PM82820: MULTI-CELL ELASTICITY OPERATIONS DO NOT FUNCTION WHEN USING APC.PREDICTOR=CPU

Fixes are available

APAR status

Closed as program error.

Error description

When using APC.predictor=CPU, the power allocation does not
account for CPU overload protection.

Local fix

Problem summary

****************************************************************
* USERS AFFECTED:  All users of the application placement      *
*                  controller in WebSphere Virtual Enterprise  *
****************************************************************
* PROBLEM DESCRIPTION: This APAR corrects problems with the    *
*                      application placement controller CPU    *
*                      predictor, multi-cell configurations,   *
*                      and AIX/POWER uncapped partitions.      *
****************************************************************
* RECOMMENDATION:                                              *
****************************************************************
CPU predictor
1) When using the CPU predictor (APC.predictor=CPU) in a
multi-cell topology, the demand information for a cluster in a
point cell is not sent to the application placement controller
(APC) in the center cell. This prevents the APC from starting
additional cluster members when demand increases for the
clusters in the point cell.
Elasticity
2) When using APC elasticity with AIX/POWER uncapped
partitions, elastic adds and removes occur based on the total
available capacity of a node (partition), which includes
entitled capacity and shared capacity. However, the available
shared capacity can change quickly and leave a cluster
significantly below its required capacity.
3) After an elasticity operation completes, a waiting period
follows, as configured in the "minimum time between placement
change" setting. This time period allows the topology to
adjust to the recently completed elasticity operation before
initiating another operation. However, during certain timing
conditions, the sleep period might be incorrectly interrupted,
and placement decisions are initiated prematurely.
4) Fundamentally, elasticity add operations occur based on
demand exceeding the capacity. There are additional constraints
that allow or prevent elasticity operations, for example
maximum instances. An unnecessary constraint incorrectly
checks for demand information from non-production servers, for
example node agent. During certain conditions, the unnecessary
constraint can prevent elasticity add operations from
occurring.
Cluster capacity
5) APC uses cluster demand statistics from autonomic request
flow manager (ARFM) to determine the percentage of the node
capacity that should be allocated to each cluster member of a
node. If a cluster does not have any demand, the APC uses an
alternate calculation to determine the cluster capacity. The
wrong cluster capacity calculation is used for clusters with
demand and results in low cluster capacity calculation, which
can lead to unnecessary queuing. This issue occurs when at
least one cluster has no demand, and the remaining clusters
have demand. The clusters with demand incorrectly use the
capacity calculation of the clusters without demand.

Problem conclusion

The following code changes were made to resolve the listed
limitations:
1) Demand information is now correctly sent from point cells
to the center cell when using CPU predictor.
2) Elasticity adds on AIX/POWER uncapped partitions now occur
based on entitled capacity. When demand exceeds the entitled
capacity of a node (partition), an elasticity add operation is
initiated.
3) The timing condition was corrected to prevent the
elasticity sleep period from being interrupted.
4) The unnecessary constraint that prevented elasticity add
operations from occurring was removed.
5) The incorrect cluster capacity calculation is no longer used
on the wrong clusters.
NOTE: Background CPU and CPU overload protection (COP) values
are now included in the apc.log file to provide a better view
of the conditions in a topology.

This fix will be included in the next WebSphere Application
Server V.8.5 fix, and in the next fix for WebSphere Virtual
Enterprise V.7.0.
Refer to the Recommended Updates page for delivery information:
http://www.ibm.com/support/docview.wss?rs=180&uid=swg27004980

Temporary fix

Comments

APAR Information

APAR number
PM82820
Reported component name
WAS VIRTUAL ENT
Reported component ID
5725C9203
Reported release
700
Status
CLOSED PER
PE
NoPE
HIPER
NoHIPER
Special Attention
NoSpecatt / Xsystem
Submitted date
2013-02-14
Closed date
2013-05-13
Last modified date
2018-10-17

APAR is sysrouted FROM one or more of the following:
APAR is sysrouted TO one or more of the following:

Fix information

Fixed component name
WAS VIRTUAL ENT
Fixed component ID
5725C9203

Applicable component levels

R700 PSY
UP

[{"Business Unit":{"code":"BU059","label":"IBM Software w\/o TPS"},"Product":{"code":"SSUP64","label":"WebSphere Virtual Enterprise"},"Platform":[{"code":"PF025","label":"Platform Independent"}],"Version":"7.0","Line of Business":{"code":"LOB45","label":"Automation"}}]

Document Information

Modified date:
12 January 2022

Tips

PM82820: MULTI-CELL ELASTICITY OPERATIONS DO NOT FUNCTION WHEN USING APC.PREDICTOR=CPU

Fixes are available

Subscribe

APAR status

Closed as program error.

Error description

Local fix

Problem summary

Problem conclusion

Temporary fix

Comments

APAR Information

APAR number

Reported component name

Reported component ID

Reported release

Status

PE

HIPER

Special Attention

Submitted date

Closed date

Last modified date

APAR is sysrouted FROM one or more of the following:

APAR is sysrouted TO one or more of the following:

Fix information

Fixed component name

Fixed component ID

Applicable component levels

R700 PSY

Document Information

Share your feedback

Need support?