Coping with bandwidth hogs

Network managers are finding new ways to curb their networks' bandwidth thirst.

Let's admit it - we were all made a little drunk by the booming economy. When applications with a deep thirst for bandwidth bellied up to the network bar, the call was to just buy more bandwidth. Cheap bandwidth gushed from the taps as though the party would never stop.

"Bandwidth hasn't been so expensive that you couldn't afford to just get more to stay ahead of demand," says Kim Ross, CIO at Nielsen Media Research in Dunedin, Fla.

"The real driver here is money," agrees Corey Ferengul, an analyst at Meta Group Inc. in Stamford, Conn. "The solution used to be JBM: Just Buy More. Today, I don't know any company that has excess money to buy bandwidth."

In any case, buying more bandwidth may not be the solution, Ross says.

"In the old days, you had one main application, one pipe, one concern," he says. "But with Internet applications, more resources are shared. If you add more capacity to a shared pool without a clear idea of why or where it's needed, you're unlikely to get the result you want. And the tools just haven't been there to help you manage IP networks, to isolate problem cases."

That has left enterprise network managers seeking new ways to slake their networks' bandwidth thirst.

1. Monitoring and Enforcing

"The way we deal with it is we watch bandwidth utilization day by day, week by week and watch for spikes in utilization," Ross says. "Usually it's someone trying something new, but nine times out of 10, it's not something that's business-critical. We'll do the detective work to find out where it's coming from, but it takes way more effort than it should."

Universal Health Services Inc. in King of Prussia, Pa., has 65 remote sites on its wide-area network. "It's sometimes tough to tell what or where the problem is," says CIO Linda L.E. Reino. "We've put a sniffer on connections, and it's ID'd some rogue devices on the LAN, polling every few minutes and using a lot of bandwidth."

But monitoring isn't foolproof, she says. During a Nimda virus scare, a virus checker on Universal Health's LAN was examining every packet of HTML traffic. "This thing made the network crawl. Our clinical order management system is HTML-based," Reino says.

Online gaming and peer-to-peer networking services that allow downloading of audio files, software and movies, such as those offered at the Morpheus, Kazaa and Grokster Web sites, can quickly gobble up bandwidth.

"We have had a Morpheus problem, and we handled it by blocking it at the firewall so it can't connect," says David Leuck, technical services manager for the Culver City, Calif., municipal government.

Part of tuning the network at investment management services firm Ark Asset Management Co. in New York was "toughening up on users," says Danny Shpak, manager of Ark's IT group. "That means no gaming, for one thing."

Ross adds, "One aspect of our control policy is to be sure the IT governance body knows the guidelines. Clear, well-understood policies can help you prevent some problems from even coming up."

2. Streamlining Networks

Ark uses Eye of the Storm from New York-based Entuity Ltd. to monitor its network and locate and ease the pain points. Streaming stock quotes constitute the single greatest use of bandwidth, says Shpak, but "no one thing really put us over the top."

Monitoring helped him identify bottlenecks such as those resulting from misconfigured connections between clients and servers. Sometimes the fix was as simple as moving an application server closer to those workers who used it most. In other cases, he had to invest in larger routers or switches or more bandwidth.

"Tuning was most important," Shpak says. "I'm interested in solving problems, not measuring bandwidth. For me, the more important issue is to have my network run cleanly."

A well-tuned network gives him another edge, he adds. "When application volume increases and you're running out of bandwidth, you have the data to present to management to get more bandwidth," he says.

3. Centralizing Operations

Centralizing some network functions can trim bandwidth use while addressing some security goals, says Tom Revak, domain architect at GlaxoSmithKline PLC, a pharmaceutical company based in Research Triangle Park, N.C.

"If you centralize data center functions, you cut down on the amount of replication and synchronization of data," Revak says. "It also helps maintain data integrity, if only because there aren't so many copies to synchronize and replicate."

4. Testing and Tuning Applications

Test applications before installing them enterprisewide to make sure they aren't going to swamp your network. Then tune existing applications for the same reason.

Atlanta-based United Parcel Service Inc. does "extensive studies of the behavior of applications," says Mark Morelli, telecommunications director for the global shipping company. UPS not only tests but also assiduously monitors applications' bandwidth use, he says.

"We have to; a lot of our sites only have 56K connections," Morelli says.

"Many of the Web-based HTML applications have become real bandwidth hogs," he adds. Those that demand too much bandwidth are simply banned from the network.

"We're waiting for the day when true class-of-service management can be done in the network," Morelli says. "Meanwhile, the business determines the importance of applications. Based on those priorities, we use throttling tools to carve out specified amounts of bandwidth that the applications can use."

Ferengul says that other companies are taking the same approach. "We're seeing a lot more interest in traffic-shaping, because people aren't as interested in buying more of anything," including bandwidth, he says.

"There are plenty of good tools that can help you optimize your network," says Dennis Drogseth, an analyst at Boulder, Colo.-based Enterprise Management Associates Inc. "Which one is best for you depends partly on your ability to make the most of what you buy."

Pat Miller, telecommunications manager at Baltimore-based mobile and modular building supplier Williams Scotsman Inc., bought Austin, Texas-based NetQoS Inc.'s SuperAgent application response-time analyzer to identify the cause of slowdowns on her company's WAN. But she's also using the tool to examine the performance of thin-client application traffic.

"Very large packets and very small packets were fine," she says. "But certain medium-size packets were clogging the network." Application developers correlated the packets to the code that produced them and reworked the code.

Williams Scotsman's bandwidth experience isn't unique.

"One client was going to buy more bandwidth for an application," explains Ferengul. "But we looked at the code and found it did things like use six calls to a database just to do the log-in." Tuning the application obviated the need for more bandwidth.

A new pharmacy system went live at Universal Health Services last year, and "the hit on the bandwidth was obvious," Reino says. "We weren't meeting our service-level agreements. We got the vendor to accept packet accelerators on the application. No one wants to let you in their application, but sometimes you just have to do it."

5. Just Buy More

The days of simply overprovisioning are far from over, says Elisabeth Rainge, an analyst at IDC in Framingham, Mass. Sometimes there's no getting around the need for more bandwidth.

"When you've got some mission-critical application that isn't optimized for networks and you can't get the vendor to optimize it, the only way around it is to make your pipes so huge that they can handle it," Leuck says.

Lais is a Computerworld contributing writer in Takoma Park, Md.

Special Report

Taking Control

Stories in this report:


Copyright © 2002 IDG Communications, Inc.

Shop Tech Products at Amazon