Project

General

Profile

Feature #24441

Increase the number of Gridmanagers in the Factory

Added by Marco Mambelli 6 months ago. Updated about 2 months ago.

Status:
New
Priority:
Normal
Assignee:
-
Category:
Factory
Target version:
Start date:
05/19/2020
Due date:
% Done:

0%

Estimated time:
Stakeholders:

Factory Ops

Duration:

Description

In the pre 3.5 Factory, there was at least one Gridmanager per user.
With the new single-user Factory we may end up with a single Gridmanager for the whole Factory.

Condor by default has for each unix account a single gridmanager.
This is controlled by an expression that can be set in condor config: gridmanager_selection_expr - setting, knob for the schedd

A single Gridmanager could become a bottleneck when sending many jobs and it there are problems w/ an Entry or a type of Entries (e.g. w/ CREAM submission), then the whole Factory is subject to the problem.
Multiple Gridmanagers could help inproving scalability on the submit host.
But if there are multiple gridmanagers sent to the same Entry (remote site), then the benefit of communicating w/ a single Gridmanager start to drop.

A Factory could be split:
  • to have 1 gridmanager per site
  • to have one Gridmanager per Grid type

This is done already in the condor configuration of some Factories, maybe should be included in the GWMS configuration.

History

#1 Updated by Marco Mambelli 2 months ago

  • Target version changed from v3_6_4 to v3_6_5

#2 Updated by Marco Mambelli about 2 months ago

  • Target version changed from v3_6_5 to v3_6_6

Also available in: Atom PDF