« New to the list - InfoBright & Aster Data |

Green Data Warehouse Top 10

I just published the second in the series of “Green Data Warehouse” articles in BeyeNetwork.  This article, “Top 10 Things You Can Do to Improve Energy Efficiency“, provides a pragmatic list of 10 things you can do immediately to reduce consumption at the system level.

Joe Foley (CTO Illuminate) left a comment on the benefits of “overloading processors rather than I/O”.  The EPA study cited in my article found that CPUs consumed about 31% of the energy in an average system, more than any other component and more than 6 times the amount consumed by disks.  So I think making more efficient use of CPU (and therefore reducing the number of CPUs) would be more beneficial than reducing disk.

Also, I want to thank Scott Humphrey (Humphrey Strategic Communications) for helping pitch my article series to BeyeNetwork.

This entry was posted on 23. February 2009 at 13:47 and is filed under Articles. You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response or trackback from your own site.

3 Responses to “Green Data Warehouse Top 10”

  1. jtfoley says:
    24. February 2009 at 09:59

    It’s quite true that the processor is the most power hungry component in a computer. The point is that by loading the CPU rather than the I/O system you make most efficient use of that power. The chip uses essentially the same amount of power when it is running at 99% as it does when it is idling. In a massively parallel system you will have many CPUs running at 5% - 10% of capacity and I/O channels maxed out. With the correlation database you will have a few CPU’s running at 90% - 100% of capacity and a few I/O channels managing the entire I/O load.

  2. Rick Abbott says:
    3. March 2009 at 13:44

    Joe- Thanks for the clarification. By overloading the processors you reduce the need for adding additional CPUs thus improving the overall efficiency of the “server units”.

Leave a Reply

You must be logged in to post a comment.