PowerScale 9.x performance impact on mixed node clusters with archive nodes

Performance issue on OneFS 9x clusters containing archive nodes and one other node type.

For Example:
A cluster that contains both A200(0) archive nodes and H500 nodes.
A cluster with both A2000 archive nodes and F800 nodes
A cluster with A200 archive nodes, F800 nodes, and H400 nodes

PSCALE-140986

Setting a system flag on archive tiers can cause performance degradation on mixed node clusters. 

To verify:

Determine which disk pool the ifsvar directory is located:

# isi get -D /ifs/.ifsvar/isi_license.xml | grep pools

Precheck:

Prior to completing the resolution steps in this Knowledge Base article, check the size of the audit logs. Verify whether there is sufficient free disk space on the none-archive node pools for the audit logs.

This command checks Disk Usage size of the Audit logs:

# du -sh /ifs/.ifsvar/audit/logs

If the size of the Audit logs causes the none-archive node pool to fill up, follow the steps in KB Powerscale: How to Remove Audit Log Filesto purge the Audit Logs. 

=============

Resolution:
Identify which node pool is configured as System. System node pools are where the contents of the Ifsvar are located. This can be in multiple pools.

# disi -I diskpools ls -v 

Remove the System flag from the Archive node pool:

# isi_gconfig smartpools.diskpools.manually_manage_system_flags=true
# disi -I diskpool modify <diskpool_name> --system=false

After changes have been made, Run a SmartPools or SmartPoolsTree job to move the Ifsvar directory into the nonarchival node pools.

Command to use SmartPoolsTree to move the data:

# isi job start smartpoolstree --paths=/ifs/.ifsvar

If SmartPools is not licensed, run a SetProtectPlus job instead.

IMPORTANT NOTE:
The above SmartPoolsTree job only moves the /ifs/.ifsvar and its contents.
To move other SYSTEM LINs and SBTs, run a SmartPools job to completion - which could take longer. 

# isi job start smartpools