IBM Performance Report: SAP Banking on IBM Z with Shared Memory Communications (SMC-Dv2)

Abstract

IBM Z servers offer multiple network configuration options and are capable of hosting large SAP workloads in a single Central Electronic Complex (CEC), making these servers a viable platform for customers to implement server consolidation and co-location. The co-location of application servers and database server onto a single IBM Z CEC allow for internal cross memory communication, optimizing network performance and added security by not exposing data to an external network.

The latest IBM z16 can further optimize intra-system communication with Shared Memory Communications (SMC). The SMC architecture is an innovative communication protocol that preserves TCP/IP qualities of services and provides optimized communications allowing applications within separate operating system instances to communicate directly through shared memory. SMC protocol uses shared memory architecture which eliminates processing layers of the TCP/IP data path, thus may reduce CPU processing cost and latency. By switching to use SMC, the improvements can be realized without changing existing application software.

A variety of application workloads on IBM Z can benefit by utilizing SMC network connections. In this paper only the SAP application workloads are focused on and assessed.

To quantify the SMC performance benefit, the IBM SAP on Z Performance Team, located in Poughkeepsie, NY, conducted a series of experiments to evaluate SMC performance with customer-like workloads running on an IBM z16. Two SAP core banking workload types were used: Account Settlement (batch), and Day Posting for online transaction processing (OLTP). The SAP Banking Account Settlement workload was batch processing of account balancing. Key performance metrics of this workload are the batch elapsed time used to address critical batch window concerns many customers face, due to operational and regulatory requirements, and CPU utilization. The SAP Banking Day Posting workload was used to simulate customer OLTP activity and evaluate key performance metrics that include average transaction response time and CPU utilization. Less latency and lower CPU usage, from deploying SMC, helps system performance and may result in a processing cost reduction.

Additionally, cost savings can be improved by exploiting specialty engines like IBM Z Integrated Information Processors (zIIP) and Integrated Facility for Linux (IFL) processors. They are designed to help improve resource optimization and lower the cost for eligible workloads. zIIPs can be exploited by the SAP database server; and IFLs, by the SAP application servers running Linux on Z (LoZ). Offloading processing from general processors (GPs) to zIIPs on the SAP database server is an attractive option and recommended. The test environment utilizes both zIIPs and IFLs.

This paper documents these tests and findings. The measurements that were done were stress tests, not certified benchmarks.