On Jun 17, 2008, at 6:40 AM, Dietmar Wolz wrote:
Hallo,
my name is Dr. Dietmar Wolz, I am working at SOPERA GmbH for the
Eclipse Runtime SOA Framework in the project Swordfish.
At the OSGi Community Meeting in Berlin last week I
talked with Peter Kriens about one of the OSGI research challenges
for 2008, the "OSGi Resolver" problem, see
http://www.osgi.org/blog/2008/02/research-challenges-for-osgi.html .
I have some experience with similar kinds of problems so I would like
to investigate this a little further.
First I would like to collect any written information about this
problem,
if available.
My approach to these kind of problems is quite pragmatic, even for
generally NP-complete problems there can often be found good
solutions in reasonable time for the specific instances of the problem
which occur in practice.
I favor a test driven approach, so
my first aim would be to set up some kind of test environment which
"generates" randomly reasonably sized "typical" instances of the
problem.
Next I would like to formalize the parameters to optimize in terms
of test code. Last would be the test of different algorithms
designed to
solve the "resolver challenge". From the results we obtain it may be
possible to
derive a more theoretical paper, but this would be last on my priority
list.
For my first aim, the generation of test scenarios, it would be
helpful to
collect some typical instances, can anyone point me to existing
projects
where a good OSGi resolver would be of some value?
Next is the definition which needs exactly to be optimized, can
anyone provide me with more detailed information regarding this?
I am starting a project at the Apache Software Foundation, CAT-Scan.
It's project proposal can be found in http://cwiki.apache.org/confluence/display/FELIX/CAT-Scan+Project+Proposal
.
"All too often one hears anecdotal information about how many bundles
get loaded and when. There have been claims that there is a need to
handle tens of thousands of bundles at one time. Apache Geronimo, a
JEE server, fully trussed up, does not load that many jars. Attendees
at the 2008 OSGi conference didn't even know or were hesitant to
divulge their bundle usage.
"CAT-Scan is a way for developers to collect information on the traits
of an application or set of applications deployed on an OSGi
framework. This information can be used by developers, framework
implementors, researchers to obtain a clearer, more accurate, picture
of their OSGi usage rather than the anecdotal information one gets
from, say, conferences.
"Information that is collected can be used by the developer to
appropriately configure their OSGi frameworks. Snapshots can be sent
to OSGi vendors to assist them in support and help make framework
design decisions. Characteristics of applications collected from this
tool can be used by researchers to provide concrete data as to the
traits of real world applications. This data can be made anonymous to
protect installations trade secrets while still providing great value. "
I will be fleshing out the rest of the details through this week.
Please feel free to contact me, or if no one minds post on this list,
with comments. Once the proposal is complete, I intend to run it by
the Felix community to "incubate" the API. My own OSGi
implementation, Papoose, will support this. It is my hope that other
implementations will as well.
Regards,
Alan
_______________________________________________
OSGi Developer Mail List
[email protected]
https://mail.osgi.org/mailman/listinfo/osgi-dev
