---------- Forwarded message ---------
From: Ahmed Ali <[email protected]>
Date: Sun, Jun 10, 2018, 8:08 AM
Subject: [STE||AR Gsoc] [Ste||ar GSoC] All-to-All Communications - First
phase report
To: <[email protected]>


Hello everybody,

Hope all your projects are going well.

Here is briefly a report about the first phase in my project.

*A brief description of the project:*

FFLIB is a library demonstrating a model for collective communications. It
has 2 versions: one was specified to demonstrate their model on Portals 4
layer, and the other was general to get the model working on different
layers but it's still in an early stage. In HPX there is an implementation
of the libfabric layer. So the main goal of the project is developing a
backend for the FFLIB library model working on libfabric layer to be
integrated with the HPX implementation.

*What I've done so far:*

   - Understand the paper of FFLIB proposed model
   - Read and understand the Zero Copy serialization paper proposed by HPX.
   - Understand Portals 4 model. I went through the documentation of the
   layer to become familiar with their model and their APIs.
   - Worked on dissecting the FFLIB code of the version working on Portals
   4 layer. I started with it because the other version had too much to be
   done. Understanding the FFLIB implementation was not too easy. There wasn't
   a documentation for the code. So I had to ask the author of the library
   some questions and go through the code to understand it. The code had
   collective communication algorithms besides the implementation of the FFLIB
   model and the APIs working with portals.Going through all of this without
   documentation and understanding the code was too hard. But finally I
   managed to get my way through it.
   - Lastly I've made a call with the author of the library and we were
   advised to go on with the FFLIB version 2 (the generic version not specific
   for Portals 4). Although it's still in an early stage but it has the send
   and recv working. So I was advised to get something working with FFLIB2
   then getting the collective communication from FFLIB1 working in FFLIB2.


*What will be done next:*

   - Going through the FFLIB2 code. It shouldn't be that hard because it
   won't be much different from the FFLIB1 code.
   - Getting a ping pong example working on libfabric layer with the send
   and recv of FFLIB2
   - Implementing collective communication APIs and communication
   algorithms in FFLIB2


I think this phase of the project is the hardest as I had a lot to
understand and we had to choose the best way to go on with through the
project. Also I had some exams at college during this phase. So, working in
the next phase will not be too hard like this phase and I will have much
more time to work.

Best,
Ahmed Samir
_______________________________________________
gsoc mailing list
[email protected]
https://mail.cct.lsu.edu/mailman/listinfo/gsoc
_______________________________________________
hpx-users mailing list
[email protected]
https://mail.cct.lsu.edu/mailman/listinfo/hpx-users

Reply via email to