AIX

AIX

Connect with fellow AIX users and experts to gain knowledge, share insights, and solve problems.

 View Only

Smarter Workload Placement on AIX: The Role of ASO in Affinity and Clustering

By Yasser Sait posted yesterday

  

Efficient workload placement in AIX is critical for maximizing performance for user workload on multi-srad or multi-node systems. IBM’s Active System Optimizer (ASO) introduces strategies like cache affinity and memory affinity to improve job locality and system throughput. This blog explores the new feature in these strategies to suit them better for customer workloads. 

  

AIX's Active System Optimizer (ASO) previously treated each shared memory segment (SHM) and each process as separate jobs. This isolated view limited placement strategies from understanding inter-process communication via shared memory. 

To solve this, a SHM clustering mechanism has been introduced. 

 

How It Works 

 

  • ASO now identifies processes that access multiple SHM segments across job boundaries. 

  • These jobs are clustered together based on their shared memory interaction graph. 

  • The processes of the cluster are placed on a single SRAD or node there by benefitting from reduced cache misses. 

  • After the processes are placed, they are profiled to remote memory accesses using PMU. 

  • If the PMU data indicates good amount of cross node access, the memory is migrated to the domain where the processes were placed. 

  • The cluster is now affinitized to a single SRAD or node within a multi-node LPAR, improving data locality. 

Example 

Imagine 8 processes interacting across 4 SHM segments. If P2 accesses both SHM1 and SHM2, these jobs are now grouped and placed together to reduce cross-node traffic. 

  

Picture 1, Picture 

  

  

  

Benefits 

  

Clustering based on shared memory and affinitizing workloads to a common domain significantly enhances data locality, leading to tangible performance improvements. Performance engineering tests have shown: 

  • An approximate 11% gain in SAPS/CPU Util for SAP-SD workloads 

  • Up to a 15% improvement in TPS for DT7 workloads 

These gains highlight the effectiveness of ASO’s clustering and affinity strategies in real-world, performance-sensitive environments. 

 

 Usage 

  

This optimization is enabled by default. ASO automatically identifies a domain large enough to accommodate all threads within a cluster, ensuring balanced placement without manual intervention. 

By design, ASO limits this clustering logic to processes that share memory segments, ensuring it focuses on workloads that benefit most from locality-aware scheduling. 

  

  

Final Thoughts 

  

This enhancement represents a meaningful advancement in how AIX, through ASO, intelligently places workloads across system resources: 

  • Default cache and memory affinity provide improved performance out of the box. 

  • SHM clustering introduces awareness of inter-process communication patterns. 

  • Cluster-based metric aggregation enables smarter, holistic workload placement decisions. 

Together, these improvements deliver a more efficient and scalable AIX environment—especially for workloads where memory access patterns directly influence performance. 

 

0 comments
1 view

Permalink

https://community.ibm.com/community/user/blogs/yasser-sait/2025/08/13/smarter-workload-placement-on-aix-the-role-of-aso