Meet the new enterprise network
Next-generation network gear aims for a bigger slice of the processing pie
- By Paul Korzeniowski
- May 08, 2006
When will a network switch no longer be just a switch? The answer is now. Network equipment vendors have been developing a new generation of products that do more than mindlessly move bits and bytes from place to place. Instead, these devices can scrutinize data and its destination, then perform a variety of functions, such as checking security and giving data a makeover for a more efficient delivery.
“In trying to improve performance, users are no longer focused on bandwidth but instead are looking at issues such as network latency, security and application performance,” said Joe Skorupa, a research director at Gartner.
Those new user priorities are prompting vendors to recast the role of networks by adding tasks traditionally done by application software and servers. Making networks more aware of the types of data and applications running across them will enable them to handle more of the low-level communications and protocol-manipulation grunt work. That should improve application response times, ease overworked servers and lower overall network bandwidth needs. The expected payoff is more productive users and lower information technology bills.
However, this new age of more intelligent networks is still in its infancy. A crowd of start-ups and industry veterans touts a variety of approaches, making agency purchase decisions tricky. “Because the market is just emerging, no one is really sure how it will eventually shake out,” said Joel Conover, principal analyst for enterprise architecture at Current Analysis.
But everyone agrees that a few factors are driving these changes in network equipment design. The most significant is a shift in customer thinking. The traditional way to solve network problems was to add more bandwidth, but that fix is becoming a rather blunt and less cost-effective instrument given recent trends in enterprise computing.
During the past few years, government agencies centralized and consolidated their servers and applications with the goal of reducing costs and system management work. Those changes created a ripple effect.
Many of the newly centralized applications weren’t designed for environments in which servers were far from users’ computers. Those programs typically rely on protocols such as Messaging Application Programming Interface, which tends to push a lot of traffic onto the network because it assumes inexpensive local-area networks link clients and servers. When agencies relocate servers in a consolidation project, network performance can quickly degenerate.
Another factor stems from changes in application design. “Organizations have been moving away from monolithic to distributed applications that share a variety of services,” said Bill Ruh, vice president of technology at Cisco Systems. Coordinating services, such as unwrapping Extensible Markup Language messages, requires more processing power than traditional application functions.
The growing acceptance of service-oriented architecture and Web services applications has made information’s movement from place to place more flexible than it had been in the past.
“Because these applications are new, [enterprises] don’t yet understand how they will perform in production environments,” said Mike Banic, a director of product marketing at Juniper Networks. “They may expect a Web services application to use about 25 percent of the available resources, but once it is deployed, it quickly requires 100 percent of the resources.”
Network latency has also become a bigger problem. With more distributed applications, the number of hops among network devices increases. For example, a government worker might use an application that collects information from a central server in Washington, D.C., adds other data in Virginia and then completes the transaction with information housed in San Diego. Making all of those connections may take only a few milliseconds, but the cumulative impact of traversing so many systems can cripple time-sensitive applications, such as video.
So far, the public-sector outfits that have ventured into the application acceleration market have done so to solve specific performance problems, not to reinvent their entire network infrastructure.
For example, at the University of Maryland University College (UMUC), changing application requirements and rapid enrollment growth created performance challenges as the school sought to serve its 95,000 online students. In summer 2004, the university opted for Juniper’s DX3250 application accelerator to improve response times for the school’s enterprise resource planning and Web-based learning management applications.
“Almost immediately, we saw performance gains of 25 percent or more,” said Gregg Chottiner, UMUC’s vice president of IT.
Wide-area network bandwidth constraints prompted the Pierce County Library in Tacoma, Wash., to examine application accelerators. The library was struggling to provide adequate response time for applications that served its 17 locations.
“We needed a solution that did not require upgrading our WAN links,” said Dan Fineman, IT systems administrator at the library. Instead, the library selected Riverbed Technology’s Steelhead system in summer 2004 and increased its network bandwidth utilization by more than 50 percent.
Such decisions are expected to become more common. Pricing for stand-alone application accelerators starts at about $5,000 — two are necessary for each connection — although they can rise to $50,000 depending on the speed and type of applications supported.
“For a lot of organizations, the accelerators represent a much cheaper solution than upgrading their WANs,” Conover said.
As workload share between network and applications shifts, agencies may find that they need to make staff adjustments. “In many organizations, the application and networking folks talk different languages, but now they need to speak with the same voice,” Ruh said. Because of the differences, turf wars may break out. For example, rather than weighing one vendor’s switch vs. another’s, purchasing decisions might hinge on whether to add an intelligent network device or another server.
Organizations that need to enhance communications between their network and data center teams have time. The market for the new devices is so small that many market research firms have not even quantified it yet. They expect the market to grow as enterprises install more Web services applications.
“The novelty effect now present with application accelerators will disappear in the next few years as more and more organizations deploy them,” Conover said.
Korzeniowski is a freelance writer in Sudbury, Mass., who specializes in technology issues. His e-mail address is firstname.lastname@example.org.