the ultimate goal is achieved. "It's more important that Web transactions are happening as quickly as possible," Kerravala says.
Semilof is a freelancer writer living in Watertown, Mass. She can be reached at firstname.lastname@example.org.
2001-2002- The number of organizations adding multiple data centers is increasing and by 2003, about 70 percent to 75 percent of mid- to top-tier applications will be distributed across at least two data centers.
2001-2004- Prices for load-balancing products will gradually decline 10 percent to 15 percent annually.
2002-2003 - Global load balancers and local server load-balancing products will incorporate wireless protocols as growth of all wireless access devices accelerates.
Source: Meta Group
How global load balancers work
Global load balancers use proprietary protocols to collect information about the health of a site or to derive basic location information, says Peter Firstbrook, an analyst at Meta Group Inc.
Although there is some support for exchanging metrics between global load-balancing gear using XML, Firstbrook says communications between devices are not expected for a few years, so customers are better off sticking with one vendor for traffic distribution.
Today's global load balancers divert traffic using one or more of several techniques:
-- Most common for Internet applications is DNS redirection, in which the global load balancer becomes the authoritative name server. In this scenario, client URL queries traverse Internet DNSs until the IP address of the global load balancer is returned. The appliance then gives the client the IP address of the best data center.
One disadvantage of DNS redirection is that it is time-sensitive. The appliance might not have the most up-to-date information on the status of each server or database. But DNS redirection is relatively fast and easy to deploy and requires less system intelligence than other methods.
-- DNS redirection is often used in conjunction with HTTP redirection. This method sends traffic to the most available site based on HTTP header information. It's also considered quick to implement, but the downside is that it's only good for HTTP traffic, not for FTP streaming, for example.
-- Triangulation is another DNS method that works when the client sends out a request, which is shipped to multiple sites that contain the requested content. The best available server sends back data the fastest. Radware uses a triangulation method where the load balancer directs traffic to the least loaded site, while masking that address with the address of the redirecting site.
Near-term advances will focus on improving each device's overall ability to check the health of an application.
F5 Networks plans to enrich its ability to control traffic so that users create localized custom topologies. IT managers can route traffic on a country level, but soon will be able to drill down on a more specific geographical level, says Jason Needham, a product manager at F5.
The company also will improve its "persistence" features, which control the ability to make sure users who input data on one screen can be routed back to those locations. This is particularly important in a financial application, when a customer fills out an online credit card profile, for example. The ability to synchronize that data with other data centers in real time is something that F5 will address this quarter.
Cisco also will improve the geographic accuracy of where content is best reached for fastest response. In particular, companies will be able to update a database to identify content that is cached on a particular local device and not just at an origin server.
But the next challenge will occur when corporations want to conduct fulfillment activities out of multiple servers across more than one data center. That will make it necessary to synchronize databases across the company, says Thomas Nolle of the CIMI Group and a Network World columnist.
"Most applications today are designed to load balance a display-only Web server, rather than really support load balanced e-commerce where there is a transaction and not just a cataloged Web page," Nolle says. "But ultimately the value of an electronic catalog is limited if you cannot execute off of it."
There is a lot to distributed load balancing that has nothing to do with networks and everything to do with the overall framework of the applications, Nolle says.