Power Management at the Data Center: From Capacity to Control

One of the most interesting companies using the Intel Data Center Manager (DCM) technology is CiRBA, whose Data Center Intelligence (DCI) platform can analyze, assign, and move workloads with maximum efficiency. This article reviews how CiRBA is working to change the way data centers are managed and the role that Intel DCM—embedded in DCI—plays in providing an unprecedented amount of information about energy usage.

CiRBA Data Center Intelligence

CiRBA, a Toronto-based software company, offers an advanced analytics software package called DCI, which enables organizations to run their computing landscapes at a high level of efficiency without taking on undue risk by conducting intelligent workload placement.

CiRBA’s approach is like the game Tetris. Computing workloads are like Tetris pieces. They all have personalities and attributes that fit together in certain ways, controlled by limiting factors, such as capacity of hardware, configuration, and business policy. Playing Tetris poorly in the data center can result in stranded capacity—with machines just sitting there, consuming power without doing much computing.

CiRBA DCI gathers data from the array of assets supporting workloads, incorporates business and regulatory policy, and creates placement plans for workloads across in-house hardware, local clouds, and the public cloud. This allows CiRBA’s customers to transform legacy infrastructure into virtual infrastructure for their customers. DCI takes into account all of the implications of moving a workload, from applications’ relationship to databases, accelerators, hardware capacity, load balancing, and bandwidth, and with the help of Intel DCM, to actual hardware power consumption. This level of sophisticated analytics means that data center managers can now transfer workloads from physical servers to the cloud (P2C) and from mid-level Unix hardware to Intel x86 architecture (P2P) at the same time, boosting efficiency in the process.

Additionally, DCI can manage cloud assets, watching over the various environments, notifying users when opportunities arise to optimize more virtual and physical resources, based on both technical and nontechnical attributes, such as business rules.

Historically, workload management has been handled using a combination of spreadsheets and the trial-and-error recombination of logical and physical clusters. The result of this has been like that Tetris screen with a big gap in the middle—stranded capacity. Many organizations that have virtualized resources have as much as double the capacity they actually need in their data centers, according to CiRBA’s head of operations and product management, Chuck Tatham.

“Because they are paralyzed by the complexity of [managing workloads] empirically, they just throw capacity at the problem,” Tatham says.

The Role of Intel DCM

In today’s power-constrained environment, buying more gear is not a sustainable solution to the workload problem. Some number-crunching financial institutions are already at the point where their host municipalities can no longer supply them with as much power as they need for their data centers, Tatham says. CiRBA’s technology includes the ability to benchmark the likely power-consumption impact of moving workloads across equipment. But the information input for power consumption was typically what the manufacturer “plated” as the top wattage rating for the unit, or the predicted CPU utilization. Use of this value typically leads to overestimation, because machines are rarely used at full power, full-time.

Intel DCM allows CiRBA’s platform to communicate directly with a server and take in real-time power consumption data, associating a real power curve with server behavior. This allows CiRBA to optimize infrastructure around “compacting, de-fragmenting, or de-balancing” infrastructure, Tatham says. For example, workloads can be moved during off-peak hours, when machines are not as busy, to a much smaller group of servers, powering down the machines that are not being used. CiRBA DCI can also predict which workloads will behave in a certain way, providing reliable advice about when machines can be sent into a low-power state.

Intel DCM also enables CiRBA DCI customers to charge back infrastructure with richer data than before. For example, the power consumption of an individual virtual machine on a server can be determined so that the owner of that virtual machine can be charged by IT with fine granularity, empowering the owner to make more informed decisions about how to balance workloads.

Intel DCM is able to do this because it draws power consumption and thermal data from the server itself. Through a web service interface, this data is fed into CiRBA’s analytics engine.

The two companies discovered each other when Intel became involved in helping strategic customers move from mid-range Unix-based systems such as Sun SPARC to Intel x86 machines. Intel wanted to make the workload analytics provided by CiRBA available to its customers, and CiRBA saw that it could benefit from adding real-time data on power to its analytics portfolio.

Benefits of Embedding Intel DCM

Now that Intel DCM is embedded in the offering, customers of CiRBA DCI can factor power into their analyses to further green IT agendas and spend less on power and cooling as they virtualize their workloads. The partnership is not only useful for reorganizing or placing workloads in the cloud; customers can also avoid designing infrastructure that is too dense from a power-consumption perspective or too hot from a thermal dynamics perspective.

“If you’re a CIO and you’re sitting on a data center that is absolutely tapped, the city won’t give you any more power, and you’re faced with building a $25 or $30 million data center in order to get more capacity, there’s a pretty strong argument to avoid building a new data center by getting smarter with power,” Tatham says.
Intel DCM is helping partner applications such as CiRBA to become an important decision-making tool for datacenter managers with daunting power management issues—an audience that previously was only guessing at the scale of those issues.

Related Articles

Seeing Is Believing in Data Centers
High Performance Computing Vendors Gain Energy Advantage
Eliminating Massive Amounts of Energy Micro-Waste