Distributed High Performance Computing
CBRAIN is a flexible software platform with a small footprint and minimal requirements which allows us to integrate the often extremely heterogenous research High-Performance Computing (HPC) facilities across Canada and the World. Moreover, not limited to deploying a National and International distributed data and compute grid for our neuroimaging partners, the CBRAINengine drives an online collaborative web platform from which users remotely and transparently control data, compute and results, including various forms of 2D and 3D data visualisation, regardless of where these resources and data are.
CBRAIN approaches the issues of modern brain imaging research on two fronts. The first is to connect users to the tools and processing power required to handle large datasets that have become the norm in the field. The second is to reduce the amount of technical expertise required to perform computational analysis of said datasets. CBRAIN brings together data management, the advanced networking capabilities of CANARIE’s CAnet, analysis tools from groups from around the world, the processing power of 8 HPCs across Canada and Europe, and cutting-edge 2D and 3D real time visualization, ultimately making this integrated platform available to users through a standard web browser. A key challenge to this type of integration is the aforementioned heterogeneity of HPC environments. CBRAIN addresses this problem with its own meta-scheduler and data grid which interact with various HPC centers in a way that is transparent to the users.
Distributed Computing:
Our team has build a web-enabled platform which allows scientists to distribute their calculations to various compute facilities with the click of a button. We have currently integrated 6 Compute Canada HPC clusters, one German HPC cluster and 3 clusters local to McGill University Campus, totaling more than 80,000 potential CPU cores. Users have full control of their distributed jobs through the web interface, no matter where the HPC center is located and which scheduler it uses, thus greatly reducing the need for advanced IT knowledge to launch calculations.
Distributed Data:
One major issue with remote computing is getting your files where they need to be and getting the results back. We do not think this should be the responsibility of users. A scientist can choose files registered in his/her profile and perform computations without worrying about data transfer details, even when the files are in geographically distinct location. Furthermore, once computation is completed, the results are synchronized back to a user specified location. We have implemented a secure transfer layer which relies on caching and synchronization of data, minimizing unnecessary transfers.
Distributed Users:
In order to make the user adoption of our platform easy in any kind of environment, we focus on the most common client of all: the web-browser. Whether users perform file & project organization, launching and monitoring of HPC jobs, simple results visualisation or even 3D brain model exploration, everything is done within their web-browser. Privileges and rights management are applied through the concept of Virtual Organizations. All system resources, from file to compute tools, file storage to cluster access is managed in such a way that users only see what they have the right to. CBRAIN allows scientists from around the World to share data, tools and compute resources.
