Running Video over the LAN Paper by consultant Nathan Muller that looks at the problems this can cause and possible solutions. QoS over the Internet: The RSVP protocol The NetCorp NetPlatform Description of a proprietary QoS LAN plan, from NetCorp. Efficient Quality of Service Support in OS for Multimedia Applications Series of papers from the Washington University Applied Research Laboratory. Gigabit Ethernet overview From the Gigabit Ethernet Alliance. Prodan is vice president of marketing at Extreme Networks, a Layer 2 and 3 Gigabit Ethernet switch maker in Cupertino, Calif. He can be reached via the Internet at gprodan@ extremenetworks. com.

By George Prodan
Network World, 6/9/97

Until recently, the ability to manage network resources and control bandwidth allocation and traffic prioritization on Ethernet networks has been something network managers could only dream about. But with policy-based quality of service (QoS), the dream becomes reality.

Policy-based QoS allows a network manager to allocate bandwidth and prioritize traffic within the network based on a set of administrative policies and usage patterns. This ability to control the ebb and flow is now more important than ever: Network traffic is at an all-time high due to the influx of Web-based data and video streams crossing the corporate backbone.

Consider, for example, a manufacturing company that has good network performance during normal day-to-day operations. But during the crunch of end-of-month shipping and billing processes, the network slows to a crawl and other company functions are put on hold to recover the required bandwidth.

With policy-based QoS, bandwidth can be allocated evenly across all departments for the first three weeks of the month, then preset to favor the accounting and shipping departments during the last week. Traffic originating from the manufacturing subnet can be given priority, while ensuring that all other traffic continues to get through, even if at a lower performance level.

What's good for ATM . . .

Building on emerging standards including the IETF Resource Reservation Protocol (RSVP) and IEEE 802.1p and 802.1Q, policy-based QoS is a high-level capability that adds detailed traffic grouping, assignment of QoS profiles and mapping to queues associated with specific switch or router ports.

This higher level capability means added control for the network manager. RSVP alone, for example, provides explicit QoS. Explicit QoS is endstation initiated and gives the power to request bandwidth to the individual user. Left unchecked, RSVP-enabled endstations can usurp bandwidth and starve other - possibly more critical - applications.

Implicit QoS is policy based; it returns control to the network manager. Policies are set centrally based on the network's unique traffic patterns.

QoS traffic groups

These traffic groups are mapped onto multiple queues associated with a policy-based, QoS-enabled switch. For each queue, the QoS profile can include minimum bandwidth, maximum bandwidth, peak bandwidth, relative priority and maximum delay.

Depending on the unique needs of the network, the network manager may choose to set all of these variables or use a subset. Or, perhaps only one or two traffic groups need QoS profiles assigned - all other traffic on the network would then fall into ''best effort,'' as if no QoS scheme were employed.

Setting minimum and maximum bandwidth will guarantee average bandwidth to a particular traffic group over time. Although short bursts of data may actually exceed the maximum, the system will contain bursts to an average over time. Setting a limit on peak bandwidth will permit the bursts of data to bring the connection up to average bandwidth more quickly.

When multiple traffic groups are assigned QoS profiles, bandwidth is allocated equitably across all the queues. All things being equal - that is, when the total of all the minimum levels of bandwidth assigned do not exceed 100 % - each queue is guaranteed its minimum bandwidth.

The tie-breaker

For delay-sensitive traffic, including that generated by applications using an admission control mechanism such as RSVP, a maximum delay setting will einsure timely delivery. This is critical for video and multimedia traffic.

Because policy-based QoS is implemented through virtual queues on the enabled switch, networkwide QoS can be achieved without altering endstations. This represents a breakthrough in the ability to manage and control the increasing diversity of traffic found in today's LANs.