[mvapich-discuss] Parameters for build

[Mark Potts]

Lei,
    Thanks for your very prompt responses.

    Is there more info or a writeup on the magnitude of the importance
    of the VCLUSTER value?  I am trying to assist a computer manufacturer
    in their adoption of MVAPICH and need to understand how dramatic an
    effect users might observe in latency/BW if, e.g.,  we were to choose
    SMALL and the system contained 2K nodes or, the reverse, we chose
    LARGE for a 100 node cluster system.

    BTW, did your numbers of <256, 256-1024, >1024 refer to nodes or to
    cpus?  Given a system with N cpus per node, it makes a difference.

    I note that OFED builds of MVAPICH using udapl device don't even ask
    one to choose a size, so that implies a potentially bad default for
    any given system.

           regards,

LEI CHAI wrote:
> Dr. Potts,
> 
> Thank you for trying out MVAPICH 0.9.9 with uDAPL. The answers are inline below:
> 
>>    (1) What are the rough limits for cluster sizes in the build 
>> script        prompt for "small", "medium", and "large" and does 
>> choosing the
>>        inappropriate size have major or minor consequences in
>>        performance or does one hit limits in the app being able to
>>        work at all?
> 
> You can choose "small" if your cluster size is smaller than 256, and "medium" if 256-1024, and "large" if larger than 1024. Generally speaking, "small" configuration will deliver best performance, while "medium" and "large" configurations are more scalable. They should all work though.  
> 
>>    (2) For what reason would one choose to _not_ define the -D_SMP_
>>        and -D_SMP_RNDV_ CFLAG parameters in the build?
>>
> 
> Usually people should always define D_SMP_ and -D_SMP_RNDV_ flags.
> 
>>    (3) Are there performance or other issues associated with
>>        the --enable-sharedlib and --enable-debug parameters in the
>>        configure, i.e. why are these not set as default options in
>>        the build?
> 
> There should be no issue with --enable-sharedlib. We will consider to set it by default in future releases. However, you may want to not set--enable-debug if you want to get the best performance.
> 
> Thanks.
> Lei
> 
>  
> 
>>    Thanks.
>>           regards,
>> -- 
>> ***********************************
>>>> Mark J. Potts, PhD
>>>>
>>>> HPC Applications Inc.
>>>> phone: 410-992-8360 Bus
>>>>        410-313-9318 Home
>>>>        443-418-4375 Cell
>>>> email: potts at hpcapplications.com
>>>>        potts at excray.com
>> ***********************************
>> _______________________________________________
>> mvapich-discuss mailing list
>> mvapich-discuss at cse.ohio-state.edu
>> http://mail.cse.ohio-state.edu/mailman/listinfo/mvapich-discuss
>>

-- 
***********************************
 >> Mark J. Potts, PhD
 >>
 >> HPC Applications Inc.
 >> phone: 410-992-8360 Bus
 >>        410-313-9318 Home
 >>        443-418-4375 Cell
 >> email: potts at hpcapplications.com
 >>        potts at excray.com
***********************************