[infinispan-dev] Cloud-TM & Infinispan

[Manik Surtani]

Greetings and welcome to this list, Paolo.  :)

As you said your starting point is looking at the replication mechanisms.  We discussed the current 2-phase scheme in detail when I was in Lisbon, and I am very keen on an alternate atomic broadcast style approach.  You presented a few different approaches even within the broader atomic broadcast umbrella, so it makes sense to make this layer pluggable so we can work with different implementations.

Have you had a look at the existing 2-phase scheme to see how an alternate scheme can fit in, and where we'd need to introduce layers of abstraction?

Cheers
Manik


On 3 Oct 2010, at 19:18, Paolo Romano wrote:

> Hi all,
> 
> I am new here, so let me first introduce myself. I am Paolo Romano, a 
> researcher working at INESC-ID Lisbon, you can find more about me and my 
> research activities at my webpage: http://www.gsd.inesc-id.pt/~romanop.
> 
> I am posting to this mailing list to introduce the Cloud-TM project 
> (http://www.cloudtm.eu), a EU funded project started in June which 
> brings together Red Hat, INESC-ID Lisbon (http://www.gsd.inesc-id.pt), 
> Rome University "La Sapienza" (http://www.dis.uniroma1.it/~hpdcs) and 
> Algorithmica (http://www.algorithmica.it).
> 
> Citing the project's abstract:
> "Cloud-TM aims at defining a novel programming paradigm to facilitate 
> the development and administration of cloud applications. It will 
> develop a Self-Optimizing Distributed Transactional Memory middleware 
> that will spare programmers from the burden of coding for distribution, 
> persistence and fault-tolerance, letting them focus on delivering 
> differentiating business value. Further, the Cloud-TM platform aims at 
> minimizing the operational costs of cloud applications, pursuing optimal 
> efficiency via autonomic resource provisioning and pervasive self-tuning 
> schemes."
> 
> Infinispan is expected to play a key role in Cloud-TM, as it has been 
> chosen as the reference platform to integrate the main research results 
> achieved during the project.  Specifically, our plan is to extend 
> Infinispan along the following main directions:
> 1. Build a library of alternative replication mechanisms optimized for 
> different workload scenarios (e.g. hi/low conflict rate, read/write 
> intensive) and scales of the platform (e.g. few/many nodes, 
> local/geographical distribution)
> 2. Developing self-scaling mechanisms aimed at elastically allocating 
> nodes from Cloud computing platforms to Infinispan caches depending on 
> the current workload.
> 3. Developing self-tuning mechanisms that will adaptively alter the data 
> replication and distribution algorithms depending on the current 
> workload characteristics and scale of the platform.
> 4. Providing programmers with a Distributed Software Transactional 
> Memory interface via a wrapper over Infinispan. This wrapper would be 
> close  in spirit to what  PojoCache is for TreeCache, though we are 
> currently oriented towards using a Domain Modelling Language and a 
> precompilation phase to generate the code to interact with Infinispan 
> (along the lines of what is done in the Fenix framework, 
> https://fenix-ashes.ist.utl.pt/trac/fenix-framework). Note that we are 
> still at very early design phase, so we are open to ideas, comments and 
> especially to learn from your experiences with PojoCache.
> 
> As developers of Infinispan, your feedback is extremely valuable to us. 
> On one hand, as nobody better than you could provide us indications on 
> how to fit within Infinispan's codebase any new experimental feature we 
> will be developing in the least intrusive fashion. On the other hand, as 
> you can help us to identify what are the most critical issues for 
> realistic deployments of Infinispan in Cloud environments, pointing out, 
> for instance, which ones, among the current Infinispan 
> paramers/functionalities, would benefit the most from self-tuning 
> approaches.
> 
> We have already started looking at the internal structure of the 
> replication's modules of Infinispan, and in the next days we will be 
> posting more about the kind of replication schemes (see point 1 above) 
> we would like to integrate in Infinispan, and how we are planning to do so.
> In the meanwhile, as a teaser :-), I am sending a reference to a couple 
> of recent papers of ours if you are curious to know what kind of 
> replication solutions we are currently working on:
> - http://www.gsd.inesc-id.pt/~romanop/files/papers/prdc09.pdf
> - http://www.gsd.inesc-id.pt/~romanop/files/papers/middleware10.pdf
> 
> Cheers,
> 
>    Paolo
> 
> -- 
> 
> Paolo Romano, PhD
> Researcher at INESC-ID
> Rua Alves Redol, 9
> 1000-059, Lisbon Portugal
> Tel. + 351 21 3100300
> Fax  + 351 21 3145843
> Webpage http://www.gsd.inesc-id.pt/~romanop
> _______________________________________________
> infinispan-dev mailing list
> infinispan-dev at lists.jboss.org
> https://lists.jboss.org/mailman/listinfo/infinispan-dev

--
Manik Surtani
manik at jboss.org
Lead, Infinispan
Lead, JBoss Cache
http://www.infinispan.org
http://www.jbosscache.org