Data center facilities pro

Feb 11 2009   2:41PM GMT

Measuring data center performance

Mark Fontecchio Mark Fontecchio Profile: Mark Fontecchio

There has plenty of talk around the green data center and data center energy efficiency, and one of the players has been The Green Grid, a nonprofit group focused on the topic. The Green Grid came out with power usage effectiveness (PUE), which compares total facility power to IT equipment power.

But in the end, what matters is what your data center does, not how much energy it consumes. You might have a PUE of 1.1, but if your servers just sit there idly all day long, who cares? Data centers are built to perform work, and if they don’t do that, energy consumption doesn’t mean squat. Re-enter The Green Grid.

Over the last year or so, there has been discussion over defining data center performance compared to energy consumed, often referred to as a data center’s useful work or data center productivity. The Green Grid has now come out with eight different proposals for “proxies,” which the group describes as approximations for comparing data center production to data center energy consumption. It compares them to the stickers in car lots that claim a certain miles-per-gallon rating, right down to the warning that “your mileage may vary.”

And before companies start comparing each other’s data center production, a warning from The Green Grid:

Comparisons between data centers would be valuable in a marketing or evaluation sense, but it is unlikely that any proxy for data center productivity will be comparable across multiple data centers. Rather, the primary use for a proxy will continue to be an indication of improvement over time for a single data center, and very constrained comparisons between data centers that perform the same function.

The data center production proxies

Proxy 1: User-defined measure of useful work divided by energy consumption. This proposal defers to the user to define useful work in a data center. That could be the number of emails sent, or the number of database queries handled. Whatever the case, it’s up to the user to define and measure it.

  • Pro: User gets to define useful work
  • Con: That definition could vary from application to application and server to server, making an overall measure of the data center difficult
  • Measured in tasks per kilowatt-hour (kwh)

Proxy 2: Green Grid-provided measure of useful work divided by energy consumption. Through The Green Grid, Intel Corp. will provide a software development kit with an application programming interface that you can install on a subset of your IT equipment. It will report data from software running on those servers, which can be converted to useful work and compared to energy consumption for that subset. That number can then be extrapolated for the entire data center.

  • Pro: Provides a standard way to measure useful work across applications
  • Con: Requires download and running of external software
  • Measured in tasks/kwh

Proxy 3: Sample workload divided by energy consumption of a subset of servers. The Green Grid says it will provide a bunch of sample workloads that users can run on a subset of servers. The user decides which sample workload best describes what the overall data center does. The workload is run to get a measurement of work completed. That is divided by energy consumption and extrapolated for the entire data center.

  • Pro: Similar to current benchmarking tests in the data center
  • Con: Mixed-workload data centers might not benefit as much, and sample workloads must be made to work on as many server platforms as possible
  • Measured in tasks/kwh

Proxy 4: Bits per kilowatt-hour. Add the total number of bits coming out of all outbound routers, and divide by energy consumption.

  • Pro: Easy to set up and measure, with an easy-to-understand result
  • Con: Uncertainty about whether all bits are created equal
  • Measured in megabits/kwh

Proxy 5: Server utilization using SPEC’s CPU benchmark. Measure CPU utilization over a period of time with the existing CPU benchmark from the Standard Performance Evaluation Corp. (SPEC), and divide by energy consumption.

  • Pro: Easy to implement, schedule and understand
  • Con: Benchmark is not available on all server platforms, and only measures the CPU utilization of the application in question, and not underlying framework such as the operating system and systems management tools
  • Measured in jobs/kwh

Proxy 6: Server utilization using SPEC’s power benchmark. Same as the previous proxy, but this time using the SPEC power benchmark, which measures performance compared to power for a server.

  • Pro: Easy to implement, schedule and understand
  • Con: SPEC power results depend on manufacturers publishing updated measurements for their server products
  • Measured in power-weighted jobs per kwh

Proxy 7: Compute units per second trend curve. Group your servers by the year purchased. A server produced in 2002 equals one million compute units per second. The value then increases or decreases by a factor of seven every five years depending on when the servers were purchased. Add the total number of compute units and divide by energy consumption.

  • Pro: No software needed and no benchmarks to run
  • Con: Bias toward newer servers and a lack of comparison of different servers released in the same year
  • Measured in millions of compute units per kwh

Proxy 8: Operating system workload efficiency. Calculates the number of operating system instances and compares that number to the power being used at that time.

  • Pro: Provides good high-level estimate of efficiency and utilization
  • Con: Not as granular as some might want (ie., what if the operating systems aren’t even running applications?)
  • Measured in operating instances per kwh

Wow, that’s a lot of proxies! Let us know which one or ones you like, and which ones you think are stupid. The Green Grid plans to digest comments from its members and the data center user crowd for at least a few months, and then decide sometime after that which one it favors. The group has even set up an online survey on the proxies, so you can let The Green Grid know which you prefer.

 Comment on this Post

There was an error processing your information. Please try again later.
Thanks. We'll let you know when a new response is added.
Send me notifications when other members comment.

Forgot Password

No problem! Submit your e-mail address below. We'll send you an e-mail containing your password.

Your password has been sent to:

Share this item with your network: