airavata-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Suresh Marru <sma...@apache.org>
Subject Re: Mesos GPU Support
Date Tue, 08 Nov 2016 04:49:43 GMT
Hi Gourav,

I am running the master inside a virtual box and running agent natively. I wanted the 1.0.1
since it claims the GPU support. You EC2 installation seem to verify that. I am using home-brew
for the agent installation which is using the formula - https://raw.githubusercontent.com/Homebrew/install/master/install
<https://raw.githubusercontent.com/Homebrew/install/master/install> 

Suresh

> On Nov 7, 2016, at 11:41 PM, Shenoy, Gourav Ganesh <goshenoy@indiana.edu> wrote:
> 
> Hi Suresh,
>  
> I have never used Tensor Flow, so not sure if I’ll be able to help there. But with
respect to the GPU support for Mesos, the current installation on EC2 (v1.0.1) does detect
GPUs, and the same is listed on their dashboard (NOTE: our cluster is not GPU enabled, hence
mesos dashboard shows GPU=0).
>  
> Also, how are you installing Mesos on mac? Did you install any virtualization software
to create VMs (for master/slave)?
>  
> Thanks and Regards,
> Gourav Shenoy
>  
> From: Suresh Marru <smarru@apache.org>
> Reply-To: "dev@airavata.apache.org" <dev@airavata.apache.org>
> Date: Monday, November 7, 2016 at 11:27 PM
> To: Airavata Dev <dev@airavata.apache.org>
> Subject: Mesos GPU Support
>  
> Hi All, 
>  
> I am trying some tensor flow tutorials and seem to have gone on a wrong tangent. Can
any of the Mesos hackers on this list help with get me get back on track?
>  
> I want run through benchmarks in [1] on mac with a GT 650M card. I got the bundled example
working but stuck in getting the tensor flow cnn tutorial example [2] working natively on
my mac. I tried to deploy the mesos first so I could try to run tensor flow from Marathon
as described in [3]. Have any of you managed to get Mesos working on a Mac? If so, can you
please point me to instructions which worked for you? 
>  
> I want to have Mesos recognize and use the GPU card. From [4], it looks like the latest
release should have GPU support natively within mesos. Have any one tried it? I am working
through the tutorial [5] but I am blocked since mesos on mac is not working correctly in the
first place. But this tutorial also seem to suggest to use spark to run tensor flow jobs.
Any suggestions on what is the best way to run tensor flow on gpu’s and use Mesos manage
these jobs? I prefer to do it on a mac, but if thats a bad idea, can try these steps on EC2
instead. 
>  
> Thanks,
> Suresh
>  
> [1] - https://github.com/soumith/convnet-benchmarks <https://github.com/soumith/convnet-benchmarks>
> [2] - https://www.tensorflow.org/versions/r0.11/tutorials/deep_cnn/index.html#convolutional-neural-networks
<https://www.tensorflow.org/versions/r0.11/tutorials/deep_cnn/index.html#convolutional-neural-networks>
> [3] - https://github.com/tensorflow/ecosystem/tree/master/marathon <https://github.com/tensorflow/ecosystem/tree/master/marathon>
> [4] - https://issues.apache.org/jira/browse/MESOS-4424 <https://issues.apache.org/jira/browse/MESOS-4424>

> [5] - http://schd.ws/hosted_files/mesosconna2016/b9/GPU%20Support%20in%20Mesos.pdf <http://schd.ws/hosted_files/mesosconna2016/b9/GPU%20Support%20in%20Mesos.pdf>
>  


Mime
View raw message