[infinispan-dev] Improving the performance of index writers

[Emmanuel Bernard]

HSEARCH-1699 looks good. A few comments.

Maybe from a user point of you we want to expose the number of ms the user is ok to delay a commit due to indexing. Which would mean that you can wait up to that number before calling it a day and emptying the queue. The big question I have which you elude too is whether this mechanism should have some kind of back pressure mechanism by also caping the queue size.

BTW,  in the following paragraph, either you lost me or you are talking non sense:

> Systems with an high degree of parallelism will benefit from this, and the performance should converge to the performance you would have without every doing a commit; however if the frequency of commits is apparoching to zero, it also means that the average latency of each operation will get significantly higher. Still, in such situations assuming we are for example stacking up a million changesets between each commit, that implies this solution would be approximately a million times faster than the existing design (A million would not be realistic of course as it implies a million of parallel requests).

I think you can only converge to an average of 1/2 * (commit + configured delay time) latency wise. I am assuming latency is what  people are interested in, not the average CPU / memory load of indexing.

Emmanuel

On 17 Oct 2014, at 20:15, Sanne Grinovero <sanne at infinispan.org> wrote:

> Hi all,
> we have been breaking down the problem of latency during Index
> Writing into smaller manageable tasks, you can find the general
> overview JIRA here :
> 
> - https://issues.jboss.org/browse/ISPN-4847
> 
> As you can see some minor improvements have been fixed already, and
> while each of them provides only minor 10% to 30% improvements, some
> provide more and combined the composite ratio is getting interesting.
> 
> While these minor issues (even combined) won't give us the many orders
> of magnitude performance improvements that we'd like to see, they are
> important as they are paving the road to the more significant
> efficiency improvements.
> 
> I documented the main idea here, as it belongs into the Hibernate Search engine:
> 
> https://hibernate.atlassian.net/browse/HSEARCH-1699
> 
> I don't expect that to be implemented overnight, but Gustavo already
> sent a PR for the ASYNC case, which is based on the same principle of
> avoiding the commits but is simpler to implement:
> 
>  https://hibernate.atlassian.net/browse/HSEARCH-1693
> 
> We expect this one to be a proof of concept for the performance that
> we'll get from HSEARCH-1699, and also I think it's very useful on its
> own: previously users of ASYNC indexing were forced into a "very
> async" architecture which might have been a bit too hard to manage,
> while now being able to set a maximum delay for the async operation I
> also expect that to be an acceptable compromise for a much wider range
> of use cases.
> 
> Essentially this will decouple the achievable throughput of indexed
> caches from the RPC latency, although obviously this latency will
> still be the limiting factor for some dimensions, especially the
> response time for a single synchronous indexed write will still be
> affected primarily by the ability of Infinispan to improve the number
> of blocking RPCs needed for a single write.
> 
> Feedback very welcome!
> 
> Sanne
> _______________________________________________
> infinispan-dev mailing list
> infinispan-dev at lists.jboss.org
> https://lists.jboss.org/mailman/listinfo/infinispan-dev