Eclipse Community Forums
Forum Search:

Search      Help    Register    Login    Home
Home » Modeling » Epsilon » ETL equivalents performance
ETL equivalents performance [message #1058136] Sun, 12 May 2013 11:15 Go to next message
Joost van Pinxten is currently offline Joost van Pinxten
Messages: 50
Registered: November 2012
Member
I've been profiling my ETL tranformation to see why it does not perform as expected; A model containing about 12k elements takes 130 seconds to transform (and the executed rules are quite trivial). The target has roughly the same amount of elements.

The profiler shows that the rules, even there is no (explicit) call to 'allInstances' the execution time linearly increases with the number of executed rules. Looking at the code in the SVN for the EquivalentOperation, the DefaultTransformationStrategy calculates/transforms all the equivalents in a decent way, so my guess is that it might be related to copying some data structure that grows when more rules have been executed.

Is this behavior expected? Or is it possible that I missed an implicit operation that is dependent on the number of elements created in the target model?

FYI; I'm using a set of lazy, extended rules to perform transformations on expressions. Is it possible that all expressions are then combined into the same set/map, such that looking for the equivalent of the transformed expression is about O(N)?

Each transformation rule call starts at about 2ms, then increases up to 10ms after about 150 calls. With the full-blown transformation, starts out at 10ms and ends up at 44; so a factor 4x slower (per rule call!) is something to be looked into.

[Updated on: Sun, 12 May 2013 12:01]

Report message to a moderator

Re: ETL equivalents performance [message #1058239 is a reply to message #1058136] Mon, 13 May 2013 10:22 Go to previous message
Dimitris Kolovos is currently offline Dimitris Kolovos
Messages: 877
Registered: July 2009
Senior Member
Hi Joost,

Thanks for reporting this. Please file an enhancement request in the Bugzilla and I'll investigate. It would also speed things up if you could attach the relevant metamodels, sample models and the transformation.

https://bugs.eclipse.org/bugs/enter_bug.cgi?product=epsilon

Cheers,
Dimitris
Previous Topic:[Eugenia] Generating common super edit parts
Next Topic:Question about adjusting links
Goto Forum:
  


Current Time: Thu Oct 02 12:29:05 GMT 2014

Powered by FUDForum. Page generated in 0.10217 seconds
.:: Contact :: Home ::.

Powered by: FUDforum 3.0.2.
Copyright ©2001-2010 FUDforum Bulletin Board Software