Hi Clemens, Our TCP-based netmods don't have this feature of packet mmap. Are you sure that the high CPU load you are seeing is not caused by active polling for completion? Otherwise, could you share with us your profiling results? Antonio On 07/09/2014 07:57 AM, Clemens Eisserer wrote:
Hi,
We've recently conducted some tests showing that CPU overhead is significant when sending/receiving data via TCP or UDP over 10GigE cards on Linux (20-50% of an Ivy Bridge core clocked at 3Ghz). A lot of time is spent copying data from kernel to user space and vice versa.
As we would like to avoid this overhead as far as possible, does mpich2 support zero copy data transfer using packet mmap?
Thank you in advance, Clemens _______________________________________________ discuss mailing list [email protected] To manage subscription options or unsubscribe: https://lists.mpich.org/mailman/listinfo/discuss
-- Antonio J. Pe�a Postdoctoral Appointee Mathematics and Computer Science Division Argonne National Laboratory 9700 South Cass Avenue, Bldg. 240, Of. 3148 Argonne, IL 60439-4847 [email protected] www.mcs.anl.gov/~apenya