OMII-SAGA: a simple programming interface for e-Researchers

Next article: OMII-Europe’s ‘bridges’ help the fight against malaria

Previous article: Cancer experts choose Taverna

Marshalling heterogeneous distributed resources at the application level can be very challenging. In fact, there is a critical need for application-level abstractions that provide the functionality needed to access and coordinate distributed computational resources. This need is fulfilled by SAGA (Simple API for Grid Applications) – an OGF approved technical specification, which provides a simple, application-level interface to heterogeneous computational infrastructure.

The defining guideline of the SAGA API specification is simplicity. The complexity of the underlying infrastructure is hidden from the user. For example, the code to submit a remote job, or to access/stage a remote file across a distributed system, could require a couple of hundred lines of code. SAGA requires only a few lines:

#include <saga.hpp>
int main (int argc, char** argv)
{
saga::filesystem::file f (saga::url argv[1]));
f.copy (saga::url (argv[2]));
}

Even more attractive is the fact that the same code will achieve the same effect across different middleware distributions. The complexity associated with working across different distributions does not disappear magically! It is removed from the user space and addressed in the adaptor and engine space, which is only of concern to SAGA developers, deployers and infrastructure providers.

OMII-UK is funding an international effort to implement the SAGA by making it usable with OMII-UK's software and other commonly used middleware distributions, such as Globus. The OMII-SAGA project will provide application and tool-kit developers with a simple and stable programmatic interface for their development efforts.

As shown in the figure, SAGA middleware plugins ('adaptors') permit stable SAGA API functionality using a specific middleware, such as GridSAM. This allows SAGA developers and system administrators to adjust the SAGA implementation to the specific middleware present. The test suite ensures that the semantics of the SAGA API are correctly mapped to the underlying middleware, and that the user is untroubled by middleware details.

OMII-SAGA will provide both a C++ and a Java implementation of SAGA, which along with a Python wrap around, the C++ development and the wide range of adaptors, should allow the majority of UK e-Research users to tap into the power of SAGA.

Is it an Application or is it a Tool?

SAGA will facilitate the development of new e-Research applications, as well as making it easier for existing applications to use distributed infrastructure. The SAGA team, which comprises research-domain specialists, has developed sensor-based adaptive applications and an ever-growing list of distributed applications.

However, one user’s tool is another’s application. By understanding the needs of the users and researchers, the OMII-SAGA project will develop application managers that will provide end-to-end requirements. The first of these will be the NAMDmanager—a lightweight simulation manager that uses SAGA Python bindings to farm-out and monitor NAMD simulations to a heterogeneous distributed environment. The NAMDManager makes intrinsically difficult and error-prone tasks easy and transparent. All of the available middleware adaptors and SAGA's sophisticated error-handling and dynamic runtime adaptor selection mechanisms are transparently available to the application. Thanks to the coupling with Plone/Zope it will be a trivial endeavour to provide a web interface to the NAMDManager.

Shantenu Jha, SAGA Research Group.

http://saga.cct.lsu.edu