airavata-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Shenoy, Gourav Ganesh" <>
Subject Re: Running MPI jobs on Mesos based clusters
Date Fri, 21 Oct 2016 02:26:42 GMT
Hi Mangirish, devs:

The Aurora documentation for “Tasks” & “Processes” provides very good information
which I felt would be helpful in implementing gang scheduling, as you mentioned.

From what I understood, there are these constraints:

1.       If targeting single-node (multi-core) MPI, then a “JOB” will be broken down into
multiple “PROCESSESES”, each of which will run on these multi-cores.

2.       Even if any one of these processes fail, then the JOB should be marked as failed.

As mentioned in my earlier email, Aurora provides Job abstraction – “a job consists of
multiple tasks, which in turn consist of multiple processes”. This abstraction comes in
extremely handy if we want to run MPI jobs on a single node.

While submitting a job to Aurora, we can control the following parameters for a TASK:

a.       “max_failures” for a TASK – the number of failed processes which is needed
to mark a task as failed. Hence if we set max_failures = 1, then even if a single process
in a task fails, Aurora will mark that task as failed.
Note: Since a JOB can have multiple tasks, and even a JOB has “max_task_failures” parameter,
we can set this to 1.

b.       “max_concurrency” for a TASK – number of processes to run in parallel. If a
node has 16 cores, then we can limit the amount of parallelism to <=16.

I did not get much time to experiment with these parameters for job submission, but found
this document to be handy and worth sharing. Hope this helps!

Thanks and Regards,
Gourav Shenoy

From: Mangirish Wagle <>
Reply-To: "" <>
Date: Tuesday, October 18, 2016 at 11:48 AM
To: "" <>
Subject: Re: Running MPI jobs on Mesos based clusters

Sure Suresh, will update my findings on the mailing list. Thanks!

On Tue, Oct 18, 2016 at 7:59 AM, Suresh Marru <<>>
Hi Mangirish,

This is interesting. Looking forward to see what you will find our further on gang scheduling
support. Since the compute nodes are getting bigger, even if you can explore single node MPI
(on Jetstream using 22 cores) that will help.


P.S. Good to see the momentum on mailing list discussions on such topics.

On Oct 18, 2016, at 1:54 AM, Mangirish Wagle <<>>

Hello Devs,
Here is an update on some new learnings and thoughts based on my interactions with Mesos and
Aurora devs.
MPI implementations in Mesos repositories (like MPI Hydra) rely on obsolete MPI platforms
and no longer supported my the developer community. Hence it is not recommended that we use
this for our purpose.
One of the known ways of running MPI jobs over mesos is using "gang scheduling" which is basically
distributing the MPI run over multiple jobs on mesos in place of multiple nodes. The challenge
here is the jobs need to be scheduled as one task and any job errored should collectively
error out the main program including all the distributed jobs.

One of the Mesos developer (Niklas Nielsen) pointed me out to his work on gang scheduling: This code may not be fully tested but certainly a good starting point
to explore gang scheduling.

One of the Aurora developer (Stephen Erb) suggests using gang scheduling on top of Aurora.
Aurora scheduler assumes that every job is independent. Hence, there would be a need to develop
some external scaffolding to coordinate and schedule these jobs, which might not be trivial.
One advantage of using Aurora as a backend for gang scheduling is that we would inherit the
robustness of Aurora, which otherwise would be a key challenge if targeting bare mesos.
Alternative to all the options above, I think we should probably be able to run a 1 node MPI
job through Aurora. A resource offer with CPUs and Memory from Mesos is abstracted as a single
runtime, but is mapped to multiple nodes underneath, which eventually would exploit distributed
resource capabilities.
I intend to try out the 1 node MPI job submission approach first and simultaneously explore
the gang scheduling approach.
Please let me know your thoughts/ suggestions.
Best Regards,

On Thu, Oct 13, 2016 at 12:39 PM, Mangirish Wagle <<>>
Hi Marlon,
Thanks for confirming and sharing the legal link.

On Thu, Oct 13, 2016 at 12:13 PM, Pierce, Marlon <<>>
BSD is ok:

From: Mangirish Wagle <<>>
Reply-To: "<>" <<>>
Date: Thursday, October 13, 2016 at 12:03 PM
To: "<>" <<>>
Subject: Re: Running MPI jobs on Mesos based clusters

Hello Devs,
I needed some advice on the license of the MPI libraries. The MPICH library that I have been
trying claims to have a "BSD Like" license (
I am aware that OpenMPI which uses BSD license is currently used in our application. I had
chosen to start investigating MPICH because it claims to be a highly portable and high quality
implementation of latest MPI standard, suitable to cloud based clusters.
If anyone could please advise on the acceptance of the MPICH libraries MSD Like license for
ASF, that would help.
Thank you.
Best Regards,
Mangirish Wagle

On Thu, Oct 6, 2016 at 1:48 AM, Mangirish Wagle <<>>
Hello Devs,

The network issue mentioned above now stands resolved. The problem was with the iptables had
some conflicting rules which blocked the traffic. It was resolved by simple iptables flush.

