| Scalability of RAP [message #57061] |
Sat, 03 November 2007 22:23  |
Eclipse User |
|
|
|
Originally posted by: daniel.hope4all.de
I'm preparing a lecture about RAP and therefor analyzing the scalability
of RAP.
In the current Eclipse Magazine there is an article "RCP goes Web 2.0"
from Mr. Muskalla and Mr. Sternberg. It is said that a web workbench test
with 500 concurrent users causes a heap space usage of approx. 100 MB.
According to wikipedia a software with good scalability will get by with
ten times of resources when the amount of demands increases tenfold. And a
software with bad scalability will cause the same resources usage with
only double amount of demands.
(http://de.wikipedia.org/wiki/Skalierbarkeit)
What is the heap space usage of only one RAP web workbench test?
And where can I range RAP between the definition of good and bad
scalability?
Thank you!
Regards,
Daniel Schnabel
|
|
|
| Re: Scalability of RAP [message #57666 is a reply to message #57061] |
Mon, 05 November 2007 13:03  |
| Eclipse User |
|
|
|
Originally posted by: fappel.innoopract.com
Hi,
as RAP is a server-side technology you can use clustering to archieve
horizontal scalability. Using the values of the example workbench test you
mentioned, 5000 concurrent users can be handled with 10 servers of the same
capability using each 100 MB of heap space. From the performance point of
view a lot depends on the capacity of the load-balancing mechanism, but this
is not a RAP specific problem.
In practice the bottleneck with horizontal clustering is most often the
backend. It doesn't help much if the frontend can handle 5000 concurrent
users, but the data provisioning for those requests doesn't work fast
enough. But this is also not a RAP specific problem.
Improving the hardware infrastructure to archive vertical scalability is
also possible, but it's not that easy to predict results. Doubling the
available heap, doesn't inevitable help to double the possible concurrent
users. For example, if the amount of possible concurrent connections that
are supported by the underlying system are exceeded, increasing memory will
not have any effect. On the other hand it could also be possible that the
concurrent users could more than double, since the additional heap is
completely available for additional sessions. If I remember correctly in the
test example the server and framework infrastructure occupied somewhat
between 10 to 15 MB heap, which would left 80 -85 MB for sessions.
Ciao
Frank
"Daniel Schnabel" <daniel@hope4all.de> schrieb im Newsbeitrag
news:4fb42377c285664df8c123f6f56c1192$1@www.eclipse.org...
> I'm preparing a lecture about RAP and therefor analyzing the scalability
> of RAP.
>
> In the current Eclipse Magazine there is an article "RCP goes Web 2.0"
> from Mr. Muskalla and Mr. Sternberg. It is said that a web workbench test
> with 500 concurrent users causes a heap space usage of approx. 100 MB.
>
> According to wikipedia a software with good scalability will get by with
> ten times of resources when the amount of demands increases tenfold. And a
> software with bad scalability will cause the same resources usage with
> only double amount of demands.
> (http://de.wikipedia.org/wiki/Skalierbarkeit)
>
> What is the heap space usage of only one RAP web workbench test?
> And where can I range RAP between the definition of good and bad
> scalability?
>
> Thank you!
>
> Regards,
> Daniel Schnabel
>
|
|
|
Powered by
FUDForum. Page generated in 0.03186 seconds
] 
Search
Help
Register
Login
Home
