The Quality/Efficiency Product: The Reason to QoS-Enable a Network
- The Quality/Efficiency Product: The Reason to QoS-Enable a Network
- 2 Raising the QE Product of a Network
- 3 The Value of Different QoS Mechanisms in Raising the QE Product of a Network
- 4 Illustrative Examples
- 5 Sharing Network Resources: Multiple Resource Pools
- 6 Summary
The previous chapter reviewed various QoS mechanisms. Subsequent chapters will discuss the use of these mechanisms to build a QoS-enabled network. The mechanisms described can offer significant benefits, but they are not cost-free. QoS mechanisms incur varying degrees of overhead both in terms of processing and memory in network elements and in terms of administration and management.
This chapter introduces the notion of the quality/efficiency product. This metric can be used to weigh the benefit of a particular QoS mechanism so that it can be compared to the cost of the mechanism. The net value that a QoS mechanism brings to a network then can be assessed in a relatively objective manner, avoiding much of the controversy and zeal that typically surrounds QoS mechanisms. Throughout this book, the concept of the quality/efficiency product will be used when evaluating various QoS mechanisms and their application.
The QoS Controversy
The tradeoffs between the benefits offered by QoS mechanisms and the overhead associated with these mechanisms is at the root of the controversy that has always surrounded the discussion of QoS mechanisms.
As it exists today, the Internet offers a simple best-effort service with very primitive QoS mechanisms. All packets generally are handled at the same service level and are forwarded in the order in which they are submitted to the network, with no assurances regarding available bandwidth or latency. QoS mechanisms in the Internet currently are limited to physical or logical reprovisioning (for example, adding links on specific routes or reconfiguring ATM VCs) or rerouting traffic around congested network regions.
The telephone network, on the other hand, makes use of heavyweight QoS mechanisms, including sophisticated per-conversation signaling (SS7) and per-conversation traffic handling (in which a 64Kbps channel is dedicated to each phone call). As a result, the telephone network is capable of offering very strict guarantees regarding available bandwidth and latency.
At one extreme, hardcore Internet traditionalists balk at the notion of bringing the overhead and complexity of any new QoS mechanisms to their simple best-effort Internet. At the other extreme are heretics who want to convert the Internet to a fully circuit-switched network (similar to the telephone network) by forcing per-conversation signaling and per-session traffic handling into every switch and router.
Early battles have been fought. ATM was adopted only to a limited degree and only in backbone networks, a far cry from the model of switched virtual circuits to every desktop that had been envisioned by ATM zealots. RSVP never even made it to the battlefield before being pulled back by the IETFs applicability statement (see Chapter 5, RSVP for further details on RSVP and the IETF applicability statement). Yet the Internet is currently falling short of its potential. Often it is unusable for mission-critical enterprise applications (which rely instead on private leased lines), and certainly is not suitable for most multime-dia applications. Adopting some level of QoS functionality probably will enable the Internet to realize its full potential.
In addition to introducing the concept of the quality/efficiency product, this chapter briefly discusses the value of various combinations of QoS mechanisms in raising a networks quality/efficiency product. Some of the concepts discussed may be confusing at first, but these will become clearer throughout the book as each QoS mechanism is discussed in greater detail.
Recall the definition of network QoS from Chapter 1, Introduction to Quality of Service:
Chapter 1 also states that if network resources are infinite, the service needs of all applications can be met. Fortunately, it is possible to meet the service needs of important applications without infinite resources. However, in many cases, a great deal of resources may be required to do so. When QoS needs are met purely by adding resources to the network, the network is often overprovisioned. QoS mechanisms are valuable because they can make it possible to provide the QoS required by applications without overprovisioning the network. In other words, QoS mechanisms make it possible to simultaneously provide the required QoS and to operate the network more efficiently than would otherwise be possible.
Note that QoS mechanisms do not create resources. They merely reallocate existing resources among different traffic flows. When certain traffic flows are allocated more resources, other traffic flows are allocated fewer resources. However, certain applications require additional resources for their traffic flows, while others operate satisfactorily with fewer resources.
This section defines the quality of a service and the efficiency of network resource usage. Subsequent sections explain how the required quality of a service drives the trade-off between efficiency of network resource usage and the use of QoS mechanisms with their associated costs.
2.1.1 The Quality of a Service
Different qualities of service are appropriate for different applications. The quality of a service refers to the level of commitment provided by the service and the integrity of that commitment.
In this context, a service is intended to refer to the specific assurances that are provided to a set of network traffic regarding available capacity, delivery latency, and reliability of delivery. In other contexts, a service could be interpreted in broader terms to include such aspects as encryption, mean time between failure, and billing issues, among others.
