Friday, March 21, 2008

Interview on MIX Online

I have a video interview on MIX Online titled "Greg Linden: Trends in Collective Intelligence and Centralization".

I ramble on for about 20 minutes on topics in personalization, recommendations, and some thoughts on MapReduce and Dryad. Nothing earth-shattering in there, but perhaps worth twenty minutes if you just can't get enough Greg Linden in your day.

Watching the video requires downloading Microsoft Silverlight, sorry about that. If you do go ahead and install Silverlight, you might as well check out its nifty multiresolution image zoom demo on the Hard Rock site too.

Update: In the comments to this post, Simon Lyall pointed out that the video is available in many formats at the bottom of the page (under the "Downloads" section), so Silverlight is not required to watch it.


Simon said...

If you click on the "Downloads" link at the bootom of the page the video is available in 7 (!) formats. So Silverlight is not required.

Greg Linden said...

Thanks, Simon! I missed those links. Much appreciated!

Shane said...

greg, could you elaborate on the types of computations that are suited to dryad over mapreduce?

Greg Linden said...

Hi, Shane. The Dryad EuroSys 2007 paper (PDF) goes into details and has several examples.

I could give you an answer, but they do it better than I would. Here is an excerpt from the paper:

Google's MapReduce ... addresses a similar problem domain. The fundamental difference between the two systems is that a Dryad application may specify an arbitrary communication DAG rather than requiring a sequence of map/distribute/sort/reduce operations.

In particular, graph vertices may consume multiple inputs, and generate multiple outputs, of different types. For many applications this simplifies the mapping from algorithm to implementation, lets us build on a greater library of basic subroutines, and, together with the ability to exploit TCP pipes and shared-memory for data edges, can bring substantial performance gains. At the same time, our implementation is general enough to support all the features described in the MapReduce paper.