GEOS icon indicating copy to clipboard operation
GEOS copied to clipboard

Update memory statistics table

Open victorapm opened this issue 2 years ago • 6 comments

What is the requested feature? Add minimum and average (across ranks) memory utilization statistics from umpire and update format of the statistics table.

Is your request related to a specific problem? No

Describe the solution you'd like Here is an example of what we currently have:

Umpire          DEVICE sum across ranks: 15293.4 GB
Umpire          DEVICE         rank max:   31.3 GB
Umpire       DEVICE::0 sum across ranks: 15293.4 G
Umpire       DEVICE::0         rank max:   31.3 GB
Umpire            HOST sum across ranks: 7272.2 GB
Umpire            HOST         rank max:   14.7 GB
Umpire HYPRE_DEVICE_POOL sum across ranks: 7178.4 GB
Umpire HYPRE_DEVICE_POOL         rank max:   14.8 GB
Umpire          PINNED sum across ranks:  256.3 GB
Umpire          PINNED         rank max:  589.6 MB

This is what I propose:

+-------------------+---------------+---------------+---------------+--------------+
|       Umpire Pool |    Min (GB/%) |    Max (GB/%) |    Avg (GB/%) |   Sum (TB/%) |
|                   |    over ranks |    over ranks |      per rank |    all ranks |
+-------------------+---------------+---------------+---------------+--------------+
|            DEVICE |   31.3  (49%) |   31.3  (49%) |   31.3  (49%) |  15.2  (24%) |
|              HOST |   14.7 (2.9%) |   14.7 (2.9%) |   14.7 (2.9%) |   7.2 (2.9%) |
| HYPRE_DEVICE_POOL |   14.8  (23%) |   14.8  (23%) |   14.8  (23%) |   7.1  (13%) |
|            PINNED |  0.589 (0.1%) |  0.589 (0.1%) |  0.589 (0.1%) |   0.3 (0.1%) |
+-------------------+---------------+---------------+---------------+--------------+

Describe alternatives you've considered N/A

Additional context Having access to minimum, maximum, and average utilization leads to a better understanding of load balancing.

victorapm avatar Feb 20 '23 22:02 victorapm

Yes, this would definitely help.

We could also add the number of ranks in the output log to be able to trace that info more easily.

jeannepellerin avatar Mar 02 '23 16:03 jeannepellerin

@MelReyCG you shouldn't put an issue in the ready status without filling the Impact / ... / component fields. Also, ready means that you're going to work on this soon. Is it what you meant when changing the status?

TotoGaz avatar Mar 02 '23 17:03 TotoGaz

I'm not sure what you're talking about, I don't think I did that myself, I just assigned myself to this issue because I will start working on it soon.

MelReyCG avatar Mar 02 '23 17:03 MelReyCG

Didn't you do image ?

TotoGaz avatar Mar 02 '23 19:03 TotoGaz

Yes I did, we moved it. I just filled the fields.

MelReyCG avatar Mar 03 '23 08:03 MelReyCG

@MelReyCG any update?

paveltomin avatar Sep 20 '23 23:09 paveltomin