IBM has patented two techniques that use analytics to improve cloud performance and efficiency in the data center. These patents probably won’t be put into practical use for at least a few years, but their development is noteworthy. Essentially, they are part of IBM's plan to stay on top of data center performance demands and needs as the Internet of Things gets into full swing.
"What we are doing is looking at trends we expect to see in the cloud market five years from now," Frank De Gilio, IBM STG's Chief Cloud Architect, told CMSWire.
These trends include massive amounts of data that will require processing -- albeit not necessarily on a time-sensitive schedule. Another trend IBM has its eye on is the rise of a "spot" market, so to speak, for data center services, with users shopping around for the best price at that particular point of the day.
One patent creates an "express lane" for "simple" data to be quickly analyzed and processed, while setting aside more complex data that may require additional processing. The other optimizes cloud performance by dynamically managing workloads between or within cloud data centers in response to availability and cost.
Data Center Express Lanes
Using U.S Patent #8,639,809, "Predictive Removal Of Runtime Data Using Attribute Characterizing," a data center can more effectively analyze data that comes in a from a variety of sources with an eye to avoiding performance inefficiencies and processing delays. De Gilio likened it to an express checkout lane at the supermarket where people line up to pay for a few items. Easy in, easy out, in other words.
"When data comes in it is sometimes hard to process it because there are added complexities," he said.
Typical data might be online transactions, financial quotes, video streams, and increasingly reading from sensors. "This patent allows the data center to attach the metadata to the stream and then manage those incoming streams in ways that are extremely efficient," he said.
It does this by identifying patterns in the data values that have correlated with slower processing in the past. The beauty of the system is that with the "easier" data separated out it can be processed first to avoid getting caught in a bottleneck behind the more complex data, De Gilio continued, mixing in another metaphor for illustration.
In tech-speak, the center automatically channels each piece of data -- called a tuple -- into the right analysis path. Tuples with values known to be problematic or time consuming to analyze are sent to specific analysis paths.
Such applications of the patent might include processing data from sensor-based highway toll collection systems, where some license plates images may be harder to analyze than others.
Another example might be an IoT application that monitors air quality in a clean room, De Gilio said.
"We can assume there are sensors placed every few feet. Now, let's say we notice a flutter in one area and the system determines that additional analysis needs to be made of four specific sensors and the area they are monitoring," he said. That data would be sent to a different analysis path.
The Upcoming Cloud 'Spot' Market
IBM's second patent, U.S. Patent #8,676,981 B2 "Routing Service Requests Based on Lowest Actual Cost within a Federated Virtual Service Cloud" uses analytics to increase performance and reduce costs by dynamically moving workloads -- between or within cloud data centers -- based on what is the most cost-efficient availability at that time.
Eventually data centers are going to develop a "follow the sun" model of operating, De Gilio said, with workloads routed to a center in, say Silicon Valley at one point in the day and then to India at another, based on which is cheapest at the moment. "Maybe there is free cooling available at a certain time of day at one center," he said. "Or maybe a company has a reason to analyze data in Asia on a particular day for time-sensitive reasons."
In short, the system is able to make IT decisions that are not functionality-focused but rather align with the company’s business needs. It is an automatic process, De Gilio said.
"We won't have a person watching over the data center operations and saying, 'oh wait, it is now cheaper to do this in Maine instead of Silicon Valley.'"
The system could also monitor spot market activities, jumping in to take advantage of a sudden drop in costs at a particular data center, he said.
As the amount of data generated by IoT grows, this will be an important asset, De Gilio said.