Quality of Service: Directing Data through the Network

• 10/06/1999

Bandwidth is the oil of information technology, and network managers can’t provide enough of it. Since slow servers and poorly written applications never catch the blame for application response time problems, it is typically incumbent upon the network group to provide a solution.

The usual way network executives provide users with the bandwidth their applications need is to try to expand the pipes as much as the technology allows. Applications such as streaming video and voice over IP, however, stretch the limits and tend to rip open the seams of corporate networks.

Much has been made recently of quality of service (QoS) technology. QoS promises to mitigate bandwidth problems by prioritizing traffic so the most important stuff gets through first. Microsoft Corp. plans to include QoS technology in Windows 2000 and Windows 98. Despite the early stages of the technology, Microsoft is pushing QoS hard. Network equipment vendors are also preaching QoS to meet the demands of multimedia applications and the convergence of voice and data. Therefore, now is the time to become familiar with QoS and how it may help manage bandwidth across your enterprise.

The key goal of QoS is to help network managers establish bandwidth policies that ensure the data a company considers most important gets through first. For example, a company running a large SAP application might consider it important that SAP transactions execute ahead of an FTP file transfer; or company executives might deem that a videoconference running over their newly converged voice/data network gets priority over SAP. Characteristics such as tolerance for delay, throughput and jitter are taken into consideration by a QoS-enabled network to determine what applications get precedence.

Several mechanisms, backed by emerging Internet Engineering Task Force (IETF, www.ietf.org) standards, work in harmony to ensure prioritization. An application on a Windows 98 workstation, for example, uses protocols such as the resource reservation protocol (RSVP) to request network resources. Policies defined by the network group reside in network devices, such as switches and routers, that get information from policy servers running common open policy service (COPS) and LDAPv3-compliant policy repositories. As applications traffic traverses the network, the policies are consulted and prioritization is applied to the data. On the LAN, class of service (CoS) bits are set at the media access control (MAC) layer and the IEEE 802.1p protocol ensures that data arrives in the correct order. Across IP networks, above Layer 2, protocols such as DiffServ map the CoS bits to the never-before-used DiffServ field -- previously known as the TOS field -- in the IP header. The end result is that application usage of network resources gets smoothed out, with mission-critical applications getting priority.

While your network vendor will make this effort seem absolutely necessary and trivial, you know better. "While QoS sounds great in theory, it truly approaches rocket science in reality," says David Passmore, research director and founder of NetReference (www.netreference.com), a networking consulting firm. "The learning curve is steep. Take for example, the idea of directory-enabled networks and the storing of policy information in LDAP directories. Now, network managers have to start tackling concepts such as schemas, stuff that they thought only database types had to worry about. The list of new concepts is pretty big," he explains.

In addition to complexity, Passmore cites three more reasons why QoS implementations still have a way to go. The first is interoperability, which "is non-existent right now," Passmore notes. "It’s hard enough for vendors to handle QoS in their own equipment let alone someone else’s. Furthermore, for true end-to-end QoS, components ranging from the application itself, the network interface card, switch and router will all have to participate, requiring the cooperation of a host of vendors."

The second reason is the issue of emerging standards. "For example, COPS isn’t handled by most switches and routers yet. RSVP [which Microsoft is implementing in Windows 2000 and Windows 98] is very resource intensive," Passmore notes.

The third big obstacle is office priorities. "Perhaps the greatest challenge revolves around the politics of QoS. Who gets to divvy up the bandwidth and to who's benefit? Imagine the meeting with your company’s business unit directors and asking them to figure out who’s application is the most important and deserves the best treatment."

While bandwidth management via QoS faces some impediments, these haven’t stopped vendors from forging ahead with policy management systems. And it's not just the networking equipment giants. Other vendors are investing heavily into QoS product development, citing the non-proprietary nature as a major reason to consider them.

"We see the acceptance of QoS technology occurring in phases," notes Charles Muirhead, founder and president of Orchestream (www.orchestream.com), a QoS technology vendor. "The first critical need is to manage application availability. QoS implementations will focus on increasing application uptime by insulating the application from changes in network congestion, such as frame relay problems."

"Almost every company wants to reduce the number of networks they run by consolidating protocols on a single backbone," Muirhead says. "Today there is no way to guarantee the uptime and performance of SNA applications. Therefore, QoS is critical to this consolidation process. Being able to consolidate networks and guarantee application uptime and performance can result in a huge return on investment in QoS technology."

