DX Application Performance Management

Hi, is there anyone out there who’s an expert on Introscope EPA and using it with network sources?

We’re experiencing a logical limit on the number of network connections that can send data to one instance of EPA. (Ours are long-running socket connections from IBM mainframe systems, sending metrics in XML format - not the HTTP Get connections.)

Having any 2 mainframe systems connected at once works fine, and Investigator shows the metrics from both systems.


When we try and connect a 3rd mainframe system to EPA, this is what happens:

-
Physically, the EM server accepts the TCP connection to the EPA port.
Netstat on the EM server shows the connection between the 3rd mainframe and the EPA port with a status of ESTABLISHED, the same as the first 2 successful connections are.
Packet trace of this connection shows metrics in correct XML format being sent to EM every 15 seconds, and EM responding with an Ack to each of these sends.

-
However, logically, the EPAgent Network Server never becomes aware of this 3rd TCP connection. For the first 2 connections, a corresponding message like this appears on the EPAgent log:
[EPAgent] Network Server processing connection from Socket[addr=/141.202.65.31,port=46035,localport=8000].[font]
For the 3rd connection, this message never appears, and the data that arrives over this connection is never processed by EPA.


We're not sure that this limit of metrics from only 2 connections is deliberate behaviour, and we definitely don’t want to require users to install multiple EPAs.

-
Introscope talks about its scalability so much, how one EM can handle thousands of agents simultaneously etc etc – so why would an EPA instance be limited to 2 connections ??
-
The TCP listener for EPA is accepting more than 2 simultaneous connections and is acknowledging the data coming over the 3rd connection – why would it do this, if it didn't intend to process that data?



Does anyone have any comments on this EPA behaviour? Many thanks from a team of Introscope newbies….

Hi Wendy,

We have seen the same limitation. I suspect it is because CA want to you to by more EPagents.

EPAgents are free to deploy. You pay per CPU (2 cores = 1 CPU; it may be different depending on your agreement) for your Java and .NET agents.

Hi Wendy,

We ran into the same problem as well. You can send in as many connection requests as you want but the EPAgent thread pool that processes the request does have a size limit of 2. There doesn't appear to be a way to tweak the number of threads but by increasing the number of plugins, you can trick the Wily EPAgent to think that it’s got more to process ( and hence its spawning additional threads ). The down side is that you will get some additional metrics under plugins node which you can reduce by writing your own plugin and sending just one metric.

Jay