Performance in a Router Network
A customer perceives the network performance by the response time seen for the application. The IBM 2210 and 6611 Network Processor are some examples of the many contributors to response time, and their performance in the customer environment is really defined by their contribution to the overall system response time.
Performance varies by protocol and packet size in the router network. Kevin Tolly, in an article in Network World (Aug. 10, 1992), states that “traffic on a typical corporate internet consists of larger frames, such as those used in Novell, Inc. NetWare or Microsoft Corp. LAN Manager file transfers. Measuring performance using 64-byte packets is largely of academic interest, since virtually no applications use such small packets.”
Performance information then must be specified in such a way that expectations for a given LAN or WAN interface are factored into the network design. WAN speed is often the limiting factor in throughput or response time measurements. Traffic patterns of the protocols being supported are also a factor when designing a network to support the necessary response time for an application.
Another performance consideration that is added to the aggregate of the performance and capacity is the overhead of router-to-router exchanges. As each protocol is added, a new flow of exchanges between the routers must occur. Network management is another internal flow that must be given consideration when determining the necessary box performance to support the system performance requirements.
Routers were designed with connectionless-oriented protocols in mind. Each data packet finds its way through the network with the information available in its header. Path information is maintained via router information exchanges outlined in protocols such as TCP/IP¢s RIP and OSPF. Broadcasts are used to learn locations of other end stations, services, and zones. Today's connection-oriented protocols are encapsulated to take on these characteristics. Performance under this encapsulation may be more comparable from the end stations point of reference, such as response time, as each implementation provides unique function in its support of connection-oriented protocols ranging from spoofing to broadcast and flow control.
Performance is limited by the hardware on which the function is being tested. Each interface must be tested with the protocols and function supported on it. After looking at the card level performance, box level performance information becomes one of the entities in the overall network performance information. Only then can performance be estimated for the network and bottlenecks pinpointed. End-to-end response time is the last of the performance information that can be gathered.
The primary metrics used to characterize the performance of a device are:
· Responsiveness is the time interval between an input and the corresponding output.
· Productivity is the amount of work done during a time interval.
· Utilization is the ratio of busy time to elapsed time.
Each is used to describe some aspect of how well a device processes jobs (units of work to be performed by the system). For routers, a job is usually synonymous with a routed packet.