Snow in Denver

Author: Nick Johnson
Posted: 21 Nov 2013 | 16:18

It's snowing!

Today is the last day of the exhibition at SC13 so whilst I'm manning the booth (#3932) I thought I'd talk about some of the sessions I've been to and people I've met since my last post.

Yesterday I and some of the other members of the Adept project met with Kirk Cameron and his students from Virginia Tech. For the past few years they've been working on a project developing methods for measuring power consumption and power modelling in x86 multicore devices. You can see the results here: They have also done work integrating the PowerPack measurement tool with scorep.

Related to this, we met with Penguin Computing who also offer a system for performing component-level power. Phil Pokorny developed this system in conjunction with Sandia NL so it was great to be able to meet him (and see the hardware).

So, you might be wondering, what's the difference? Penguin Computing prefers a non-invasive approach and uses a BeagleBoard in conjunction with Hall effect sensors. This gives a really accurate measurement of current but necessitates some calibration of the sensors. The other approach is to use interposers (risers) fitted with current sense resistors (CSRs). This can be a slightly easier tack as it doesn't require calibration but it can be difficult to measure high currents with a precision resistance.

On Tuesday, I went to the CASC luncheon with EPCC colleagues Alison Kennedy and Toni Collis. It was nice (if a little daunting) to meet the directors of the big US facilities. Hopefully this can help to foster some links with US labs as there are definite links between projects that would be good to exploit. In a similar vein, Tuesday night was the UKUSAHPC networking dinner. With representatives from the NSF, NIST, DoE and EPSRC, one could get a good feel for how things work in the US. Hopefully this will allow us to explore some ideas for joint funding for transatlantic projects.

Anyone with an interest in power modelling and measurement should feel free to get in touch with me.


Nick Johnson, EPCC

Blog Archive