Use Cases

[Werner Keil]

Curious to learn, if something like this tool is possible on top of
Parfait:
http://www.appdynamics.com/lite.php?utm_source=InfoQ&utm_medium=Banner&utm_campaign=Free

And since it is a Monitoring tool, I assume, the question "Why Not Use
a Profiler?" can be answered similar for Parfait?;-)

[Paul Cowan]

Hi Werner,

On 13/09/10 04:35, Werner Keil wrote:
> Curious to learn, if something like this tool is possible on top of
> Parfait:
> http://www.appdynamics.com/lite.php?utm_source=InfoQ&utm_medium=Banner&utm_campaign=Free

Interesting, there's definitely some overlap there. Looks like it has
quite a similar approach in some ways, especially the (very clever
looking) auto-detection of meaningful 'business events' (struts actions,
spring controllers, etc). This is the path we've started to go down to
make our monitoring easier.

It's certainly vastly more developed than parfait in terms of polish and
usefulness for diagnostics; I guess the thing we use parfait for
primarily, though, is long-term data collection.

AppDynamics looks most useful (at least from the docs/videos) as a
'what's wrong now' tool. At Aconex, we primarily use parfait as a 'what
are the trends?' tool; by logging all our collected metrics to PCP (and
while we collect a lot of the parfait standard metrics we also have
hundreds of custom counter-values, etc -- e.g. # of user logins, # of
PDF conversions done, # of socket timeouts connecting to a remote
server...) we can really easily use the PCP tools to track trends over
time, do capacity planning, etc. Not sure if this is where AppDynamics
focuses.

That said, some of the stuff that's been added to parfait recently (e.g.
the in-progress event monitoring) is certainly moving more towards that
'instant diagnostics' goal rather than 'collect stats and work out what
to do with them later'.

> And since it is a Monitoring tool, I assume, the question "Why Not Use
> a Profiler?" can be answered similar for Parfait?;-)

Sure. This is partially answered above, but I guess for us the big
different is that parfait's goal (and it's not there yet, not by a long
shot) is that (like AppDynamics) we're aiming for low-overhead,
continual production data collection. We're also focused more on
meaningful business/technical counters (e.g. files converted to PDF, JMS
messages sent) in the scope of meaningful business transactions (e.g.
user logs in, user uses bulk-document-download functionality), rather
than just method executions (which may or may not map well to these
counters and transactions).

I think there's overlap, but parfait certainly gives us a lot more
flexibility (with a lot lower overhead) than simple profiling.

Thanks for the link, dying to try out that tool.

Cheers,

Paul