jmeter-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From sebb <seb...@gmail.com>
Subject Re: PERFORMANCE => Changing JMeter defaults to ensure better performances by default
Date Thu, 10 Jan 2013 12:26:10 GMT
On 24 December 2012 13:20, Philippe Mouawad <philippe.mouawad@gmail.com> wrote:
> Hello,
>
> I am kind of annoyed of reading articles, blogs that say JMeter cannot
> perform high Load Tests, consumes lot of memory, generates OutOfMemory ...
>
> This has become a kind of "Urban Legend" partly due:
> - to issues that have been fixed for a while now
> - and partly In my opinion to some default configuration parameters that
> lead to these issues
>
> In my opinion, we should:
>
> 1) change these defaults to avoid new comers, beginners fall into all these
> traps and others check they are using it well:
>
>    - Save Service using XML output => Change to CSV

OK

>    - Distributed Mode that uses the Standard which is far from being the
>    best performing Sample Sender => Change to Batch or StrippedBatch

OK

> 2) Add warnings on GUIs of all elements that are more suited during
> Scripting than during Load Test :
>
>    - View Result Tree (I keep seeing people use this element during High
>    Load Test ! )
>    - View Results in Table
>    - Graph Results
>    - ...

Adding this information to the individual manual pages should be sufficient.

> 3) Add a popup warning when Start and Remote Start are clicked from GUI to
> encourage NON GUI mode use (we could add a checkbox Remind Me later which
> could be unchecked to avoid it again, but at least user would know about
> it).

That could quickly become very annoying.

> 4) Finally use some kind of visual indicator (RED background) on some
> options that have high impact on performance:
>
>    - Javascript as scripting language
>    - Body (unescaped) in Regular Expression Extractor (*this one is a real
>    performance killer !*)
>    - Encourage JSR223 Samplers + Groovy  + Caching instead of Beanshell
>    - ...

I don't think that is a good idea, as the relative performances may well change.

However, we can extend the "Lean and mean" documentation, so long as
we clarify to which versions the caveats apply.

>
> Maybe we should post this mail on user mailing list to see what users think
> about it.
>
> --
> Regards
> Philippe M.

Mime
View raw message