Intermittent large IO throttling resulting from no shares being defined in db resource plans. (Doc ID 2858014.1)

Last updated on JULY 23, 2024

Applies to:

Symptoms

• In a High Capacity Cell, Flash intermittently gives multi-second queue time and displays inconsistent metrics in cell metrics output. 
  • Cell total throughput was less than PDB throughput
  • Large IOPS + SMALL IOPS != total Flash Cache IOPs
  • A single Consumer group or subset of consumer groups heavily doing large IO's have much larger queue time than other consumer groups which are doing OLTP / Small IOs.
• In most cases the cell was not exceeding capacity for the flash CACHE throughput numbers overall, and we did not see any flash DISK exceed 51% of it's capacity.
• Flash cache was 99% allocated to oltp and on average ~92% dirty.
• The database Resource Manager plan directives are configured all to 0's for the consumer groups(i.e. there is no P1/P2/P3, etc for MGMT_* or CPU_*)
• IORM objective is basic

Summarizing cell metrics for an example:

• PDB mypdb Throughput: MBPS=8919
  • Consumer Group mypdb.APP_GROUP Throughput: MBPS=143, Avg qtime=3.7ms
  • Consumer Group mypdb.BATCH_GROUP Throughput: MBPS=5012, Avg qtime=2005ms
  • Consumer Group mypdb.ETL_GROUP Throughput: MBPS=1089, Avg qtime=28.4ms
  • Consumer Group mypdb.OTHER_GROUPS Throughput: MBPS=2665, Avg qtime=34.9ms
• Cell Total Throughput: MBPS=125.132

• Pluggable Database: mypdb
• Flash Cache: IOPS=137866 Space allocated=11183215MB
  == Flash Device ==
  Utilization: Small=0.579% Large=38.221%
  Throughput: MBPS=8919
  Small I/Os: IOPS=1856 Avg qtime=0.0ms
  Large I/Os: IOPS=0 Avg qtime=0.0ms  <-- notice how the PDB has less large IO IOPs than it's consumer groups above

Changes

Cause

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