Eclipse Community Forums: Remote Application Platform (RAP)

Home » Eclipse Projects » Remote Application Platform (RAP) » Performances for huge ERP

Performances for huge ERP [message #108827]

Thu, 09 October 2008 08:36

Eclipse User

Hi,

we plan to migrate our RCP application to RAP but 2/3 major risks still
afraid us to do.

Our application is a medical ERP aims to be used by thousand of users and
must have an high availability (its use by emergencies units).

First concern : memory usage on server. As an instance is used the whole
the day (and night), how many memory will be used on server ? For each
instance of the application, what is share on server/client side ? For
example, is the table checked elements are saved on server or client side
?

Second concern : the huge number of events request that could possibly
flood the server.
Did someone have already implement an application used by thousand of
people and could make feddback on server load ?
Technicaly, for a component, is each kind of event sent to server even if
theres no listenner on it ?

Last point, is there a way to customize the white error page when session
lost (and possibly recover state) ?

Thanks for help !

Re: Performances for huge ERP [message #108912 is a reply to message #108827]

Mon, 13 October 2008 03:17

Eclipse User

Originally posted by: rherrmann.innoopract.com

Hi,

please see my comments below.

HTH
Rüdiger

Thomas Moreeuw wrote:
> Hi,
>
> we plan to migrate our RCP application to RAP but 2/3 major risks still
> afraid us to do.
>
> Our application is a medical ERP aims to be used by thousand of users
> and must have an high availability (it�s use by emergencies units).
>
> First concern : memory usage on server. As an instance is used the whole
> the day (and night), how many memory will be used on server ? For each
> instance of the application, what is share on server/client side ? For
> example, is the table checked elements are saved on server or client side ?
For a discussion on memory consumption, please see here:
http://dev.eclipse.org/newslists/news.eclipse.technology.rap /msg03745.html

> Second concern : the huge number of events request that could possibly
> flood the server. Did someone have already implement an application used
> by thousand of people and could make feddback on server load ?
> Technicaly, for a component, is each kind of event sent to server even
> if there�s no listenner on it ?
No, events are only sent if they are of interest to the server side.

>
> Last point, is there a way to customize the white error page when
> session lost (and possibly recover state) ?
The session-timeout page can be localized, but apart from that it
cannot be customized.
There are ideas to implements a client-side session-recovery, but it
is not a committed plan item. Please see these threads:
http://www.eclipse.org/newsportal/article.php?id=4457&gr oup=eclipse.technology.rap#4457

>
> Thanks for help !
>
>

Re: Performances for huge ERP [message #108936 is a reply to message #108827]

Mon, 13 October 2008 03:48

Eclipse User

>> Did someone have already implement an application used by thousand of
>> people and could make feddback on server load ?

We are planning such an application as you can see by our
prototypes/measures mentioned in the other post by ruediger.
Our nonfunctional req. demand not to exceed 1.5GB for 2000 concurrent
sessions on one server. The application will run the whole day as well.

So far we believe to achieve that - provided that some patches from
https://bugs.eclipse.org/bugs/show_bug.cgi?id=240143 are integrated.
Nevertheless - you have to prepare your application for that - which needs
some more time of thinking and architecture - just hacking together your
RAP-application wouldn't be enough.

For instance:
- we don't use the full RAP-stack. We 'just' use RAP in 'standalone'-mode
with swt + jface.jars. No workbench, as we don't know how memory-hungry
this one is.

- we distribute information among many horizontal+vertical-tabs - so never
have too much information visible at once (though the user thinks there
are a lot of open tabs).
This allows us to dispose the contents of the invisible tabs. If the user
clicks on such a tab the content is recreated on the fly.
Common memory<->time tradeoff.

- we use single-sourcing techniques, which means we can start our app as
an SWT-app as well.
This is good for automated testing (using 'abbot for swt) up to a certain
degree but far better: in the worst case (which we don't assume) we can
deploy the application as a desktop-version via webstart. So all the work
we are doing so far won't be lost.

Re: Performances for huge ERP [message #108950 is a reply to message #108912]

Mon, 13 October 2008 08:15

Eclipse User

"No, events are only sent if they are of interest to the server side."

Can you be more specific about this ? What kind of events is of interest
for the server ?

For example when I select an item in a tree (not expanding, just
selection) a request is sent to the server (I use firebug xhr request
listener) but nothing is done except colouring the item. Why is this of
interest to the server ?

If i put a keylistener on a field, does it mean a request will be sent
each time I strike a key in this field ?

