Symptom
Initial performance observations for SAP 7.40 Process Orchestration
on HANA (SAP BPM 7.40 on HANA, SAP 7.40 Process Integration Advanced
Adapter Engine on HANA)
Please refer to SAP Note 2039820 (Sizing HANA Memory for SAP Process Integration Advanced Adapter Engine (PI-AEX) , Business Process Management (BPM), and Process Orchestration) if you want to perform HANA memory sizing.
Please refer to SAP Note 2039820 (Sizing HANA Memory for SAP Process Integration Advanced Adapter Engine (PI-AEX) , Business Process Management (BPM), and Process Orchestration) if you want to perform HANA memory sizing.
Other Terms
Sizing, resource requirements, performance, update, hardware, BPM, PI-AEX, Process Orchestration on HANA
Reason and Prerequisites
Customers plan hardware expenditures prior to implementing SAP 7.40 Process Orchestration on HANA
Customers optimize system performance for their own specific use-cases and scenarios
Understanding of performance characteristics and hot-spots helps the optimization of performance
Customers optimize system performance for their own specific use-cases and scenarios
Understanding of performance characteristics and hot-spots helps the optimization of performance
Solution
Overview of performance on HANA, basic characteristics and
expectations (such as % of load on APP/database, performance during
job/archiving runs, differences due to primary use-case (such as PO, BPM
w/OPint, etc..)
HANA Performance Factors
savepoint_interval – HANA uses savepoints to periodically write a consistent image of the in-memory database to the file system. The savepoint is a write intensive operation that benefits significantly from fast disk I/O. During the critical phase of the savepoint there is a lock on database updates. It is important to keep this lock duration at a minimum to avoid impacting the system operation. The standard savepoint_interval is 5 minutes (300 seconds). For high-volume scenarios it might be necessary to reduce the interval in order to reduce the amount of data written during each savepoint. Adjustments to the interval should be evaluated in your own landscape as the performance impact is dependent on the physical hardware infrastructure. Tuning of the savepoint is similar in concept to tuning of Java garbage collection. With more frequent savepoints you have more pauses but of a shorter duration (tens of milliseconds). With less frequent savepoints you have fewer pauses but they are longer in duration (tens or hundreds of milliseconds). You can check the savepoint performance using the HANA system view M_SAVEPOINTS which provides current and historical savepoint statistics. The CRITICAL_PHASE_DURATION is the time during which the database is locked. The value is given in microseconds. It should ideally be only a few milliseconds. Longer durations suggest an I/O bottleneck. Longer durations might also suggest that the interval is too long and thus the data which needs to be flushed to disk is too large.
You can check the savepoint performance using the HANA M_SAVEPOINTS view. Please refer to the current HANA manuals for further details. http://help.sap.com/hana_platform
Disk I/O Capacity – Your disk I/O capacity will directly impact the HANA performance. The use of high I/O disk systems will greatly improve the performance of SAP HANA and the applications. Disk write operations in SAP HANA occur during log writing and during savepoints. You should carefully monitor your system I/O statistics to ensure you do not have bottlenecks which will impact your performance. It is normal to see spikes in I/O operations and Disk Write volume at the period intervals of the savepoints. If during the savepoint your I/O exceeds 80-90% of capacity for more than a few milliseconds then there may be an impact to users.
Storage of LOBS with SAP HANA – As of SAP HANA SP7 you can configure columns which contain large data object to be stored on disk rather than in memory. This can significantly reduce the memory used by the application schema within the HANA database. SAP recommends that the PI table BC_MSG be configured to use disk storage for all messages in order to keep conserve the HANA memory. The related note 2039873 Table classification and index optimizations and Memory LOBS for SAP Process Integration Advanced Adapter Engine (PI-AEX) , Business Process Management (BPM), and Process Orchestration provides an attachment which configures the LOB setting per SAP recommendation on several key tables.
Related Documentation:
SAP HANA SQL and System Views Reference Guide
SAP HANA Technical Operations Guide
SAP HANA Memory Usage Explained
SAP HANA Troubleshooting and Performance Analysis
HANA Performance Factors
savepoint_interval – HANA uses savepoints to periodically write a consistent image of the in-memory database to the file system. The savepoint is a write intensive operation that benefits significantly from fast disk I/O. During the critical phase of the savepoint there is a lock on database updates. It is important to keep this lock duration at a minimum to avoid impacting the system operation. The standard savepoint_interval is 5 minutes (300 seconds). For high-volume scenarios it might be necessary to reduce the interval in order to reduce the amount of data written during each savepoint. Adjustments to the interval should be evaluated in your own landscape as the performance impact is dependent on the physical hardware infrastructure. Tuning of the savepoint is similar in concept to tuning of Java garbage collection. With more frequent savepoints you have more pauses but of a shorter duration (tens of milliseconds). With less frequent savepoints you have fewer pauses but they are longer in duration (tens or hundreds of milliseconds). You can check the savepoint performance using the HANA system view M_SAVEPOINTS which provides current and historical savepoint statistics. The CRITICAL_PHASE_DURATION is the time during which the database is locked. The value is given in microseconds. It should ideally be only a few milliseconds. Longer durations suggest an I/O bottleneck. Longer durations might also suggest that the interval is too long and thus the data which needs to be flushed to disk is too large.
You can check the savepoint performance using the HANA M_SAVEPOINTS view. Please refer to the current HANA manuals for further details. http://help.sap.com/hana_platform
Disk I/O Capacity – Your disk I/O capacity will directly impact the HANA performance. The use of high I/O disk systems will greatly improve the performance of SAP HANA and the applications. Disk write operations in SAP HANA occur during log writing and during savepoints. You should carefully monitor your system I/O statistics to ensure you do not have bottlenecks which will impact your performance. It is normal to see spikes in I/O operations and Disk Write volume at the period intervals of the savepoints. If during the savepoint your I/O exceeds 80-90% of capacity for more than a few milliseconds then there may be an impact to users.
Storage of LOBS with SAP HANA – As of SAP HANA SP7 you can configure columns which contain large data object to be stored on disk rather than in memory. This can significantly reduce the memory used by the application schema within the HANA database. SAP recommends that the PI table BC_MSG be configured to use disk storage for all messages in order to keep conserve the HANA memory. The related note 2039873 Table classification and index optimizations and Memory LOBS for SAP Process Integration Advanced Adapter Engine (PI-AEX) , Business Process Management (BPM), and Process Orchestration provides an attachment which configures the LOB setting per SAP recommendation on several key tables.
Related Documentation:
SAP HANA SQL and System Views Reference Guide
SAP HANA Technical Operations Guide
SAP HANA Memory Usage Explained
SAP HANA Troubleshooting and Performance Analysis
Header Data
| Released On | 15.08.2014 13:57:21 | ||
| Release Status | Released for Customer | ||
| Component | BC-BMT-BPM-SRV Process Server | ||
| Other Components |
| ||
| Priority | Recommendations / Additional Info | ||
| Category | Performance |
No comments:
Post a Comment