eG Monitoring
 

Measures reported by BobiAPSPerfTest

One of the key factors influencing the performance of the Adaptive Processing Server (APS) is the usage of its JVM memory heap. This is because, APS is a pure Java based process, initialized with 1 GB of Java Heap. Naturally therefore, the lack of adequate free memory to the JVM, faulty and frequent garbage collections, and JVM deadlocks can all have an adverse impact on the health of the APS. Likewise, if critical services hosted on the APS – such as the BEx web application server and the Client auditing proxy server – are not correctly configured to handle the requests they receive, then again APS performance will degrade. This is why, the eG agent periodically runs the BobiAPSPerfTest test.

This test enables administrators measure JVM health and the correctness of the configuration of the critical APS services, thus helping them rapidly detect dips in APS performance and the possible reasons for it.

The measures made by this test are as follows:

Measurement Description Measurement Unit Interpretation
freeJVMMemory Indicates the amount of memory available to the JVM for allocating new objects. GB Ideally, the value of this measure should be high.
pctFreeMem Indicates the percentage of memory available to the JVM for allocating new objects. Percent A value close to 100% is a cause for concern, as it indicates rapid erosion of the JVM memory heap. Without sufficient memory, the APS and its services will not be able to operate optimally.
cpuUsagePCT Indicates the percentage of time the CPU was used by the APS during the last 5 mins. Percent This measure considers all processors allocated to the JVM.

A value close to 100% indicates excessive CPU usage, probably owing to CPU-intensive operations performed on the JVM. If more processing power is not allocated to the JVM, the APS may hang.
stoppedSystemPCT Indicates the percentage of time that APS services were stopped for Garbage Collection in the last 5 minutes. Percent A critical stage of garbage collection requires exclusive access and all APS services are halted at this time. This value should always be less than 10. The value of 10 and above indicates a low throughput issue and requires further investigation.
pageFaultsDuringGC Indicates the number of page faults that occurred while garbage collection was running during the last five minutes. Number Any value greater than 0 indicates a system under heavy load and low memory conditions.
jvmLockContention Indicates the current number of JVM lock contentions. Number This represents the number of synchronized objects that have threads that are waiting for access. The average value of this measure should be 0. Consistently higher values indicates threads that will not run again. You may want to take a thread dump to investigate such issues.
jvmDeadLockedThreads Indicates the number of active sessions within a Business Explorer Web Applications Service. Number These threads are indefinitely waiting on each other for a common set of resources. Average value should be 0. Consistently higher values warranties further investigation using thread dumps.
bExWASSessions Indicates the number of active sessions within a Business Explorer Web Applications Service. Number BEx Web applications are Web-based applications from the Business Explorer (BEx) of SAP NetWeaver Business Warehouse (BW) for data analysis, reporting, and analytical applications on the Web.

This measure is a good indicator of the load generated by this service on the APS.
auditEventRate Indicates the rate of client auditing events received within the last measure period. Events/Sec If the event is configured to be audited, the client sends the event information to the web application server, which passes it to the Client Auditing Proxy Service (CAPS) hosted in an Adaptive Processing Server (APS).

This measure is used to monitor the configuration and load on the Client auditing proxy service. Values greater than 0 indicate that the service has been configured correctly.
currAuditEventInQueue Indicates the number of auditing events that the APS has recorded, but which have not yet been retrieved by the CMS Auditor. Events/Sec If this number increases without bound, it could mean indicate that auditing has not been configured properly or that the system is heavily loaded and generating auditing events faster than the auditor can retrieve them. When stopping servers, It is advisable to disable them first and wait for auditing events to be fully retrieved and this queue becomes empty. Otherwise, they may be retrieved only when this server has been restarted and the CMS polls for them.
fullGC Indicates the rate of full garbage collections performed in the last 5 minutes. GCs/sec A rapid increase in this value may indicate a system under low memory conditions.