When I implemented a R-based stats web service infrastructure (
http://dx.doi.org/10.1021/ci700188u and
http://www.chembiogrid.org/projects/proj_statistics.html) my approach was to
have an Rserve instance running somewhere (in our case it was on a remote
machine that stored our pre-built models). The web based application
(originally Axis based SOAP but I think IU has now provided more RESTful
wrappers or implementations) would then send queries to the Rserve instance.

As a result of this, we only needed a single Rserve instance running
independent of the frontend (which was a web page but could have been
anything), rather than multiple instances being started up. You can see the
code for this at
http://cicc-grid.svn.sourceforge.net/viewvc/cicc-grid/cicc-grid/rws/trunk/

More recently, in support of the Open Notebook Solubility Challenge, I
looked at an alternative approach based on mod_python based web
applications. In this case, I used the rpy2 Python package to interface to
R, to load models and get solubility predictions. A very nice API, and
somewhat simpler than my original Rserve based solution, but it does imply
that each time the app is run, an R instance is reloaded (I'm not sure that
I could share instances between sessions). See
http://github.com/rajarshi/rest-ws/tree/master/predict/ for code

One other approach, that I haven't explored yet is the use of rapache, which
is a DSO for Apache that runs R. In this approach, you'd avoid the start up
cost of R instances, since they'd already be running.Furthermore, you could
easily serve up pages directly from R code.

On Tue, Dec 22, 2009 at 9:35 AM, surajit ray <mr.surajit.ray at gmail.com>wrote:

> I am using JRI with the situation that the Java code (for prediction ) is a
> desktop aplication which is being initiated by the ./run script in rJava
> (within the JRI subfolder). This means that the R engine is being started
> by
> the script and then the java code is initiated.
>
> This is ok in working for the desktop application. But since we are going
> to
> develop an online service - I have to reprogram the code in JSP etc.
> However
> this means I cannot use the JRI run script (since the code has to be
> initiated by tomcat).
>
> So I would like to start the engine from within the Java Code (JSP). I have
> tried this - but this is not recommended by the author of JRI. My attempts
> have failed till now. I can forward the exact code and the details of the
> errors.
>
> Cheers
> Surajit
>
> On Tue, Dec 22, 2009 at 2:12 PM, Rajarshi Guha <rajarshi.guha at gmail.com
> >wrote:
>
> >
> > On Dec 22, 2009, at 4:32 AM, Egon Willighagen wrote:
> >
> >
> > >>> 5) Unable to make the R part of the code native to java as yet.
> > >>> Still
> > >>> trying a few things. This hinders our ability to provide the API
> > >>> interface.
> > >>>
> > >> I am aware of at least one Java - R integration, namely The CDK
> > >> library
> > >> (*http://cdk*.*sourceforge*.*net*/)  has R integration since few
> > >> years -
> > >> it might help if you consult the code there. And AFAIK the author of
> > >> CDK-R code is on this list as well, so please do ask specific
> > >> questions
> > >> on the development mailing list.
> > >
> > > Not sure if Rajarshi is on the list yet, but he indeed wrote various
> > > levels of integration, calling R from a Java environment is one.
> > >
> > > Have a look at http://www.omegahat.org/RSJava/, which has a section on
> > > 'Calling R from Java'.
> >
> > I'm here :)
> >
> > Indeed, RSJava is one way to make calls to R from Java. However, the
> > preferred solution is to us JRI (which is what the CDK-R bridge uses)
> > which has a much more robust interface
> >
> > ----------------------------------------------------
> > Rajarshi Guha        | NIH Chemical Genomics Center
> > http://www.rguha.net | http://ncgc.nih.gov
> > ----------------------------------------------------
> > A committee is a group that keeps the minutes and loses hours.
> >        -- Milton Berle
> >
> >
> > _______________________________________________
> > Development mailing list
> > Development at opentox.org
> > http://www.opentox.org/mailman/listinfo/development
> >
>
>
>
> --
> Surajit Ray
> _______________________________________________
> Development mailing list
> Development at opentox.org
> http://www.opentox.org/mailman/listinfo/development
>



-- 
Rajarshi Guha
NIH Chemical Genomics Center