I have been thinking quite a bit recently about how to make services not only dynamically discoverable, but also dynamically usable. I think the first part is done pretty easily via a registration service. Each new service that comes online is able to send information about itself to the registration service. That way, each other service can also subscribe to the registration service and become aware of new services that have come online.

So basically we need to create a dynamic services catalog that other services in the network can become aware of and know about. Getting services to be aware of each other is only the first step; we then need to ensure that the services can be used on demand. To me this involves the following elements:

1. What protocol is used to communicate with the service?

Does the service require HTTP, CORBA, MQ, JDBC, File, etc.?

2. What are required and optional data elements and how do we validate them?

I would also tend to define what validation is done here. Do we just require a field to have data, i.e. not  null, or do we validate that the data matches a specific format like a social security number, a zip code, or a phone number with or without a country code.

3. What other if any dependencies with additional services are required?

For example if you want to use a create account service is that dependent on an address validation and a credit check service. Basically we would know what the services hierarchy looks like.

4. What is the required format of the messages?

Are we using XML message, a specific proprietary format, or something else?

5. What does the response look like?

Would the calling service be able to handle the response from the new service? Like the requirements for talking to the service, what does the response look like, what is the schema, and how is the message formatted.

6. How does a service acquire the required security credentials to utilize the new service?

This is critical as we need to ensure that services are truly authorized to use the new functionality.

What brought this to mind is getting the best utilization out of a service catalog. If we auto populate the catalog we are still dependent on humans to determine if a service can be utilized or not. This creates several problems.

1. The person must make a determination if the service is applicable or not. If there is any confusion based on the name or description of the service this can lead to error.
2. The interface has to be understood by the user. This includes all of the elements above including what they need to do to utilize the service.
3. A value must be seen by the user to utilize the service. I believe that users will tend to discount a service based on their own personal views rather than evaluate it dispassionately.

There are several methods to make the proposal work. I am still working out the best way to make service discovery work. We could use a registration service, or possibly use something like dns to expedite the discovery of new services. I am interested in comments.