Analytics: Infrastructure Matters
John Easton 12000062GC JKJ@uk.ibm.com | 2012-07-20 07:42:59.0 | Tags:  workload optimized-systems analytics business-analytics system-z infrastructure | 0 Comments | 4,313 Visits
Something has been bothering me for a number of weeks now, and it's just come into sharp focus.
I've recently been working with some colleagues to address some system performance challenges at a client. I have no idea who designed the infrastructure that they are currently using, but it's definitely a case of not properly thinking through the infrastructure implications for their analytical applications.
My frustration is this: What is it about “analytics” that makes people think that any old hardware will do?
I've spent more years than I care to recall designing and building systems that have stringent performance and availability requirements. Consequently, we've always used appropriate hardware for the job: incorporating the correct choice of computational, storage and network elements early in the solution design process.
Since starting a new role at the start of the year, building the next generation of infrastructures to support analytical workloads, I'm increasingly finding that common “analytical” sense seems to have been forgotten by the people designing and building these systems.
Maybe it's because many of the organisations building these solutions don't have a heritage in complex or high performance systems design, or maybe they are developing applications without the knowledge and experience of anything more than commodity hardware?
I really don't know, but something is certainly amiss.
Anyway, back to my troubles. At some point in time in the dim, distant past this system might have worked. However, given the volumes of data that are now passing through it, and the requirements of the business to update their analytical environment more frequently than before, the current system plainly isn't able to do what it needs to.
Put simply, to move and process this volume of data, in the time period that the business requires, is physically impossible given the hardware infrastructure available to us.
We're not just talking about a bit of tuning or configuration changes to solve this. As the legendary Mr. Scott of Star Trek fame would put it, “Ye cannae change the laws of physics!” It is physics that we are up against here.
Yes, the workload requirements have grown along with the business. What the client is doing with it now has changed too. But why wasn't the ability to support growth and business change also factored into the design?
I've always had the belief that infrastructure matters.
For analytical workloads, this is plainly still the case. Scalability, in this era of big data, matters more than ever. Smarter analytics systems need to be as resilient as any other core enterprise system, and built from workload-optimised infrastructure components. All this implies a deliberate IT architectural approach to reap the benefits and drive business outcomes.
I ask you… How do we change this perceived wisdom?
For more information:
· Read the advisory that gives guidance on designing an integrated business analytics environment with high performance, resiliency, and a scalable growth path.