Re: [clif] About probes

[Bruno Dillenseger <bruno.dillenseger@xxxxxxxxxxxxxxxxxxxx>]

Emmanuel Cecchet:

> I quickly looked at CLIF source code and I was particularly interested 
> in the probe code.
> I would like to know why all probes (cpu, mem, ...) are merged inside 
> the same component that samples all values at fixed intervals?
> Wouldn't it be better to have a probe to monitor a single ressource? 

Both the implementation and concept of probe in CLIF will have to be 
re-engineered (see below).
This suggestion sounds good. Any extra thoughts are welcome.

> Another question that arises from internal discussions is: shouldn't 
> we extract the monitoring infrastructure in a separate project? 

Definitely yes. The concept of probe should be:
(1) generalized, in the following sense: a probe monitors a "resource" 
of any kind: system (hardware/OS), middleware, application...;
(2) decoupled from CLIF, since resource monitoring need is more general 
than the performance test area. However, I would like to keep the idea 
of considering probes as Fractal components that can be (un-)deployed 
during runtime.

Hence, this discussion has much to do with the question of distributed 
system administration. How to abstract this notion of probe, with which 
conceptual framework? Which interfaces to define? Shouldn't we get 
inspiration from management standards like SNMP, JMX...?
Sure there is a place here for a standalone project...

Best regards,
-- Bruno.