MiniCluster: Planning for and monitoring shared storage write utilization (Doc ID 2521607.1)

Last updated on APRIL 04, 2023

Applies to:

Symptoms

MiniCluster provides HA NFS shared storage as a means of sharing files between various MiniCluster virtual machines.  This storage is built upon ASM utilizing 6 x 8TB or 10TB HDDs included in the MiniCluster Storage Tray (12 disks if two trays are attached).  It is Mirrored and Striped storage but for write operations both mirrors must be written to in order for a write to be complete so you effectively have 3 HDDs bandwidth for writes.  Also there are slices of the same disks utilized for voting purposes in the KZ (OCRVOTE diskgroup) and in Customer DB VMs (SYSTEM diskgroup), so one cannot expect to obtain near peak write IOPs given likely long head seek times.

If a customer uses shared storage for heavy write operations, disk timeouts can occur.  In the worse scenarios, disks can be dropped from shared storage and one could lose shared storage until administrative remediation is performed.  In worst case scenarios, disks could be dropped from customer DB voting disk group and these DBs could be taken offline.

If these disk timeouts occur, one will see them in MC node 1 or node 2 global zone /var/adm/messages* files similar to this, and one should immediately take action to reduce write traffic to the shared storage mounts :

Cause

My Oracle Support provides customers with access to over a million knowledge articles and a vibrant support community of peers and Oracle experts.