Skip to content

Conversation

@artpol84
Copy link
Contributor

Force only procs that are participating in the ne Comm to decide what
CID is appropriate. This will have 2 advantages:
* Speedup Comm creation for small communicators: non-participating procs
  will not interfere
* Reduce CID fragmentation: non-overlaping groups will be allowed to use
  same CID.

Signed-off-by: Artem Polyakov [email protected]
(cherry picked from commit 50a6d10)

@artpol84 artpol84 changed the title ompi/comm: Improve MPI_Comm_create algorithm ompi/comm: Improve MPI_Comm_create algorithm (v2.x) Apr 19, 2017
@artpol84 artpol84 modified the milestones: v2.0.3, v2.1.1 Apr 19, 2017
@artpol84
Copy link
Contributor Author

Efficiency verification

#3375 (comment)

    Force only procs that are participating in the ne Comm to decide what
    CID is appropriate. This will have 2 advantages:
    * Speedup Comm creation for small communicators: non-participating procs
      will not interfere
    * Reduce CID fragmentation: non-overlaping groups will be allowed to use
      same CID.

Signed-off-by: Artem Polyakov <[email protected]>
(cherry picked from commit f1f7f20)
@bwbarrett-bot
Copy link

Test Passed

@artpol84
Copy link
Contributor Author

artpol84 commented Apr 21, 2017

MTT verification

#3375 (comment)

@jsquyres
Copy link
Member

Do not merge until #3375 is merged.

@jsquyres
Copy link
Member

Per discussion on the 2017-04-26 webex and per further discussion by the RMs on 2017-04-26, this appears to be an enhancement that isn't classified as a bug fix (e.g., no customer problems were cited). As such, we're dropping it from v2.1.x.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants