Message from the CTO: The Future of Cloud

Author:  David Jackson, CTO, Adaptive Computing
David has been a pioneer of HPC, grid, and cloud solutions. He was just awarded his 80th patent on cloud management, cloud bursting, and cloud brokering.

Cloud is everywhere…It is pervasive and it is rapidly changing both our professional and personal lives. For IT professionals, instant access to remote data, services, and compute power alters how we program and how we architect solutions.

Before cloud, accessible compute capacity was static. IT decision makers would anticipate future needs and size compute infrastructure accordingly. These decisions would be made months or even years in advance based on rough estimates and highly subjective details. Information was poor and bad decisions could prove costly.

Enter the cloud. With cloud computing, short-term resourcing decisions could be made, with information that was far more up-to-date and estimates of needs that could be far more accurate. Although cloud resources were expensive, decision makers could scale resources up or down according to changing needs and could determine the optimal balance of on-premise and off-premise resources. Risk was removed and expert staff with appropriate financial models could allow tuning of cloud allocations so as to minimize cost and maximize SLA delivery.

Of course, technology never stands still. The unstoppable force of automation dictates that anything an expert does will eventually be replaced with an algorithm. Already, solutions exist in which data center and HPC management software is able to monitor current workload backlog and dynamically bring additional cloud resources online as needed. This method can be far more efficient and agile than having dedicated staff decide once a month or once a quarter whether or not to allocate another 100 servers from the cloud. These automated systems can monitor workload moment by moment, and respond in seconds, allocating exactly what is needed, exactly when it is needed. Just as valuable, these systems can also release cloud resources the instant they are no longer needed.

These management systems can track every job and service, creating dynamically customized cloud allocations to meet the specific SLA’s of particular groups or the specific compute needs of specific applications. Such micromanagement and moment to moment rebalancing would overwhelm even the best human operator with facts, details, and considerations, but is not a problem to an intelligent workload management system.

It is interesting that this great advance in technology forces a secondary advancement. Previously, workload management systems operated within set bounds, controlling how existing resources are utilized. Humans determined capacity, and software determined how it was used. But now, software can determine capacity and thus, through the cloud, software can spend money. But, if software is spending money, it must understand costs. Further, any decision involving costs must also understand the impact of that decision on value.

This makes a fundamental change. Management software no longer operates within a fixed capacity. Rather, it must broaden its view beyond the question, “How do I utilize what I have?” to “What more could I do and is it worth it?” This implies that completing workload and applications have value and servicing these workloads have cost. What is the value of running a computation one day earlier? Of a researcher being 15% more efficient? Of a drug coming to market 3 months ahead of the competition? Of a Hollywood movie rendering in time for its release date?

On the flip side, what is the cost of staging data into the cloud? What is the cost of on-premise power per application? Or the cost of applications running slower in a non-optimized cloud environment? Or the cost of increased latencies associated with remote services? How do users, admins, and CIOs express the value and model these costs?

Answering these questions is not easy. There is no clear model representing value of time to delivery and there is no clear model of opportunity cost.   It is difficult for experts to model these and even a larger challenge to automate this. Still, this is the direction the market is heading. For now, it is baby steps as resourcing costing and workload value become better defined. With this, automated cloud bursting systems will mature, becoming more efficient and more business savvy. While we are not there yet, the time is now to start thinking about what you want to have happen when you hand your corporate credit card over to your newly installed management software.