Jprofiler perfino...

[Kevin Burton]

Has anyone here used Jprofiler's Perfino in production?

https://www.ej-technologies.com/products/perfino/overview.html

The pricing seems decent to run with one server per role, maybe with less load (to account for profiling overhead).

We would probably end up paying 2k for our entire stack (500 per role).

[Kirk Pepperdine]

it comes with the same overhead as other profilers. Depending on what you’re looking for I’d suggest you construct your own targeted profiler.

— Kirk

--
You received this message because you are subscribed to the Google Groups "mechanical-sympathy" group.
To unsubscribe from this group and stop receiving emails from it, send an email to mechanical-symp...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Trask Stalnaker]

If you are looking at APM tools, I’d love to get your feedback on the open source project I’ve been hacking on for the past too many years https://glowroot.org.

I’ve learned lots from following this list that has helped me reduce monitoring overhead over the past year.  You can see current overhead benchmark results at https://glowroot.org/overhead.html.

If anyone has comments on improving the benchmark, or knows of any existing benchmarks in the APM space, that would be very helpful.  I realize monitoring overhead can depend on so many factors, but think it's still important to build solid benchmarks and publish concrete results.

Thanks,

Trask

--

[Kevin Burton]

On Friday, February 27, 2015 at 1:37:33 AM UTC-8, Kirk Pepperdine wrote:

it comes with the same overhead as other profilers. Depending on what you’re looking for I’d suggest you construct your own targeted profiler.

Huh.  I was hoping it was smart and didn't require the overhead and instead did something similar to Java Mission Control. 

I like the idea of having one node that runs 1/10th (or so) of your load but has the profiler enabled. 

[Ben Evans]

If it has 0.1X of your load, then any statistics it generates are quite probably lying to you.[0]

Ben
[0] Statement contains nuts &/or zealotry. If you need clarification of this footnote then these aren't the droids you are looking for.

--

[Justin Mason]

I like the look of this -- 2% overhead is a pretty acceptable hit!  Nice dashboarding too.  I could see myself putting this on 1 server in an LB fleet to allow profile captures in production.

It'd be great to get more of an idea of performance impact in a production scenario.  ie. what impact does it typically have in GC terms?  when does it flush to disk (if it does flush to disk)? etc.

--j.

[Trask Stalnaker]

The memory footprint is generally around 20mb (runtime weaving state + non-flushed aggregates + embedded h2 database cache).

As far as the GC impact, the benchmark sustains 90 requests/second for 20 minutes (CPU utilization at around 80%), so I think the GC impact is more or less already included in the benchmark numbers.  Though I agree it would be useful to know what portion of the overhead is due to GC impact.  I’ll capture total young and old gc times in the benchmark harness next time I run it.

Aggregate data is flushed to disk once a minute.  Individual transaction traces that exceed their threshold are flushed to disk immediately on completion.  The benchmark harness is not currently capturing disk I/O but that’s a good idea, I’ll add that also next time I run it.

Thanks,

Trask