Coherence Evaluation help

Question

I'm currently undergoing a technical evaluation of Coherence and am finding a number of problems, the most pressing of which is that the documentation is not wholly helpful as it doesn't explain some basics about the software (at least, I can't find the information I want).
As part of the evaluation, I am attempting to implement a distributed cache of product data which can be read from a number of different applications.
PRICING:
It is not clear to me that if an application is looking up data in the cache, whether Coherence's pricing model means that we must pay for the CPU of that application (i.e. applications could be thought of as running inside a Coherence container which must be paid for). The alternative scenario is that the data resides on a number of Coherence servers (for which we must pay) but that any number of clients can access the cache without paying for every single client.
We have a (I would have thought) modest number of internal applications which want to access data on 6 blades, each of which has 2 quad-core processors. With the enterprise edition, the "Container" model would cost us $240,000 for the Enterprise edition (assuming a quad-core processor = 2CPUs) which is a mind-boggling amount for what is (I would have thought) a small deployment. From my experience thus far, however, Coherence would seem to follow this model
BASIC QUERYING:
I have implemented a basic cache of product data and have encountered a problem with terrible querying performance using ValueExtractors. Basically a pretty basic non-key-based lookup is taking around 100ms (this is with 2 nodes on the same machine), even if I repeatedly lookup data using the same extractor and even when my data is not being written to. Why is this? Am I missing some configuration to use Smart-caching of query results. Surely the client API should be aware of whether the data has been changed (jn another node) and therefore whether it can just run the query of the data in the local VM? An ID-based lookup is sub-millisecond (but only just) by comparison and taking the cache's size takes about 3ms.
Of course, this could be the serialization overhead of the ValueExtractor but that would mean that querying for individual items using an extractor is effectively unusable.
I'm not dissing Coherence here as I'm sure it's a fantastic product and that it's just the case that I can't find the information I need to answer my questions
Edited by: user10653192 on 26-Nov-2008 03:31
Edited by: user10653192 on 26-Nov-2008 03:31

Answer 1

robvarga wrote:
In case your query depends only on the content of the cache, but no outside or variable factors, like time, then Coherence provides the Continuous Query Cache which reevaluates changed entries for being included or evicted from query results and practically maintains an up-to-date query result on the local node.
The trouble here is that I am maintaining a cache of (financial) product data and wish to be able to query for individual products by any one of a number of "code types" individually - e.g. ISIN, RIC, TICKER etc. Keeping alive a continuous query to perform what is essentially cache.lookup(CodeType.RIC, "JAVA.O") would not be an option because an application may need to query tens of thousands of products.
With indexing, I have got the query down to around 3ms but this really not the sort of performance I was looking for. I would have to build a caching layer on top of Coherence (i.e. on top of a cache)!
And I expect you wanted to write Coherence not Confluence :-)Oops :-)
Also, I found Coherence pretty easy to evaluate even 3 years ago and it only became easier since then due to the abundant information in the Wiki pages.I have been plagued by appalling page load times, various error pages on the Oracle website. From the Oracle website, the documentation is difficult to find; the docs bundled with the install are much more helpful. That said, the docs for GigaSpaces are better laid out and better structured from an evaluation point of view. (GigaSpaces seems to fundamentally lack a feature that we need by the way - I am not "bigging" up the GigaSpaces product!)