Questions on JMX and management-config

Question

I was wondering if anyone can clarify the usage of management-config.
We've got a setup with managed-nodes=all and allow-remote-management=true (other params defaulted). We've got processes that occasionally ping the cluster (via a single node) to log performance statistics and to look for problems that might require intervention (across all nodes via remote management).
I'm trying to determine if the JMX subsystem is completely isolated from runtime cache performance.
1) Does the setting for the read-only flag impact performance in any way? If set to false, is there any sort of locking or synchronization that takes place when querying cluster nodes?
2) Do calls to JMX (and presumably the Invocation Service) share any resources with a Distributed Cache Service?
3) the Default management-config settings in Tangosol-coherence.xml point to the "Management" service - which is not actually defined elsewhere in the configuration file. Is this actually an implicit link to the Invocation Service?
4) Is there a single "Service Thread"? or does each distinct service have its own service thread?
Thanks in advance,
Dave

Answer 1

The performance impact of management is relatively small. The consumption is based on the information "read" (getAttribute/getAttribute) and the frequency of the poll. Polls of the same data element within 1 second will receive the same result for remote nodes.
The formula to determine the network consumption estimation formula for Coherence 3.3 on a single poll of the information is:
Estimated network bytes per poll =
Number of remote Node MBeans retrieved * 725
+ Number of remote Cache MBeans retrieved * 225
+ Number of remote Storage Manager MBeans Retrieved * 720
+ Number of remote Service MBeans Retrieved * 144
+ Number of remote Point to Point MBeans * (Number of Nodes in Grid -1 * 250)
Given the order N size of the Point-to-Point MBeans, It is not recommended to put that information in a loop for a large cluster.
1) Does the setting for the read-only flag impact performance in any way? If set to false, is there any sort of locking or synchronization that takes place when querying cluster nodes?
EW: The Read-only flag does not impact performance. There is no synchronization on read of the objects. Synchronization on update is on an attribute by attribute basis.
2) Do calls to JMX (and presumably the Invocation Service) share any resources with a Distributed Cache Service?
EW: The JMX calls are processed through the Management (Invocation) Service. This shares system (memory, CPU, network) resources with other services. The management information is not processed out-of-band and under extreme load performance will be impacted.
3) the Default management-config settings in Tangosol-coherence.xml point to the "Management" service - which is not actually defined elsewhere in the configuration file. Is this actually an implicit link to the Invocation Service?
EW: The Management service is a non-configurable Invocation service.
4) Is there a single "Service Thread"? or does each distinct service have its own service thread?
EW : Not exactly sure I understand the question. Each service has its' own "Service" thread and thread pool. However, there are some shared components at the TCMP (network) layer.
Hope this helps,
Everett Williams
Message was edited by:
user631017
Message was edited by:
user631017