Default memory reserved for Platform function in All-in-one and worker should be reduced

Bug #1850209 reported by Tee Ngo
10
This bug affects 1 person
Affects Status Importance Assigned to Milestone
StarlingX
Fix Released
Medium
John Kung

Bug Description

Brief Description
-----------------
Currently the amount of memory reserved for platform function on the controller node is 16500 MiB (14500 MiB on numa0 and 2000 MiB on numa1) for All-in-one and 11000 MiB for low-core All-in-one by default. These are over allocated and should be reduced to 8000 MiB and 6000 MiB respectively to give more memory to user applications. The users can always allocate more memory to platform function if and when there's a plan to add more worker nodes.

Similarly, the amount of memory reserved for platform function on the worker node is 1000 MiB (8000 MiB on numa0 and 2000 MiB on numa1) by default. This should be reduced to 4000 MiB.

Severity
--------
Minor

Steps to Reproduce
------------------
N/A

Expected Behavior
------------------
Default resources allocation should be optimal

Actual Behavior
----------------
Default memory setting is not optimal for a platform without OpenStack services

Reproducibility
---------------
100% reproducible

System Configuration
--------------------
All-in-one simplex and duplex

Branch/Pull Time/Commit
-----------------------
Master

Last Pass
---------
N/A

Timestamp/Logs
--------------
N/A

Test Activity
-------------
Evaluation

Revision history for this message
Frank Miller (sensfan22) wrote :

This is a performance tuning LP. Marking stx.3.0 gating.

Changed in starlingx:
assignee: nobody → John Kung (john-kung)
importance: Undecided → Medium
status: New → Triaged
tags: added: stx.3.0 stx.config
Tee Ngo (teewrs)
description: updated
summary: - Default memory setting for All-in-one system should be reduced
+ Default memory reserved for Platform function in All-in-one and worker
+ should be reduced
Tee Ngo (teewrs)
description: updated
Changed in starlingx:
status: Triaged → In Progress
Revision history for this message
OpenStack Infra (hudson-openstack) wrote : Fix merged to config (master)

Reviewed: https://review.opendev.org/691992
Committed: https://git.openstack.org/cgit/starlingx/config/commit/?id=9e75a019eb6468b84a518d2872dcc8d040f21454
Submitter: Zuul
Branch: master

commit 9e75a019eb6468b84a518d2872dcc8d040f21454
Author: John Kung <email address hidden>
Date: Tue Oct 29 17:22:11 2019 -0400

    Reduce system configuration memory reserved defaults

    The AIO system configuration defaults are reduced:
      COMBINED_NODE_CONTROLLER_MEMORY_RESERVED_MIB 4000 MiB
      COMBINED_NODE_CONTROLLER_MEMORY_RESERVED_MIB_XEOND 3000 MiB
      COMBINED_NODE_CONTROLLER_MEMORY_RESERVED_MIB_VBOX 3000 MiB

    For workers:
      DISK_IO_RESIDENT_SET_SIZE_MIB 1000 MiB

    This results in default kubelet --system-reserved based on
    get_required_platform_reserved_memory() of:

    For AIO in MiB:
       node0: 7000 = 1000 (disk IO RSS) +
                     2000 (PLATFORM_CORE_MEMORY_RESERVED_MIB) +
                     4000 (COMBINED_NODE_CONTROLLER_MEMORY_RESERVED_MIB)
       nodeX: +=1000

    For AIO low-core in MiB:
       node0: 6000 = 1000 (disk IO RSS) +
                     2000 (PLATFORM_CORE_MEMORY_RESERVED) +
                     3000 (COMBINED_NODE_CONTROLLER_MEMORY_RESERVED_MIB_XEOND)
       nodeX: +=1000

    For worker:
       node0: 7000 = 1000 (disk IO RSS) +
                     2000 (PLATFORM_CORE_MEMORY_RESERVED_MIB) +
                     4000 (NETWORK_METADATA_OVERHEAD_MIB)
       nodeX: +=1000

    These defaults may be overridden via system host-memory-modify.

    Verify installation, application (stx-monitor) apply on AIO
    and Standard system with default system values.

    Change-Id: Ifd76b45b8a1a7632872874b31082b5ec591ac965
    Closes-Bug: #1850209
    Signed-off-by: John Kung <email address hidden>

Changed in starlingx:
status: In Progress → Fix Released
To post a comment you must log in.
This report contains Public information  
Everyone can see this information.

Other bug subscribers

Remote bug watches

Bug watches keep track of this bug in other bug trackers.