1.9 Performance Characteristics

1.9 Performance Characteristics

Services may include one or more of the performance characteristics we have mentioned so far in this chapter: capacity, delay, and RMA. Each characteristic is actually a label for a class of characteristics of that type. For example, the term capacity is used as a label for the class of characteristics that involves moving information from place to place, including bandwidth, throughput, goodput, and so forth. Similarly, delay is a label for the class of characteristics that includes end-to-end delay, round-trip delay, and delay variation. RMA is a label for the class of characteristics that includes reliability, maintainability, and availability. Thus, when the terms capacity, delay, and RMA are used in this book, you can use other terms from each class, depending on your network.

There are times when it makes more sense to describe capacity in terms of throughput, for example, when developing requirements for applications. Round-trip delay is commonly used as a measure for delay, although at times delay requirements are expressed in terms of one-way delay.

1.9.1 Capacity

Capacity is a measure of the system's ability to transfer information (voice, data, video, or combinations of these). Several terms are associated with capacity, such as bandwidth, throughput, or goodput. Although we will use the generic term capacity throughout this book to reference this class of characteristics, you may choose to use another term in place of or along with capacity.

Bandwidth is the theoretical capacity of one or more network devices or communications links in the system. Theoretical, or raw, capacity does not take into account overhead from higher-layer protocols or the performance loss related to inefficiencies in the system (e.g., delays in the operating system or peripheral devices).

Throughput is the realizable capacity of the system or its network devices. Values for throughput vary, depending on system design, types and configurations of equipment, and where in the protocol stack the measurement for capacity is being taken.

Example 1.8

The bandwidth of a SONET OC-3c link is 155.52 Mb/s, which is three times the bandwidth of an OC-1 link (51.84 Mb/s). This bandwidth does not include data-link, network, or transport-layer protocol (e.g., ATM, IP, or transport control protocol/user datagram protocol [TCP/UDP]) overhead or, in the case of wide-area networks, the loss in performance due to the bandwidth delay product in the network. When a network or element is performing at its theoretical capacity, it is said to be performing at line rate. When an OC-3c circuit was tested, values of realizable capacity (throughput) ranged from approximately 80 to 128 Mb/s (measurements taken at the transport [TCP] layer of the National Research and Education Network [NREN] and Numerical Aerodynamic Simulation [NAS] networks, NASA Ames Research Center, March 1996).

A discussion on protocol overhead is presented in Chapter 11.

Both bandwidth and throughput are important terms when describing system, network, application, and device capacities. Each gives a different and compatible view of capacity.

1.9.2 Delay

Delay is a measure of the time difference in the transmission of information across the system. In its most basic sense, delay is the time difference in transmitting a single unit of information (bit, byte, cell, frame, or packet) from source to destination. As with capacity, there are several ways to describe and measure delay. There are also various sources of delay, such as propagation, transmission, queuing, and processing. Delay may be measured in one direction (end-to-end) and both directions (round-trip). Both end-to-end and round-trip delay measurements are useful; however, only round-trip delays can be measured with the use of the practical and universally available utility ping.

Another measure of delay incorporates device and application processing, taking into account the time to complete a task. As the size of a task increases, the application processing times (and thus the response time of the system) also increase. This response time, termed here latency, may yield important information about the behavior of the application and the network. Latency can also be used to describe the response time of a network device, such as the latency through a switch or router. In this case the processing time is of that switch or router.

Delay variation, which is the change in delay over time, is an important characteristic for applications and traffic flows that require constant delay. For example, realtime and near-real-time applications often require strict delay variation. Delay variation is also known as jitter.

Together, delay (end-to-end and round-trip), latency, and delay variation help describe network behavior.

1.9.3 RMA

RMA refers to reliability, maintainability, and availability. Reliability is a statistical indicator of the frequency of failure of the network and its components and represents the unscheduled outages of service. It is important to keep in mind that only failures that prevent the system from performing its mission, or mission-critical failures (more on this in Chapter 2), are generally considered in this analysis. Failures of components that have no effect on the mission, at least when they fail, are not considered in these calculations. Failure of a standby component needs tending to but is not a mission-critical failure.

Reliability also requires some degree of predictable behavior. For a service to be considered reliable, the delivery of information must occur within well-known time boundaries. When delivery times vary greatly, users lose confidence in the timely delivery of information. In this sense the term reliability can be coupled with confidence in that it describes how users have confidence that the network and system will meet their requirements.

A parallel can be seen with the airline industry. Passengers (users) of the airline system expect accurate delivery of information (in this case the passengers themselves) to the destination. Losing or misplacing passengers is unacceptable. In addition, predictable delivery is also expected. Passengers expect flights to depart and arrive within reasonable time boundaries. When these boundaries are crossed, passengers are likely to use a different airline or not fly at all. Similarly, when an application is being used, the user expects a reasonable response time from the application, which is dependent on the timely delivery of information across the system.

Along with reliability is maintainability. Maintainability is a statistical measure of the time to restore the system to fully operational status after it has experienced a fault. This is generally expressed as a mean-time-to-repair (MTTR). Repairing a system failure consists of several stages: detection; isolation of the failure to a component that can be replaced; the time required to deliver the necessary parts to the location of the failed component (logistics time); and the time to actually replace the component, test it, and restore full service. MTTR usually assumes the logistics time is zero; this is an assumption, which is invalid if a component must be replaced to restore service but takes days to obtain.

To fully describe this performance class, we add availability to reliability and maintainability. Availability (also known as operational availability) is the relationship between the frequency of mission-critical failures and the time to restore service. This is defined as the mean time between mission-critical failures (or mean time between failures) divided by the sum of mean time to repair and mean time between missioncritical failures or mean time between failures. These relationships are shown in the following equation, where A is availability.

A = (MTBCF)/(MTBCF + MTTR) or A = (MTBF)/(MTBF + MTTR)

Capacity, delay, and RMA are dependent on each other. For example, the emphasis of a network design may be to bound delay: A system supporting point-of-sale transactions may need to guarantee delivery of customer information and completion of the transaction within 15 seconds (where the network delay is on the order of 100s of ms); a Web application can have similar requirements. However, in a computeintensive application we may be able to optimize the system by buffering data during periods of computing. In this case, delay may not be as important as a guarantee of eventual delivery. On the other hand, a system supporting visualization of real-time banking transactions may require a round-trip delay of less than 40 ms, with a delay variation of less than 500 s. If these delay boundaries are exceeded, the visualization task fails for that application, forcing the system to use other techniques.

1.9.4 Performance Envelopes

Performance requirements can be combined to describe a performance range for the system. A performance envelope is a combination of two or more performance requirements, with thresholds and upper and/or lower limits for each. Within this envelope, levels of application, device, and/or network performance requirements are plotted. Figures 1.28 and 1.29 show two such envelopes. The performance envelope in Figure 1.28 consists of capacity, in terms of data sizes transferred across the network, and end-to-end delay. In this figure, delay is shown as 1/delay for consistency.

Figure 1.28: Example of a 2D performance envelope.

Figure 1.29: Example of a 3D performance envelope.

Figure 1.29 is a 3D performance envelope, showing capacity, delay, and RMA. This envelope also describes two regions of performance, low and high performance, which are functions of the limits and thresholds for capacity, delay, and RMA.

Performance envelopes such as these are useful for visualizing the regions of delay, capacity, and RMA in which the network will be expected to operate based on requirements developed for that network. In Chapter 2 we will discuss how requirements are developed for a network.