Performance analysis and monitoring of a Forteapplication

Question

Hello,
It would be good if one could do some performance analysis and monitoring of
a Forte application at production time.
By performance analysis I mean measure the time some selected methods take
to return. In a CS application such a method would be some selected method
of a key remote service representative of the application's activity.
One would like to mesure the min, max, average time, and also give a
threshhold which when reached will automatically generate an alarm or some
pre-defined processing.
The most powerful way would be to use a SNMP application through a Forte -
SNMP gateway. (see the G. Puterbaugh paper "Building a Forte-SNMP gateway"
in the 96' Forte Forum proceedings).
But before going that far some simple means accessible through EConsole
would already be great.
A collegue of mine when to that Forum and reported that Puterbaugh said that
such an agent is currently missing, but that its implementation is not
difficult.
I looked at all the agents and their instruments. I came to the following
conclusions :
1) instrumented data are available at the granularity level of a partition,
not at a smaller granularity. For example, the DistObjectMgr agent gives you
very useful information : the number of events (sent/received) and the
number of (remote) methods (called/invoked), but this for the entire
partition. Thus it prevents making tuned observations (unless you partition
in a special way your application to put in a dedicated partition the thing
you want to observe and only this thing).
2) there is no instrumented data related to processing time.
This leads me to the point that no information observed by the standard
agents help me figuring out my performances. Thus I have to add at
development time some lines of code to the methods I potentially want to
observe later at production time to generate the appropriate information a
custom agent will then display (process) with the appropriate instruments.
Does someone share this position ?
Has someone implemented such an agent and assotiated means ?
PS: I will probably implement my own one if no other way around.
best regards,
Pierre Gelli
ADP GSI
Payroll and Human Resources Management
72-78, Grande Rue, F-92310 SEVRES
phone : [phone] (direct) [phone] (reception desk)
fax : [phone]

Answer 1

Hello,
It would be good if one could do some performance analysis and monitoring of
a Forte application at production time.
By performance analysis I mean measure the time some selected methods take
to return. In a CS application such a method would be some selected method
of a key remote service representative of the application's activity.
One would like to mesure the min, max, average time, and also give a
threshhold which when reached will automatically generate an alarm or some
pre-defined processing.
The most powerful way would be to use a SNMP application through a Forte -
SNMP gateway. (see the G. Puterbaugh paper "Building a Forte-SNMP gateway"
in the 96' Forte Forum proceedings).
But before going that far some simple means accessible through EConsole
would already be great.
A collegue of mine when to that Forum and reported that Puterbaugh said that
such an agent is currently missing, but that its implementation is not
difficult.
I looked at all the agents and their instruments. I came to the following
conclusions :
1) instrumented data are available at the granularity level of a partition,
not at a smaller granularity. For example, the DistObjectMgr agent gives you
very useful information : the number of events (sent/received) and the
number of (remote) methods (called/invoked), but this for the entire
partition. Thus it prevents making tuned observations (unless you partition
in a special way your application to put in a dedicated partition the thing
you want to observe and only this thing).
2) there is no instrumented data related to processing time.
This leads me to the point that no information observed by the standard
agents help me figuring out my performances. Thus I have to add at
development time some lines of code to the methods I potentially want to
observe later at production time to generate the appropriate information a
custom agent will then display (process) with the appropriate instruments.
Does someone share this position ?
Has someone implemented such an agent and assotiated means ?
PS: I will probably implement my own one if no other way around.
best regards,
Pierre Gelli
ADP GSI
Payroll and Human Resources Management
72-78, Grande Rue, F-92310 SEVRES
phone : [phone] (direct) [phone] (reception desk)
fax : [phone]