Here is the test MPI program running on multiple machines:-

[centos@mesos-slave-1 ~]$ mpiexec -f machinefile -n 2 ./mpitest
Hello world!  I am process number: 0 on host mesos-slave-1
Hello world!  I am process number: 1 on host mesos-slave-2

The next step is to try invoking this through framework like Marathon. However, the job submission
still does not run through Marathon. It seems to gets stuck in the 'waiting' state forever
(For example Further, I notice that Marathon
is listed under 'inactive frameworks' in mesos dashboard (

I am trying to get this working, though any help/ clues with this would be really helpful.

Thanks and Regards,
Mangirish Wagle

On Fri, Sep 30, 2016 at 9:21 PM, Mangirish Wagle <<>>
Hello Devs,

I am currently running a sample MPI C program using 'mpiexec' provided by MPICH. I followed
their installation guide<>
to install the libraries on the master and slave nodes of the mesos cluster.

The approach that I am trying out here is that I am equipping the underlying nodes with MPI
handling tools and then use the Mesos framework like Marathon/ Aurora to submit jobs to run
MPI programs by invoking these tools.

You can potentially run an MPI program using mpiexec in the following manner:-

# mpiexec -f machinefile -n 2 ./mpitest

  *   machinefile -> File which contains an inventory of machines to run the program on
and number of processes on each machine.
  *   mpitest -> MPI program compiled in C using mpicc compiler. The program returns the
process number and he hostname of the machine running the process.
  *   -n option indicates number of processes that it needs to spawn
Example of machinefile contents:-

# Entries in the format <hostname/IP>:<number of processes>

The reason for choosing slaves is that Mesos runs the jobs on slaves, managed by 'agents'
pertaining to the slaves.

Output of the program with '-n 1':-

# mpiexec -f machinefile -n 1 ./mpitest
Hello world!  I am process number: 0 on host mesos-slave-1

But when I try for '-n 2', I am hitting the following error:-

# mpiexec -f machinefile -n 2 ./mpitest
[proxy:0:1@mesos-slave-2] HYDU_sock_connect (/home/centos/mpich-3.2/src/pm/hydra/utils/sock/sock.c:172):
unable to connect from "mesos-slave-2" to "mesos-slave-1" (No route to host)
[proxy:0:1@mesos-slave-2] main (/home/centos/mpich-3.2/src/pm/hydra/pm/pmiserv/pmip.c:189):
unable to connect to server mesos-slave-1 at port 44788 (check for firewalls!)

It seems to not allow the program execution due to network traffic being blocked. I checked
security groups in scigap openstack for mesos-slave-1, mesos-slave-2 nodes and it is set to
'wideopen' policy. Furthermore, I tried adding explicit rules to the policies to allow all
TCP and UDP (Currently I am not sure what protocol is used underneath), even then it continues
throwing this error.

Any clues, suggestions, comments about the error or approach as a whole would be helpful.

Thanks and Regards,
Mangirish Wagle

Error! Filename not specified.

On Tue, Sep 27, 2016 at 11:23 AM, Mangirish Wagle <<>>
Hello Devs,

Thanks Gourav and Shameera for all the work w.r.t. setting up the Mesos-Marathon cluster on

I am currently evaluating MPICH ( to be used for launching
MPI jobs on top of mesos. MPICH version 1.2 supports Mesos based MPI scheduling. I have been
also trying to submit jobs to the cluster through Marathon. However, in either cases I am
currently facing issues which I am working to get resolved.

I am compiling my notes into the following google doc. You may please review and let me know
your comments, suggestions.

Thanks and Regards,
Mangirish Wagle

Error! Filename not specified.

On Wed, Sep 21, 2016 at 3:20 PM, Shenoy, Gourav Ganesh <<>>
Hi Mangirish,

I have set up a Mesos-Marathon cluster for you on Jetstream. I will share with you with the
cluster details in a separate email. Kindly note that there are 3 masters & 2 slaves in
this cluster.

I am also working on automating this process for Jetstream (similar to Shameera’s ansible
script for EC2) and when that is ready, we can create clusters or add/remove slave machines
from the cluster.

Thanks and Regards,
Gourav Shenoy

From: Mangirish Wagle <<>>
Reply-To: "<>" <<>>
Date: Wednesday, September 21, 2016 at 2:36 PM
To: "<>" <<>>
Subject: Running MPI jobs on Mesos based clusters

Hello All,

I would like to post for everybody's awareness about the study that I am undertaking this
fall, i.e. to evaluate various different frameworks that would facilitate MPI jobs on Mesos
based clusters for Apache Airavata.

Some of the options that I am looking at are:-

  1.  MPI support framework bundled with Mesos
  2.  Apache Aurora
  3.  Marathon
  4.  Chronos
Some of the evaluation criteria that I am planning to base my investigation are:-

  *   Ease of setup
  *   Documentation
  *   Reliability features like HA
  *   Scaling and Fault recovery
  *   Performance
  *   Community Support
Gourav and Shameera are working on ansible based automation to spin up a mesos based cluster
and I am planning to use it to setup a cluster for experimentation.

Any suggestions or information about prior work on this would be highly appreciated.

Thank you.

Best Regards,
Mangirish Wagle
Error! Filename not specified.

View raw message