ATLANTA - James Wiedel might seem the sort of network manager who'd be eager to investigate quality of service (QoS) on a network that seems to grow at exponential rates.
As the director of networking at the University of Southern California (USC), Wiedel oversees a network with some 20,000 end stations and almost 30,000 switched Ethernet ports spread over several locations, many connected by gigabit links. Increasingly, end users want access to multimedia and other bandwidth-hogging applications.
But Wiedel is having nothing of QoS. He'd rather just throw more bandwidth on his network as demand grows.
It's a lot cheaper, less complex and eliminates a host of political problems that can come with deciding whose traffic gets what priority, Wiedel said in a panel discussion on QoS at NetWorld+Interop in Atlanta last week.
"I work in a university; we have these professors there and they're some of the biggest primadonnas around,'' he said. "They want everything" - and what they don't want is to have somebody else get - or seem to get - higher priority on the network, he said.
"People will be unhappy because somebody else might be getting better service," he said. "He may not be, but the other group or people or individuals think he is. ... You're going to have some unhappy folks no matter what you do.''
In contrast, simply adding more bandwidth is "the only way to really be fair," he said. With everybody assured of the same access, network managers can count on peer pressure among end users to help curb political demands, he said. In any case, "you can engineer your network to have more bandwidth than your users can use," he said. "You can do that today. It's also fairly inexpensive today and easy to implement if you have the right technologies."
These include moving to full duplex links, he said. He added that several vendors now support trunking, in which several parallel links can be grouped to form a single, logical fat pipe. Non-trunked parallel links can be used for load balancing, as well, he said.
"You're going to run out of processor (power) on your end station before you run out of bandwidth,'' he said. In fact, adding too large a pipe to a user's PC could swamp that system, causing performance problems.
Wiedel acknowledged that throwing more bandwidth at the problem may not be practical on the WAN, because of the cost. But even there, he is skeptical of QoS, because he relies on the Internet to carry traffic, and there is no guaranteed QoS from one ISP to another, he said.
Ashley Stevenson, CEO of QoS vendor Xedia, however, disagreed. The problem, especially on the WAN is that such critical applications as SNA and SAP traffic can often "get swamped out by the ESPN browser activity or large e-mail attachment activity."
"People are running more apps than ever," he said. "It's not uncommon for us to deploy 500 apps on a global WAN. They're all just sloshing around."
But even without sports downloads, QoS schemes are needed for applications that expect zero packet loss, he said.
Richard Muirhead, director of business development at QoS vendor Orchestream, agreed. He said a growing number of applications simply cannot tolerate the best-effort packet delivery of native TCP/IP.
Over-provisioning a circuit could work in theory, but in practice would simply be too expensive on the WAN, Stevenson said.
Muirhead said QoS need not be very expensive, because many networking devices on the market today support it.
RELATED LINKS
Convergence divergence
Vendor debate reveals interoperability issues, spotty product plans at N+I forum. Network World, 9/20/99.
