llnl / llnl/UnifyFS

Replace hostname gather/scatter with more scalable algorithm

Open
#191 1 comment 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
C
Stars
122
Forks
34
PR merge metrics
No merged PRs in 30d

Description

To determine which MPI ranks are on the same node, this code gathers hostnames to rank 0, rank 0 computes the local rank ids, and then the result is scattered back out.

here:
https://github.com/LLNL/UnifyCR/blob/3c7e3a941480b0d27182628d9fef0eec9d1d875c/client/src/unifycr.c#L2393

and here:
https://github.com/LLNL/UnifyCR/blob/3c7e3a941480b0d27182628d9fef0eec9d1d875c/server/src/unifycr_init.c#L277

A more scalable method would be to use MPI_Comm_split_type (added in MPI-3):
https://www.open-mpi.org/doc/v2.0/man3/MPI_Comm_split_type.3.php

or via rankstr (ECP VeloC component):
https://github.com/ECP-VeloC/rankstr

Contributor guide

No contributing guide indexed for this repository

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start by reading the hostname gather/scatter code at client/src/unifycr.c:2393 and server/src/unifycr_init.c:277, then compare the MPI_Comm_split_type and rankstr approaches linked in the issue. Done means replacing the current rank-locality algorithm with a more scalable implementation in the affected paths.

Written by the indexing model from the issue text.

Assessment

Tech stack
c
Domain
distributed-systems, hpc
Issue type
Refactor
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Mostly clear
Newbie friendliness
38/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.