Search Mailing List Archives


Limit search to: Subject & Body Subject Author
Sort by: Reverse Sort
Limit to: All This Week Last Week This Month Last Month
Select Date Range     through    

[mininet-discuss] Completion time difference for running MPI experiment on hardware and Mininet topology

evelina dumitrescu evelina.a.dumitrescu at gmail.com
Wed Apr 13 09:03:52 PDT 2016


More insight for what I trying to do:

I have a set of analytical models of the execution time from some MPI
applications that I try to validate against real measurements on different
network topologies.
However, I don't have real systems for all the topologies that I want to
model.

I tried to use Mininet since it allows the simulation of different types of
toplogies.

When comparing the execution times between Mininet and a real hardware
network topology, I noticed that the execution time of the MPI applications
is several times lower in Mininet.

Is it possible to use Mininet for what I am trying to achieve ?



On Fri, Apr 8, 2016 at 4:45 PM, evelina dumitrescu <
evelina.a.dumitrescu at gmail.com> wrote:

> Hi,
>
> I have a hardware topology with 2 hosts connected through a switch.
> The link  bandwidth between the switch and the 2 hosts is about 100Mbs;
> the observed latency measured with ping is about 0.26ms and the iperf
> throughput is about 37Mbs;
> I used the same topology in Mininet with 37Mbps bandwith and delay 0.26ms.
> I run some MPI experiments on the hardware testbed and in Mininet and
> measured the execution time.
> The time obtained in Mininet is half the time in the hardware topology.
> My questions are:
> - what exactly represents the delay in Mininet (theoretically, the delay
> is represented as the sum of putting the bits on the wire + flight time)
> - how should I setup the parameters in Mininet to obtain the same results?
> What other parameters should I take into account?
> - is it normal to get such difference in time between the hardware testbed
> and Mininet ?
>
> Thanks,
> Evelina
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.stanford.edu/pipermail/mininet-discuss/attachments/20160413/e4824a05/attachment.html>


More information about the mininet-discuss mailing list