In a previous article we showed the numerous communication flows that can occur between the different SAP systems which can impact sizing and performance of SAP systems. In this article we investigate these flows in a hybrid cloud scenario where the SAP landscape is split between on-premise and a public cloud (VMware Cloud on AWS). We will show some workload examples and introduce a solution to help monitor the SAP hybrid cloud.
VMware Cloud on AWS is not supported for SAP workloads at this time. Please stay tuned to https://twitter.com/vmwarecloudaws .
VMware Cloud on AWS can extend the on-premise data center to the cloud and enable easy migration of application workloads to VMware Cloud on AWS without conversions. With a phased migration of a large SAP landscape it is possible that some systems may be split between the cloud and on-premise and in this situation performance of communication flows between the systems may have an impact on SLAs.
Lab Setup
The goal of this investigation was to test various use cases where components/services of an SAP system were split between on-premise and cloud. The following diagram depicts the lab environment and the virtual machines (VMs) that were deployed.
Lab Environment
We tested with two SAP systems:
- SAP system 1 (ECC 6.0, SAP NetWeaver 7.4, SAP HANA) consisting of two VMs: database and application server in separate VMs
- SAP system 2 (SAP NetWeaver 7.4, Oracle 11.2) consisting of one VM: database and application server in the same Guest OS
- NFS service deployed in a single VM
- VMware Cloud on AWS SDDC Version 1.8v2
VMware Cloud on AWS enables migration of VMs between on-premise and cloud without change of the IP address – the on-premise network is stretched to the cloud SDDC.
Test Description
The following communication flows were tested where SAP systems or components/services of an individual system were split between cloud and on-premise.
Results
The following table summarizes the results of the use cases described above.
Note: the workload tests above are lab examples, mileage is expected to vary.
Results show a negative performance impact when communication flows need to cross the network from on-premise to cloud. With the RFC use case two custom programs were executed and exhibited different results because while both programs were designed to retrieve the same amount of data, program B included other processing logic that consumed more time i.e. it was less sensitive to network performance. The impact of communication flows over a long-distance network connection is dependent upon the workload and program logic. For example, program logic can be tuned to minimize the amount of data retrieved which can reduce network traffic – instead of retrieving the complete record from a business table, a program can be designed to extract only the relevant fields.
Conclusion – Migration Considerations
A long-term migration plan to cloud may result in a split of the SAP landscape between on-premise and cloud. The results here show that splitting components/services of SAP systems between a long-distance network connection can have varying impact on performance metrics. Hence a good understanding of communication flows / dependencies between SAP VMs should be factored into a cloud migration project. For example, VMs with heavy communications between them may benefit from being co-located in the same location, on-premise or cloud so they should be migrated together. VMware provides a solution that can help with this situation – vRealize Network Insight which is described in this article. This article describes how SAP communications flows can be discovered and monitored in the on-premises datacenter with the vRealize Network Insight appliance. The latest release of the appliance has integration with VMware Cloud on AWS which can provide network visibility of the SAP landscape across private, public and hybrid clouds – the following screen capture shows this functionality.
vRealize Network Insight Installed on-premise – monitoring communication flows between on-premise and VMware Cloud on AWS
The above screen capture is based on a preliminary test – more details will be provided in a future article.
Thanks to Todd Muirhead for his reviews/inputs.