Newbie help with cache design/custom filter

Question

Hello,
I'm very new to Coherence, so you'll have to forgive me if these are easy questions. I have an application that contains 1.5 million products. Each product has a map of prices for our client list. So internally, each product object I'm caching contains a pricelist map with up to 100 entries (clients):
private String productId ;
private String productName ;
private Map<String, Integer> priceList; //where the String is the client id and the Integer is the price for that clientHere's my question(s):
1. Is there any type of query that would return the product info and just the price for the 1 client (there really is no need to send all 100 of them for the individual quries)? How could this be done efficiently?
2. I need to sort by price. To do so, I've created a filter on the product keys, and a comparator that contains the client ID. This works fine, but it seems to take a long time to sort based on price using the comparator. Is there any other method of sorting that might be more efficient on the Coherence side of things? Sorting by product name is also a bit slow - is there a way to speed that up? The product name is the same for all clients, and thus is a simple string in the object being cached.
3. Indexing: Is there any way to index the internal Map of the price list? And would that help in performance?
Since this is a new implementation, the object design for the cache is open for change as well - if there are other structures that would enhance the performance. Any help would be most appreciated!
Sincerely,
Chris

Answer 1

There are actually techniques you could use to keep all the information about an article in the same object. This may be attractive in particular if the information is rarely updated (frequently updating a large object is costly and, unless every attribute changes with every update, breaking up the object graph making some of the objects into separate cache items is then preferable).
With the later Coherence releases there are for instance PofExtractor that allow you to extract selected parts of an object without de-serializing the whole object. You could probably use this to only extract the "price list"for a product. Combine this with a custom aggregator (that aggregates over all products a customer want to get a quote on) that could extract just that customers price from the "price list" and you would have a quite efficient solution...
As for indexes you COULD put the price list and other info from the large object in indexes but be aware that Coherence indexes are always stored on heap (other data can nowadays be stored off-heap) and also occupy a lot of space (Coherence always creates both index and reverse index and since Java objects are used to implement them there is quite a lot of overhead)... I would first look into either a 3NF-like solution with mainly primary indexes or a single cacnhe solution using PofExtractors etc...
/Magnus