Overview
This is a typical engagement for a database operations team that runs SAP HANA and sees memory use climb toward the allocation limit, with slow periods at month end. The team has to decide between buying memory and changing how the database is used, and wants measurements to decide on. AceMQ engineers work at the database layer with partner SAP HANA subject-matter experts.
Challenge
HANA is an in-memory database and takes memory up to a limit, so high use alone does not say much. What matters is what sits under the limit: the largest tables, the column-store deltas, and the headroom a delta merge needs, which SAP says is roughly the current main, the new main and the current delta at once. Slow periods can also come from savepoints on slow storage or from a few expensive statements, which more memory would not fix.
Environment
A SAP HANA 2.0 system with a system database and tenant databases, on-premises or in the customer's own cloud account, a SAP HANA cockpit installation, a monitoring tool that holds some history, and a team that runs the database but did not size it.
Approach
AceMQ engineers read the configuration and state with the customer's DBAs and change nothing. They compare global_allocation_limit with ALLOCATION_LIMIT and use in M_HOST_RESOURCE_UTILIZATION and M_SERVICE_MEMORY, break use down by service, component and table in the cockpit memory analysis, and read delta sizes and merge history in M_CS_TABLES and M_DELTA_MERGE_STATISTICS. Savepoint duration is compared with storage latency, and expensive statements are listed for the periods the team names. Questions that only SAP can answer are marked as questions for SAP.
Solution
- Memory picture: allocation limit against use, per service, plus the largest tables and components
- Delta and merge review: large deltas, merge history and the memory headroom a merge needs
- Savepoint duration and storage latency read together, with the periods that stall
- Ranked list of expensive statements for the slow periods, with what each depends on
- Sizing recommendation with a growth projection and the measurements behind it
- List of items that need SAP's answer, kept apart from the database findings
Outcome
The engagement delivers a written review: the memory picture, the tables and merges that drive it, the savepoint and storage findings, the expensive statements, a sizing recommendation and a short list of questions for SAP. The customer then decides whether to add memory, change the workload or both. AceMQ does no SAP application, Basis or ABAP work and gives no SAP licensing advice.