Efficiency Benchmarks: Could Storage Show Us How? by Peter Judge

Views and Opinions on Green IT (Oct 18 2010)

1. 

    

   To make data centers more efficient we have to take a closer look at the hardware. And as different parts of the hardware inventory get closer scrutiny - it looks like storage has taken the lead.  
   
   Emerald, a new scheme from the Storage Networking Industry Asociation (SNIA) aims to provide comparative benchmarks for the energy efficiency of storage equipment for data centers. It starts formally in 2011, but its site already has plenty of detail, and I spoke to one of the people who got it started, in SNIA’s Green Storage Initiative. 
   
   “We’ve spent three years on this, behind the scenes,” said Leah Schoeb of VMware, who is chair of the GSI and a veteran of benchmarking initiatives. The problem Emerald had to deal with is a tricky one - how do you compare energy efficiency in data centers which may be carrying out different jobs in different ways?
   
   The Green Grid’s Power Usage Effectiveness (PUE) takes a top-down approach. It simply  addresses the efficiency with which a building delivers energy to IT equipment - energy “wasted” in cooling and other facilities jobs counts against the center. This is good, but we’ve known for some time that there’s a need for other metrics to measure how well energy is used once it gets to the racks.  
   
   Emerald goes bottom-up, to start in on that job. It aims to benchmark energy use in different kinds of storage equipment - and will eventually publish real comparative results for energy use by different companies’ equipment. 
   
   At the moment there’s a place-holder table, with fictitious results for products like “Cool JBOD” and “Tapes-r-Us”, but in the first part of 2011, real products should start appearing there, giving data center people a chance to choose storage on its energy usage. 
   
   All benchmarking is based on defining a fixed specification, then letting companies meet that specification the best way they can. Performance benchmarks, like the TPC transaction processing benchmarks define a certain set of calculations, and hardware makers build kit which can chomp through that as quick as possible, to get a place high up the league table. 
   
   Emerald starts with a “taxonomy” defining the broad types of storage by their purpose, and their size. Storage is online, near-online, removable, or a “virtual media library”, and is for consumer, low-end, mid-range, high-end or mainframe use. 
   
   For each of these types of storage, Emerald defines performance criteria. It is then up to the hardware to meet those criteria, and submit it for testing. Performance benchmarks measure the performance of a system but in Emerald, the performance is given and the tests measure how much energy is used to achieve it. 
   
   Now, servers make up about half the energy used in a data center, but efforts to provide an  energy efficiency comparison have not get very far, although the Energy Start effort from the EPA has started the ball rolling. 
   
   In most cases, to compare energy use by servers, companies have to do their own ad hoc tests, specifying a certain TPC performance and then measuring the energy required to achieve it.
   
   There’s some signs that companies are comparing the energy use of network kit, but Ethernet only uses about ten percent of the energy which reaches the racks in a data center, so it’s lower on the priority list. 
   
   The Emerald effort looks like a good way forward. It’s got a financial model behind it - the vendor pays, with the hope of getting a positive result on a certificate. 
   
   And it will be backed by all the tedious-but-necessary machinery of auditing, appeals and so on - the kind of stuff you don’t personally want to get involved in but which we are heartily glad someone else has done. 
   
   Maybe it’s in the nature of the storage industry to get that kind of thing done. However it’s happened, it looks like storage could be leading the way to more efficient data cen

   Bookmark or Share this article

Login to comment.