Hadoop+GPU

  • March 26, 2012 · 6:00 PM
  • This location is shown only to members

Hadoop has become one of the most popular platforms for distributed 'big data' analysis. Putting 'big compute' in the form of GPUs alongside the data across a cluster offers the prospects of some interesting performance gains. Come join us for an evening of discussion on "Hadoop+GPU".

Join or login to comment.

  • A former member
    A former member

    For someone coming at this from the Hadoop side it was a good intro to GPU-ing the Hadoop.

    March 27, 2012

  • Andrew S.

    Tim Child talked about some very interesting work he has done with GPUs in the database, and Jack Pappas gave some valuable insights into finding patterns within your code that can be translated to MapReduce. A lot of the usual faces and some new people as well.

    March 27, 2012

  • A former member
    A former member

    Another great meetup. Thanks Andrew!

    March 27, 2012

  • Enzo A.

    Great speakers. Thank you Shep, Jack, Tim.

    March 27, 2012

  • Michael C.

    I am looking to understand how to bring parallelization into my modelling for research projects and am not a CS person by training. All of the speakers made it easy for me to understand the parts that might be important to me in both learning to program map/reduce for GPUs and how to better think about my application and domain problems. Very valuable to me.

    March 27, 2012

  • adusumilli

    Great talks!!

    March 27, 2012

  • deepak

    I would assume that a use case would be something like very large scale image processing (eg Satellite Images etc). Hadoop provides features like HDFS (for storing data in a fault tolerant manner), and writing distributed applications using the simple MapReduce Framework. I think the most natural way to use Hadoop with GPU is to spawn Map-only jobs where Mappers copy HDFS data to local FS and invoke GPU code. If the task on my one node fails, Hadoop would take care of re-scheduling it, etc..

    January 22, 2012

  • A former member
    A former member

    Thanks to Andrew Sheppard for starting this meetup. And to the entrepreneurs that started Meetup,

    Tim reflects my experience with using GPU's. Let's not forget they are an artifact of the billion dollar computer game industry. The original intention was to render pixels on a screen. Nvidia is leveraging this for more general purposes. We are still at an early stage. At some point we will not have to use an Excel spreadsheet to divine occupancy levels of cores when we launch a kernel.

    January 13, 2012

  • Tim C.

    There's an irony with Hadoop+GPU. I've touched on it with my work on GPU DBMS stored procedures. The irony is Hadoop is based on map reduce. Map/Reducing is splitting work into small discrete pieces for parallel processing. But with a GPU you want 100's - 1000s of work items per GPU. So if the Hadoop input is a sequence of records then you need to Map/Increase to the 100s - 1000's of work item for successful GPU throughput

    January 13, 2012

  • Gavin Y.

    This topic sounds so interesting... can not wait...

    January 13, 2012

People in this
Meetup are also in:

Sometimes the best Meetup Group is the one you start

Log in

Not registered with us yet?

Sign up

Meetup members, Log in

By clicking "Sign up" or "Sign up using Facebook", you confirm that you accept our Terms of Service & Privacy Policy