Eclipse Community Forums: EMF » [EMF Compare] Performance comparison of different Similarity Checkers

Help

Home

Home » Modeling » EMF » [EMF Compare] Performance comparison of different Similarity Checkers

Show: Today's Messages :: Show Polls :: Message Navigator

[EMF Compare] Performance comparison of different Similarity Checkers [message #754011]

Mon, 31 October 2011 13:09

Tatiana Fesenko

Messages: 62
Registered: July 2009

Member

Hi,

I am wondering if somebody has done performance comparison of different
Similarity Checkers in EMF Compare. I am particularly interested in the
performance of StatisticBasedSimilarityChecker in comparison to using
XMIIDSimilarityChecker.

Thanks,
--
Tatiana Fesenko
ModelSolv, Inc.
Model-Oriented Solutions for Service Integration

Report message to a moderator

Re: [EMF Compare] Performance comparison of different Similarity Checkers [message #754760 is a reply to message #754011]

Fri, 04 November 2011 08:51

Laurent Goubet

Messages: 1902
Registered: July 2009

Senior Member

Tatiana,

We have never compared the two as they are not comparable : the XMI ID checked will only retrieve the ID of the elements and return either "they match" or "they don't" with a simple "equals". The statistics checker, on the other hand, needs to fully resolve the elements' contents and compute all similarities until it determines a match. The statistics checker is expected to be way longer than the ID checker for all models that _do_ use IDs. For all models that do not use IDs however, the ID checker will delegate to the statistics one ... and that will amount to the same time performance.

For the record, I just did one comparison with the "nominal" model I described for http://eclipsemde.blogspot.com/2011/10/emf-compare-scalability.html . That model took 22 seconds to compare with the ID checker ... it takes 1 minute and 50 seconds with the statistics checker (plus, the result is likely to be less accurate).

Laurent Goubet
Obeo

Report message to a moderator

Re: [EMF Compare] Performance comparison of different Similarity Checkers [message #755663 is a reply to message #754760]

Wed, 09 November 2011 09:07

Tatiana Fesenko

Messages: 62
Registered: July 2009

Member

Hi Laurent,

On 11/4/11 9:51 AM, Laurent Goubet wrote:
> For the record, I just did one comparison with the "nominal" model I
> described for
> http://eclipsemde.blogspot.com/2011/10/emf-compare-scalability.html .
> That model took 22 seconds to compare with the ID checker ... it takes 1
> minute and 50 seconds with the statistics checker (plus, the result is
> likely to be less accurate).

I do understand that these checkers are used for different use cases. In
our case compared models are result of M2M transformations so elements'
XMI IDs do not really match. We were thinking of
1) setting calculated XMI IDs programmatically prior to model comparison
and compare the models with XMI Id checker.
or
2) Using statistic based similarity checker.

In your example statistics checker takes ~5 times more time than the Id
checker. So both approaches worth considering.

I talked to Cedric Brun at Eclipse Day Paris yesterday and he told me
that there is an experimental component called "Fingerprints".
Fingerprints will allow us to set different weights to properties to
during element matching. This is a very interesting and promising
approach. I hope the EMF Compare team will find time and resources in
Juno to develop it.

Best wishes,
Tanya

--
Tatiana Fesenko
ModelSolv, Inc.
Model-Oriented Solutions for Service Integration

Report message to a moderator

Previous Topic:	[EMF Compare] Support for dynamic EMF models: Indigo vs Helios
Next Topic:	EclipseCon North America

Goto Forum:

-=] Back to Top [=-

[ Syndicate this forum (XML) ] [

]

Current Time: Fri Apr 26 09:29:07 GMT 2024

.:: Contact :: Home ::.

Breadcrumbs

Sign up to our Newsletter