|
|
Designing an internetwork is just the starting point. To get the best out of it you must systematically fine-tune key areas of the design to improve performance, preserve critical bandwidth, and minimize operating costs. Network optimization is typically performed soon after the network design is complete, and it must be performed periodically throughout the lifetime of the network to ensure that changes in traffic dynamics are monitored and responded to. It is possible to improve even a good network design considerably through judicious configuration, tuning, and the use of specialized software and hardware tools. However, before attempting optimization you should know the baseline characteristics of the network, its systems, and its protocols. Typically, the acceptance stage is the best time to collate baseline data; failing that, you should at least characterize the key systems before attempting to optimize. Without this database you have nothing with which to make comparisons later, and this could lead to wasted time and effort. During the capacity-planning phase you may have also identified potential bottlenecks; again this, may help to focus your attention if performance problems occur [1].
The complex interactions between users, applications, protocols, and systems create dynamic loads on a network that are often hard to predict at the design phase. As networks expand, it becomes all the more important to understand, control, and optimize traffic over the network to preserve valuable bandwidth (and possibly meet service guarantees). This active form of control is often referred to as policy. Traffic engineering policy comprises a set of rules that govern the distribution of traffic throughout the network. Examples include the following:
To regulate backbone traffic to a maximum of 25 percent average bandwidth during the workday, with one-minute peak traffic not exceeding 70 percent utilization.
To ensure that all nonessential traffic is filtered off the backbone.
To ensure that communication between remote departments occurs transparently, irrespective of the differences in technology employed.
To implement policy there are many standards-based and proprietary tools and technologies available to the network designer—some with only system-level context and others with network-wide scope. Lack of integration in multivendor environments is currently a major issue, and careless use of these tools may result in contradictory or inconsistent behavior. Unfortunately, there is no magic wand. Table 7.1 lists a number of techniques available today and the main benefits achieved by using those techniques.
Optimisation Techniques | Reduces Latency | Conserves Bandwidth | Routing Performance | Protocol Performance | Application Performance |
---|---|---|---|---|---|
Traffic Prioritisation & Queuing | ✔ | ü | ✔ | ✔ | |
Packet Filtering | ✔ | ✔ | ✔ | ||
Load Sharing | ü | ✔ | ✔ | ✔ | |
Caching Techniques | ✔ | ✔ | ✔ | ✔ | ✔ |
Hierachical Routing | ✔ | ✔ | |||
Data Compression | ü | ✔ | |||
Protocol Tuning | ü | ✔ | ✔ | ||
Proxy Services | ✔ | ✔ | ✔ | ✔ | |
Keepalive/Local ACK Spoofing | ✔ | ✔ | |||
Storage Optimisation | ✔ | ✔ | ✔ | ✔ |
This chapter deals with the technologies that you are likely to encounter and use in the optimization process. We take a broadly hierarchical approach, starting at the backbone level and working down to specific protocol tuning options. In Chapter 8 we take a much broader view of traffic policy by introducing architectures used to provide network Quality of Service (QoS). This is an area of much recent interest and research, especially in the Internet and backbone environments. There is now a concerted effort to improve QoS provisioning in the IP environment, and this bodes well for the future. In Chapter 9 we present an overall policy management framework. For further details and a more formal discussion of performance analysis, the interested reader is referred to [2].
|
|