After companies invest in QoS for its insurance policy aspects, Muirhead believes more serious attention will be paid to the cost-optimization benefits. QoS has the ability to steer traffic into appropriate queues, thus providing better network utilization. Network executives will be better armed to make decisions on potential bandwidth upgrades and will make better use of the bandwidth they already have.

While some companies see applications such as voice over IP (VOIP) and streaming media -- which are pushed by many of the networking heavyweights -- as major reasons to implement QoS, Muirhead says there are other good reasons to employ QoS in a network.

Similar to Orchestream, Hewlett-Packard Co. is initially focusing its attention on the performance of mission critical applications, especially over WAN links. "Performance concerns of mission-critical applications across slow links is driving the main buying interest for us," notes Chuck Darst, product manager at Hewlett-Packard. HP’s PolicyXpert, a member of the OpenView family of products, is also designed to enable policy-based QoS among heterogeneous networks.

"Congestion occurs at the edge of a network, between the LAN and WAN. Our customers tell us that it is most difficult to control application bandwidth where jitter, delay and information loss during traffic bursts are unacceptable," Darst says.

Jennifer Geisler, manager of product marketing at Packeteer Inc. (www.packateer.com), agrees. "Bottlenecks occur most frequently when a high-speed LAN narrows into a lower-speed WAN. Due to the nature of IP, non-urgent applications swell to fill the available bandwidth, putting mission critical applications at risk," she says.

While implementing a full-blown, end-to-end QoS might prove daunting to many, there are technologies available that can help.

"Before embarking on a major QoS project, the question you need to ask yourself first is, ‘Do I know how my network is performing today?’" says Jim McQuaid, director of monitoring solutions at Ganymede Software Inc. (www.ganymede.com). Ganymede produces Chariot, a network testing tool, and Pegasus, which helps managers view network performance on a daily basis. "The first step in any effort to provide fair application bandwidth is to bracket your applications and characterize your network traffic."

In the past, undertaking such a traffic characterization project meant breaking out the network analyzer, creating complicated scripts and filters, capturing traffic and then sifting through hundreds of pages of network traces. But the most daunting task was to translate the data into a format that our bosses and business directors could understand.

"Most IT managers can’t settle for this type of analysis," Packeteer's Geisler says. "What they want to do is to quickly identify the applications running on their network. They want a simple list of applications and how they are consuming their network’s bandwidth."

To that end, Ganymede’s Chariot/Pegasus combination and Packeteer’s PacketShaper, a bandwidth management system, can help network managers efficiently characterize their network traffic. Chariot, for example, uses scripts to bounce traffic off of end-point agents that you distribute on boxes -- Chariot supports 16 operating systems -- across your network while collecting test data. Chariot is useful in performing benchmarking and "what if" activities. This helps network managers get a picture of how a network is performing now, and also a sense for how it will perform under different conditions.

Packeteer’s PacketShaper sits behind a WAN access router and is geared for WAN application bandwidth control. Packeteer can identify more than 150 types of traffic, including Microsoft Exchange, NetShow, Terminal Server, Oracle, SAP and Citrix WinFrame. After placing the product on your network, PacketShaper can dynamically assess the traffic flowing through it, classifying the traffic by a number of criteria, including application, protocol, Web destination, subnet and others.

Bandwidth issues aren’t going to wait until you make some sense out of the QoS blitz. Assuming that you have done a network analysis and you know where your bottlenecks are located, here are a few options for alleviating your immediate bandwidth crunch.

If your WAN is getting whacked, a product like PacketShaper can help. PacketShaper can control traffic and report trends, as well as characterize and analyze applications running over your network. Although useful throughout your network, Packeteer targets the WAN bandwidth problem -- where most data communications budgets get hit hardest. PacketShapers are simple to install and can be employed immediately because they don’t rely on the maturing of QoS standards to be useful.

If you need some help on the LAN, a product such as IP Metrics Software Inc.'s (www.ipmetrics.com) NIC Express can provide relief. NIC Express can group up to four NICs in a Windows NT server, effectively creating a RAIN (Redundant Array of Inexpensive NICs). A RAIN provides fault tolerance and additional throughput.

Bandwidth management is now more critical than ever. And sophisticated tools that can do a more effective job are finally appearing. As is characteristic of most of these issues, the technology learning curve is steep. Now is the time to get familiar with the technology. That way you will be prepared when management hands you the next killer app to run across your network.