Friday Sep 21, 2007

Energy versus Code

On Tuesday I attended the roundtables and discussions surrounding the Department of Energy's MOU to join the Green Grid. It's great to see a government agency with a department of efficiency, and even more important is that having the DOE on stage drags eco-computing front and center as a national policy issue as well.

Among the dominant themes was that of high voltage power distribution, comparing high voltage AC and DC and the potential energy savings of each. There's a study that Sun and Lawrence Berkeley Labs conducted to examine DC power distribution in the data center, and the first order estimates are that it could save 5-7% over AC. That seems significant, until you put together a pareto of other energy savings ideas. What's lost is lost -- whether it's in the conversion of AC to DC or in spinning disks to manage data that could be migrated to tape or running inefficient code.

The last point is one that was touched by both DOE Assistant Secretary for Efficiency Alexadner Karsner and the New York Stock Exchange's CTO, Steve Rubinow. Rubinow commented that we've come to take "cheap processors and memory" as an excuse to get messier with code, and as a result, we have inefficient applications consuming power on gross scales. Karsner was more direct: "The most available, cheap source of energy is that which we waste".

As we've seen with other environmental and ecological initiatives, real savings come from both reduction in demand as well as efficiencies gained in production and distribution. Managing demand is independent of supply voltages, and can be kicked-off with software tools like the Sun Studio Performance Analyzer as well as systemic inspection offered through Dtrace. Back in the 70s, Tower of Power preached that "There's Only So Much Oil in the Ground." While we've improved discovery, production and distribution of fossil fuels, it's been conservation efforts and efficiencies that extended the lifespan of dead dinosaurs.

Thursday Feb 08, 2007

Not Your Father's HPC

Marc Hamilton and I put the spin on High Performance Computing in the latest wave of our Innovating@Sun podcast program.

HPC normally conjures up images of big FORTRAN applications and ray tracing, look for radar shadows or casting light shadows to make computer animation more pleasing to our visual radar. But we're now seeing a class of HPC applications that shift the usual time-space tradeoffs, putting enough data into memory to allow near real-time analysis in areas like transaction fraud detection and logistics optimization.

About

Hal Stern's thoughts on software, services, cloud computing, security, privacy, and data management

Search

Archives
« April 2014
SunMonTueWedThuFriSat
  
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
   
       
Today