Grid overview – dispelling the black art of grid

What is grid?

Grid computing is a model that improves the utilization of the infrastructure by using networked computers to perform parallel computing tasks which can be allocated automatically independent of the application or batch.
Grid sits between the application and the infrastructure. The application submits work to the grid, the grid then submits the processing/tasks to the infrastructure. This allows the application (and therefore the business) to be less dependent on the infrastructure, less sensitive to system outages.

So how is it typically being deployed?

The banks are using grid heavily in the risk and front office trading systems, whether its a way of increasing the batch time for your risk reports, or allowing risk/trading calculations to be submitted from the desktop to the grid for quicker calculation. The key aspect being the performance improvements can be significant, and if your getting quicker/more accurate risk reports, traders are better able to hedge their risk liabilities.

Grid comprises (In a Datasynapse configuration)
A director and broker server
Batch servers (or engines) typically blades – but they can be servers or desktops

So grid means?

Ability to use numerous cheaper computers rather than several expensive ones – all your desktops doing your risk reports overnight?
More resilient than a stand alone solution – one computer failure will mean a failed task, not necessarily a failed batch, grid will reallocate work if it detects the computer is down
Compatible with many operating systems – solaris, linux and windows

How do we get there?

The application needs to be reconfigured to allow parallel processing, in this case, on a basic level, the following happens:
Application submits job to the grid – process my market risk reports
Grid through the Director and Broker accepts the job and cuts it up into smaller tasks for processing.
Engines (blades/pcs/grid servers) accept the task, process it, and report back when complete
Grid accepts the completed tasks and then submits back to the application.

Every application is different as is the respective grid for that application, but in essence this is what we’re trying to achieve: more efficient use of the infrastructure, those idle systems get used, be less platform dependent and more importantly less infrastructure aligned, server 3 being down should not affect your application, grid should see it as faulty and schedule the work to another server.

This site uses Akismet to reduce spam. Learn how your comment data is processed.