WCF Services, what's better, one big request or a lot of little ones?

Question

I'm reviewing some code where we've had some issues with return data from a WCF web service. Currently the service makes a list of objects, serializes them (as JSON for the record) and returns the entire serialized list down the wire. Obviously when there's a lot of data users run into quota limit problems.

I'm considering changing it so the service returns one item at a time which would send a bunch of requests on a loop adding one object at a time into the list until it was done.

Obviously in scenario one we're making one request to the service that has the potential to return a massive amount of data and run up against the quota. In the other scenario we never hit the quota but the requesting app will be requesting data item after data item in a stream of separate requests.

To illustrate we have a list of items which come in a variety of item types and those types come at a variety of price points. The app might want to aggregate a number of items, the customers who want that item, the types of item and price requested by the customer and their could be maybe seventy items with between five and eighty customers each requesting on average 2 types of product at 1 price each.

Taking averages at the extreme end this could make 7000 separate (very small) data requests in a single complete job. Is that a problem? It is possible to package it up a bit so that customer types and prices requested could be bundled but that's still potentially a couple of thousand requests at one time.

Am I better off with a single huge data stream? Or a couple of thousand smaller ones?

Answer 1

You're better off with the optimal sized return for your scenario :) It kinda depends on the overhead on the request. Generally the less chatter back and forth to a web service the better.

Facetious answer, so here's the rub: You're probably best off with some sort of paging system, wherein your request asks for a specific number of items, and your response returns a "n of m" in the results. That way you can tune the number of requests and size of the response to perform best in your situation.