Speaker
Dr
Aleš Křenek
(Masaryk University)
Description
Web portals became a popular interface to complicated workflows of scientific
computation in many areas due to their ability to shield the user from the
complexity of the workflow implementation and control of the used computing
resources.
On the other hand, implementation of such web portal itself involves
rather complicated technology, which becomes difficult to operate and maintain
as the number of its users grows. Moreover, user requirements may become
contradictory (e.g. stick with an old version of simulation software to retain
full backward compatibility vs. use a new version to leverage new features).
Cloud infrastructures available nowadays, together with cloud orchestration
engines allow addressing those issues in a scalable way. Rather than
maintaining "one size fits all" instance of an application web portal, we
provide templates for deployment multiple instances of a specific portal,
dedicated to smaller user groups. Such instances may vary in size, and they
can be easily customized to the needs of their users, while still reusing 99%
of the common code and configuration.
We describe in detail the solution adopted in the EU H2020 West-life project.
The current implementation uses Cloudify orchestrator, extended with OCCI
plugin, to spawn and to keep alive the set of virtual machines forming the
portal in the environment of EGI Federated Cloud. Deliberately, minimalistic VM
images containing the bare operating system only are used; all required
software is installed and configured with Puppet system management upon VM
instantiation. In this way, tedious management of huge VM images, which would
require frequent updates and which would be complicated to distribute
otherwise, is avoided.
We demonstrate the approach with a prototype of web portal on top of the
Gromacs molecular dynamics package. The original version of the portal was
developed in the WeNMR EU project, it is based on Gromacs version 4.5, and it
uses gLite workload management system to submit payload jobs to the EGI grid
resources. Besides wrapping the portal to the virtualized cloud environment we
adapted it to use the more recent version of Gromacs (5.1), to leverage GPU
acceleration when the hardware is available. Submission to the grid was
replaced with use of local batch system (Torque) which is sufficient to manage
resources at the size expected for any instance of such virtualized portal,
while being less complicated, self-contained, and more reliable.
We also outline plans for future work -- adopting the more advanced cloud
orchestrator developed in EU Indigo-Datacloud project; in particular, we will
focus on flexible sizing of the portal instances according to their current
load.
Primary author
Dr
Aleš Křenek
(Masaryk University)