Re: Performances for huge ERP [message #109091 is a reply to message #108950]

Tue, 14 October 2008 05:23

Eclipse User

Originally posted by: rherrmann.innoopract.com

John Fa wrote:
> "No, events are only sent if they are of interest to the server side."
>
> Can you be more specific about this ? What kind of events is of interest
> for the server ?
Events are considered of interest for the server-side if there is a
listener attached.

>
> For example when I select an item in a tree (not expanding, just
> selection) a request is sent to the server (I use firebug xhr request
> listener) but nothing is done except colouring the item. Why is this of
> interest to the server ?
Because someone added a SelectionListener to the tree. You will
probably find a call to tree.addSelectionListener() in the code.
Therefore RAP instructs the client to notify the server whenever the
selection changes.

>
> If i put a keylistener on a field, does it mean a request will be sent
> each time I strike a key in this field ?
First, key events aren't yet fully implemented. But yes, we are
currently working on key events that work like you described. The
focus of this effort is to bring cell editors to work and it
currently seems that this approach is sufficient.

>

Re: Performances for huge ERP [message #111005 is a reply to message #108936]

Fri, 31 October 2008 11:32

Eclipse User

Hi,

tanks for your posts, it have been helpfull to us.

We also realized small tests based on org.eclipse.rap.demo (customized
with our own components) with JMeter/JConsole. Tests have been performed
on local machine (dual-2GHz, 2Go RAM, XP SP3) so results must be
interpreted with care.
We've use Innoopract test plan for demo application (see
http://rapblog.innoopract.com/2008/10/performance-testing-ra p-on-cloud.html)
and launch test with 20, 50 and 100 users (can't go further without
crashes JMeter).

To summarize, the application use around 40 to 50Mo memory with 20, 50 or
100 users. We still need more realistic test, I'll send you summary report
as soon as the tests are performed.

I still don't understand all RAP principles, is it a good documentation
entry point to learn more about architecture and RAP fondamentals ?

best regards,
Thomas

Re: Performances for huge ERP [message #111041 is a reply to message #111005]

Fri, 31 October 2008 21:43

Eclipse User

Originally posted by: evolanakis.innoopract.com

Hi Thomas,

just wanted to say that you can add more users to jmeter if you increase
the HEAP parameter in jmeter.sh / jmeter.bat. You can also experiment
with starting jmeter "headless" (without UI). See
http://wiki.eclipse.org/RAP/LoadTesting (at the bottom).

I've been using jmeter with 800-1000 users / threads.

Kind regards,
Elias.

Thomas Moreeuw wrote:
> Hi,
>
> tanks for your posts, it have been helpfull to us.
>
> We also realized small tests based on org.eclipse.rap.demo (customized
> with our own components) with JMeter/JConsole. Tests have been performed
> on local machine (dual-2GHz, 2Go RAM, XP SP3) so results must be
> interpreted with care.
> We've use Innoopract test plan for demo application (see
> http://rapblog.innoopract.com/2008/10/performance-testing-ra p-on-cloud.html)
> and launch test with 20, 50 and 100 users (can't go further without
> crashes JMeter).
>
> To summarize, the application use around 40 to 50Mo memory with 20, 50
> or 100 users. We still need more realistic test, I'll send you summary
> report as soon as the tests are performed.
>
> I still don't understand all RAP principles, is it a good documentation
> entry point to learn more about architecture and RAP fondamentals ?
>
> best regards,
> Thomas
>

--
---
Elias Volanakis
Innoopract, Inc.
http://www.innoopract.com

Re: Performances for huge ERP [message #111331 is a reply to message #111041]

Wed, 05 November 2008 20:37

Eclipse User

Thanks, I have a look a this for next test.

Thomas

Elias Volanakis wrote:

> Hi Thomas,

> just wanted to say that you can add more users to jmeter if you increase
> the HEAP parameter in jmeter.sh / jmeter.bat. You can also experiment
> with starting jmeter "headless" (without UI). See
> http://wiki.eclipse.org/RAP/LoadTesting (at the bottom).

> I've been using jmeter with 800-1000 users / threads.

> Kind regards,
> Elias.

Previous Topic:	[ANN] RAP 1.2 M2 available
Next Topic:	Weird Characters at startup

Goto Forum:

-=] Back to Top [=-

Current Time: Mon Jul 07 21:07:29 EDT 2025

.:: Contact :: Home ::.

Breadcrumbs

Sign up to our Newsletter