- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi All,
I have a question to ask about external library.
While looking at various materials, I found out about the communication method called the SHARP(Scalable Hierarchical Aggregation and Reduction Protocol), so I'd like to ask if Intel MPI supports this external library.
I found the option to use the UCX HCOLL library in the following posting, and I would like to know if SHARP is supported similarly.
https://www.intel.com/content/www/us/en/develop/articles/improve-performance-and-stability-with-intel-mpi-library-on-infiniband.html
If not, could you tell me if you are planning to apply in the future?
Best Regards,
Kihang
- Tags:
- Cluster Computing
- General Support
- Intel® Cluster Ready
- Message Passing Interface (MPI)
- Parallel Computing
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi Kihang,
Based on the documentation of Intel MPI we didn't find any references of SHARP as of now.
We are transferring your request to the concerned team.
Regards
Prasanth
Link Copied
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi Kihang,
Based on the documentation of Intel MPI we didn't find any references of SHARP as of now.
We are transferring your request to the concerned team.
Regards
Prasanth
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi Prasanth,
Thank you for your response.
Please close the case. :-)
Best Regards,
Kihang
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi Kihang,
We do implicitly leverage SHARP via HCOLL.
Therefore setting I_MPI_COLL_EXTERNAL as described in the article you mentioned is sufficient.
Best regards,
Michael
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi Michael,
Thank you for your response.
I don't intend to advertise SHARP here (because I haven't experienced the exact functionality), but mellanox says that HCOLL and SHARP are different.
For example, SHARP has the difference of using a specific daemon (sharpd), and their material promotes the performance difference between regular HPCX and HPCX+SHARP.
as you said, we will proceed with the test using HCOLL.
Thank you.
Best Regards,
Kihang
- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page