3293198 - SLES15 - Operating system directory cache - SAP HANA Performance degradation

***Consulting KBA, not a general best practice parameter recommendation***

The described symptoms were observed altogether on a physical 12TB HPE superdome 280 flex appliance during an initial data migration using AMDPs. However, all these symptoms may also occur on virtualized environments as the root cause is related to the VFS layer of the operating system.

• AMDPs are used to load the data into the application tables.
• Data loading gradually slowing down - SAP HANA threads are in waiting status.
• Data load jobs are still running, not getting canceled.
• Hundreds/thousands of HANA delta merges are executed (this case mainly the partitioned ACDOCA was merged frequently).
• HANA delta merge is much slower than expected (e.g.: instead of 2-3 minutes, 15-20 minutes).
• No errors are shown in HANA traces or OS logs.
• xfs filesystems are used for the HANA data and log volumes.
• Command sar -v shows a gradual increase in dentunusd and inode-nr to the millions (sometimes also file-nr increase is observed).
• Command cat /proc/sys/fs/procfs-drop-fd-dentries returns 0.
• A host reboot temporarily solves the performance problem of the HANA database.

Example output of sar -v command from the affected HANA appliance

• SLES15 SP2
• SAP HANA 2.0 rev. 65
• SAP S/4HANA 2021 FPS01
• 12TB HPE superdome 280 flex appliance
• Linux kernel 5.3.18-24.86

dentunusd, SLES15, VFS, directory cache, procfs-drop-fd-dentries, /proc/sys/vm/drop_caches , KBA , BC-OP-LNX-HPE , Linux on HPE Hardware , HAN-DB-PERF , SAP HANA Database Performance , Problem