In Part Two of APMdigest's exclusive interview, Matthew Ellis, IBM Vice President of Service Availability and Performance, discusses predictive analytics.
ME: Analytics is important to all phases of operations. In all areas of business it is axiomatic that more data enables better decisions, and operations and application management are no exceptions.
Just as important, however, is sorting that data to identify the critical context for decision makers to act on, and this is where analytics come in.
IBM is investing in analytics very seriously, and from an operations management perspective, we apply analytics in three categories: Simplify Operations Management, Avoid Business Disruption, and Enable Optimization.
Simplify Operations Management is a class of analytics technology that enables our customers to do the work that they do today more easily. This includes historical analysis of data to recommend and establish dynamic thresholds, and trending of performance and capacity data to identify areas that may become bottlenecks based on historical behavior.
Avoid Business Disruption is the key driver for the predictive analytics component. The goal is early identification of environmental changes that indicate a significant change in the behavior of an application or service, and to bring this information to the attention of the operations management team so that problems can be identified and addressed before they ever impact a customer. We have identified emerging problems days before traditional management tools saw signs of trouble and in some situations, discovered problems in unmonitored resources that were affecting the behavior of critical applications.
Enable Optimization is the ability to mine collected data across multiple dimensions enabling insight and optimization of services and applications by enabling rich insight. It is also known as business analytics.
ME: At IBM, we believe there are three key capabilities that any analytics solution must have to provide maximum predictive capability:
1. Algorithms: Multivariate Analytic techniques are critical to identifying emerging problems early, while all metric data is still well within their normal range.
The key to this statistical approach is to monitor the relationships of important related data metrics and raise an exception when the relationships of data change in significant ways. Any single metric displays a wide range of variability during a normal day, increasing and decreasing with changing workloads, and daily, weekly and seasonal behavior.
In general, however, related metrics will follow the same pattern all the time in a healthy system. Successfully identifying these relationships, and accurately determining when these relationships diverge in an important way is key to accurate early identification of problems.
Our algorithms are developed and refined by one of the largest private math departments in the world; the same organization that developed Watson to win at Jeopardy.
2. Scalability: Analytics solutions work better when they have more data upon which to base their conclusions. The IBM analytics solutions directly leverage proven data collection technologies that have been in use for most of a decade and have seen continual refinement. This capability is proven to be able to collect millions of data points per second, and deliver that data to the analytics engine with very low latency offering real-time evaluation of very large data streams. We believe that the data collection technology we are using is the most scalable and high performance in the industry.
3. Breadth of Monitored Resources: One of our design requirements was to deliver an easily extensible mediation capability allowing customers (or our services teams) to connect any data source to our data collection solution in a matter of hours or days.
During our pilot, we have worked with many products from non-IBM vendors and our team has found that almost all data integration work can be done in a very short time without ever requiring a visit to the customer site, saving time and money while maximizing data availability for analysis.
ME: IBM expects that analytics tools, and the organizations that use them, will evolve rapidly over the next few years. IBM is investing heavily in providing highly scalable, flexible, and robust systems for identifying emerging problems as early as possible.
We expect analytics to evolve along multiple dimensions:
1. Improvements in analytics learning and data exchange with existing application and service discovery, topology, and CMDB data to combine the strengths of traditional IT tools with analytics learning solutions. This will accelerate the statistical learning process and allow the learned relationships to be built back into the visible topology of the environment.
2. Apply analytics solutions to additional IT management domains to include Smarter Infrastructures, improved detection of security problems, asset management and maintenance scheduling and additional problems
3. Further improve feedback and integration of learning technologies, process optimization, and analytics in general with operations processes.
Matthew Ellis is the Vice President of Development for Tivoli's Service Availability & Performance Management product portfolio with IBM. This product suite enables monitoring and modeling the utilization, performance, capacity and energy-use of distributed, mainframe and virtualized platforms and associated application software. Ellis joined IBM in 2006 through the Micromuse acquisition, where he was the Vice President of Software Development.