The following are examples of services in order of decreasing quality:
100Kbps of traffic will be carried from point A to point B with zero packet loss. Each packet will be delivered from source to destination in less than 100 milliseconds.
1Mbps of traffic will be carried from point A to point B with the appearance of a lightly loaded network.
100Kbps of traffic will be carried from point A to point B; 95% of packets will be delivered in less than 5 seconds.
1Mbps of traffic will be carried from point A to point B in less time than it would be delivered without using this service.
Traffic from point A will be delivered to its destination in less time than it would be delivered without using this service.
The first two services offer strictly quantifiable bounds on latency and packet loss. The third and fourth services offer somewhat less quantifiable service. The fifth service offers no quantifiable parameters at all. The first and second services correspond to the Integrated Services (IntServ) definition of Guaranteed and Controlled Load Service, respectively. (The relative quality of the second and third services described could be debated, depending on ones definition of a lightly loaded network). Finally, the fourth and fifth services are types of better-than-best-effort (BBE) service. The fourth service constrains the volume of traffic that will be serviced and the route along which it will be accommodated. The fifth service offers no constraints.
Generally, it is true that the stricter the constraints associated with a service, the higher the quality that can be offered. The value of the fifth service is hotly debated. I tend to refer to it as the better-than-Joe service because it seems to assure the customer only that no matter how bad the network appears to the customer, some poor soul is even worse-off. Nonetheless, I am sure that this is a marketable service. Often the terms quantitative and qualitative are used to refer to services such as Guaranteed and Controlled Load on the one hand, versus BBE services on the other hand. Other terms commonly used are hard services versus soft services.
It is fairly clear that the services listed offer progressively lesser levels of commitment (allowing for some ambiguity between the second and third services). Another aspect of the service that determines its quality is the integrity of the service. Conventional wired telephony services have high integrity in the sense that a call in progress generally does not diminish in quality after it has been granted, no matter how many other people in the neighborhood attempt to place calls.
A service quality is not necessarily related to the actual amount of resources committed, nor to the cost of the resources. In the previous example, the first service is of higher quality than the second service, even though it offers less network capacity. Furthermore, an appraisal of the quality of a service is not a judgment of its value to the end user, but rather a statement of its suitability to different applications. For example, a BBE service may be entirely satisfactory for a certain Web-surfing application, while a higher-quality service may be required to handle interactive voice traffic. It is reasonable to assume that, all else being equal, higher-quality services cost more than lower-quality services. Given this assumption, the high-quality service might be deemed excessive for the Web-surfing application. From a cost/performance perspective, the lower-quality service would be preferable to the end user.
In the context of this discussion, efficiency refers strictly to the amount of network capacity (in terms of bandwidth) required to provide a certain quality service. It does not refer to processing overhead, management burden, or any other potential inefficiencies that might be associated with providing network services. Efficiency is an important consideration because bandwidth is often expensive. Of course, in some situations bandwidth is inexpensivein this case, efficiency is less of a concern.
2.1.3 The Quality/Efficiency Product
For any given network, the higher the Quality of Service required, the more bandwidth is required.
Obviously, this is true only within certain limits. Above a certain bandwidth or network capacity, further bandwidth availability will not improve the QoS.
This is the essence of the quality/efficiency product. In a given network, with a given set of QoS mechanisms, the product of service quality and efficiency is fixed. If higher-quality service is required, the network must operate with higher capacity and, hence, less efficiently. If the network is operated more efficiently, the quality of services that can be offered is compromised. This can be expressed as follows:
This value characterizes the network from a QoS perspective and will be referred to in the remainder of this book as the quality/efficiency (QE) product of the network.
Figure 2.1 characterizes a specific network with a QE product of C. This is the QE curve of the network.
Figure 2.1 Quality/Efficiency Product for a Given Network
Networks may be operated at different points on their QE curve. For example, most LANs on the Microsoft campus are capable of offering reasonable quality for IP telephony sessions. This is because they tend to be overprovisioned. As such, the LAN operates at a point on the curve (P1) that corresponds to high quality but low efficiency. By contrast, most international WAN links are provisioned very efficiently (for cost reasons).
Consequently, these offer poor quality with respect to IP telephony applications. The international WAN links are operated at a point on the QE curve (P2) that corresponds to high efficiency and low quality. In this example, the WAN and the LAN regions of the network are shown to be operating on the same QE curve. In general, different curves will correspond to the two networks.