The oil and gas services company's bandwidth-analysis tools were showing HTTP traffic filling 30 percent to 40 percent of the WAN pipes there. And the RMON2 probes were indicating that some of Schlumberger's users in the Far East were hitting several non-business-related Web sites a little too regularly. It turns out these users were downloading massive picture files and music off the Internet, too. The company at the time was backhauling much of its Far East Internet traffic on its own international links (it now uses local ISPs there), so any errant Internet traffic meant trouble for the WAN.
Maged Elmenshawy, a network engineer for Schlumberger Network Solutions, based in Houston, pinpointed the problem during a routine check of the NetQoS NQReporter and NQAnalyzer tools; they showed irregular traffic flows in the Far East WAN links. Schlumberger Network Solutions, which runs Schlumberger's network and provides network services to outside companies, uses the tools to track trends in the performance of its own applications and WAN links. "The data is organized by what's consuming the most bandwidth," says Elmenshawy.
The digital picture and music incidents of a year and a half ago -- as well as some other bandwidth-usage spikes in the WAN -- fueled Schlumberger Network Solutions' plans to start "coloring" its traffic with class-based QoS (Quality of Service) priority labels so its business applications get priority over nonbusiness and other less crucial traffic. The company had used basic prioritization for the more congested links in its WAN, but now is evaluating and testing class-based QoS, which can provide more specific parameters for an application's claim to the pipe.
"Schlumberger has gone from throwing bandwidth at an underperforming application to the point where the network is now being prepared to be optimized for business applications," says Jeff Griesel, chief network architect for Schlumberger Network Solutions. "We want to ensure that applications play fair."
The RMON2 probes and NetQoS tools have been handy not only for identifying applications that need bandwidth adjustment, but also for tracking those applications' behavior over time, Griesel says. "It's one of the corner building blocks of the QoS strategy. The NetQoS tools tell us what an application really needs, and then monitor that application to see if the right parameters were set," he says.
Homegrown applications are infamous for being bandwidth hogs, and Schlumberger's in-house application to interpret oil-field data is no exception. It sends about 20 GB to 40 GB of data each time it synchronizes updates across the WAN, which can saturate the entire pipe, says Elmenshawy. "The [NetQoS] tool showed abnormally high and sustained traffic on an unusual TCP port, which we found was used by the application to synchronize the software baselines," Elmenshawy says.
The main limitation of the NetQoS tools is that they provide only the data you request, such as the performance of Schlumberger's backup system. And the RMON2 probes are prone to going offline and spitting out invalid data. "But the NetQoS software identifies probes and the last date it received data from them, and whether the data was invalid," Elmenshawy says.
The trade-off with the new QoS for Schlumberger is that it requires replacing many of the company's Cisco Systems routers with newer Cisco models that come packaged with the differentiated services-type QoS. It's still unclear just how the bandwidth-analysis tools will measure the QoS performance. "We're working with NetQoS [on this]," Elmenshawy says.