This theory assumes that the value a provider of a service receives is ran- dom because of market uncertainty. The theory illustrates how allowing users to pick the best approach from many ways to provide a similar service provides a good chance of achieving a service with features that are a supe- rior market match. The theory accounts for technical and management
Theories about How to Manage Network-Based Services 97
advantages from the centralized architecture for network-based services, comparing them to the benefits of easy experimentation with the distrib- uted management structure. The theory states that when the advantages of the more centralized service architecture outweigh the benefits of many experiments, a centralized management structure may be justified. Finally, the theory accounts for how services evolve from generation to generation.
It states that at each generation of a service, service providers learn from the previous generation about what will work better for the next generation.
Here are two fundamental assumptions and a simple theory:
A S S U M P T I O N 1 The market demand for network-based services has market uncertainty. This means that service providers (which includes
entraprise users) are unable to accurately predict the value they will receive for providing a service. This market uncertainty is the subject of Chapter 6.
A S S U M P T I O N 2 Experimentation with services is possible, and a market exists to value the experiments. The value of a particular experiment is the success of its adoption. This experimentation is used to determine what service best matches the current market conditions in the context of what features will be the most popular.
T H E O RY 1 The expected value of the best of nsimultaneous attempts at providing a service is likely to exceed the expected value of any single
experiment. As n increases, the possibility of a truly outstanding market match grows. This result is illustrated in Figures 6.1 and 6.2 in Chapter 6.
One way to view Theory 1 is in the context of options — having a choice is analogous to having an option. This theory demonstrates the value of net- work architecture promoting many experiments, compared to management structure, where experimentation is more difficult. One example of this is the choice between two standards for Voice-over IP, SIP, and megaco/H.248. As discussed in Chapter 10, SIP allows both end-2-end applications and appli- cations with a more centralized structure, but megaco/H.248 does not.
Theory 1 illustrates the value of protocols, such as SIP, that allow flexibility in the management structure they allow applications to have.
What follows is a set of stronger assumptions allowing a deeper theory considering the management structure of services based on the degree of market uncertainty. It defines more precisely how services have different management structures and how each structure has different values and attributes.
98 Chapter 7
A S S U M P T I O N 3 The payout to the service provider offering the best of n choices is nonlinear. More experimentation and greater uncertainty increase the expected value. The service provider receives this value by providing the service that best matches the market. This is explained in Chapter 5 on options.
A S S U M P T I O N 4 The less disruptive and less expensive it is to develop and deploy a service, the more experiments there will be. Experiments in networks with infrastructure allowing applications with end-2-end architecture requiring no alteration to the network infrastructure are generally less
expensive and less disruptive than environments where a more constraining centralized architecture requires infrastructure change and permission, as explained in Chapters 2, 3, and 4.
A S S U M P T I O N 5 For some services there exist business and technical advantages (BTA) that push providers to offer services that are more centrally managed.
Services with the property of Assumption 5 are interesting because high market uncertainty implies that distributed management will produce services with the most value, even given that it will be more expensive to provide these services because of the inefficiencies of distributed manage- ment. If the market uncertainty for these services is zero, then a centralized management structure will work best.
Assumption 5 focuses on services that have advantages to a centralized management structure. If centralized management has no advantage, then it is never a good choice for a particular service. The choice of best man- agement structure for services interesting to us is not clear because of the trade-offs between the efficiency of centralized management and the bene- fit of more experimentation. With services like this, the situation is counter- intuitive because services that are managed inefficiently do better in the market than services with a more efficient management structure. This book concentrates on situations such as this where the arguments support- ing centralized management are strong, yet distributed management is a better choice.
Next is a discussion considering under what conditions the preceding advantage of experimentation and choice is not enough to outweigh the inefficiencies of managing a distributed service.
Theories about How to Manage Network-Based Services 99
T H E O RY 2 If high market uncertainty causes the difference between the expected value of the best of nexperiments and the expected value of each individual experiment to exceed the business and technical advantages of the centralized management structure, then a service provider should consider providing this service with a more distributed managed architecture. When market uncertainty is low enough that the advantage of having nchoices is less than the business and technical advantages of a more centrally managed service, then providing the service with centralized management architecture makes the most sense.
Theory 2 demonstrates the value of a management structure that pro- motes experimentation by users compared to that of a management struc- ture that allows experimentation only by centralized authorities. One way to view Theory 2 is to compare the value of many users experimenting to the value of one central authority undertaking a single experiment. It is market uncertainty that determines which has the most value: the effi- ciency of centralized management when market uncertainty is low or the flexibility of distributed management when market uncertainty is greater.
So far this theory looks at a single generation of a service, yet services evolve over time. Each generation has many attempts (experiments) to provide a service offering with a good market match. Thus, each service generation is composed of many service instances from simultaneous experimentation (that is, a group of services), which are the efforts of many different contributors. This theory incorporates service providers learning from the previous generation of experiments, thus reducing the market uncertainty from generation to generation.
A S S U M P T I O N 6 Those experimenting and providing services learn from experience, causing a decrease in market uncertainty.
T H E O RY 3 If market uncertainty is decreasing, then service providers are likely to succeed at providing a service with centralized management when the advantage of market uncertainty and parallel experimentation no longer outweighs the business and technical advantages (BTA) of the centralalized management architecture.
Market uncertainty can also increase if technology changes, as Clark [2]
points out, leading to the following:
100 Chapter 7
A S S U M P T I O N 7 Technology changes and alters the space of what services are possible, which causes market uncertainty to increase. One example of this occurred when PBXs became computerized — the whole scope of possible features changed, as discussed in Chapter 9.
T H E O RY 4 If market uncertainty is increasing, then service providers are likely to succeed at providing a service with distributed management when the advantage of market uncertainty and parallel experimentation outweighs BTA.
Theories 3 and 4 imply that market uncertainty is catalytic in the migra- tion of users from one management structure to another. When market uncertainty is increasing, then the ability to experiment, which distributed management allows, becomes more valuable than any advantages to the centralized management structure. A reduction in market uncertainty, though, causes the shift of management structure in a network service from distributed to centralized. Reducing market uncertainty implies that the business and technical advantages of centralized management become more important than the benefit of increased experimentation from dis- tributed management.
This theory is fundamental to understanding how to design the infra- structure used to build services based not only on business and technical advantages, but also on market uncertainty. It provides a framework to ana- lyze the value of network infrastructure in the context of the type of man- agement structure this infrastructure allows. It illustrates the trade-offs between centralized and distributed management structures with respect to market uncertainty, the number of experimental attempts to provide the service, how many generations the service evolves for, and the advantage of centrally managing the service. It shows that when a centrally managed ser- vice has an advantage from a business and/or technical perspective, the market for the service may still be better met with services that have less central management but allow more innovative features because of experi- mentation. It illustrates the value of flexibility in network infrastructure with respect to the type of management structure services can have.