BlazeDS and scalability (i.e., concurrent users)

BlazeDS relies on the Servlet API...so, it means about a few hundred concurrent users, maximum -- which isn't very encouraging. This is because, in BlazeDS, one user equals one, or maybe two/three threads...
Has anyone else considered this and/or thought of a work around? I presume BlazeDS cannot load balance, etc, so having a proxy like Apache point to different servers isn't an option.
Anyone have any suggestions?
Thanks,
Cormac

>>BlazeDS relies on the Servlet API...so, it means about a few
>>hundred concurrent users, maximum -- which isn't very encouraging.
>>This is because, in BlazeDS, one user equals one, or maybe
>>two/three threads...
How many concurrent users a BlazeDS server can support will depend a lot on what channels/endpoints you are using. You are right that with a streaming endpoint you are probably limited to a few hundred concurrent users because each connection is persistent and is going to tie up one of your app server's request handler threads. If you use polling however you can support many more concurrent users. This is because connections are not persistent and the poll requests made to the server are short-lived. Web/application servers are very good at handling these types of requests so you should be able to handle a few thousand concurrent users using polling (although a number of factors such as your polling interval, type and size of data/messages being sent around, individual app server, server hardware, network configuration etc. will all affect this). The trade off with using polling is that it is not as real time as streaming. If a message shows up on the server it won't get delivered to you right away. Instead you'll get it the next time your poll request reaches the server. This might be ok for some people. If you need "real time" delivery of messages to large numbers of clients, you can use LiveCycle Data Services ES which has an NIO based version of the streaming endpoint. This endpoint can support many more concurrent users (probably a few thousand although caveats listed above also apply here) than the HTTP based streaming endpoint because each connection doesn't tie up a request handler thread.
>>Has anyone else considered this and/or thought of a work around? I
>>presume BlazeDS cannot load balance, etc, so having a proxy like
>>Apache point to different servers isn't an option.
You can load balance BlazeDS servers. Make sure your load balancer uses session pinning/sticky sessions because for messaging to work you need to keep communicating back to the same server. We've seen some issues with load balancing streaming connections because some load balancers buffer the data being sent back to the client over the streaming connection. I believe Apache is one of the servers that does this so you'd definitely have to do some testing to find a proxy server that worked for you if you wanted to load balance streaming connections.
>>Anyone have any suggestions?
BlazeDS actually scales quite well (at least in my opinion) when using non-persistent connections with short-lived requests, such as with polling. If you want to support large numbers of concurrent users with streaming/persistent connections, I'd look at using LiveCycle Data Services ES.
-Alex

Similar Messages

A quick question about WebDynpro SLD and R/3 with concurrent users

Hello ,
I have a very quick question about Webdynpros and SLD connecting to an R/3 system, when you configure a webdynpro to connect to an R/3 system using SLD, you configure a user name and password from the R/3 for the SLD to use. What I would like to know is when I have concurrent users of my webdynpro, how can I know what one user did in R/3 and what another user did? Is there a way for the users of the web dynpro to use their R/3 credentials so SLD can access the R/3? Like dynamically configuring the SLD for each user?
- I would like to avoid leaving their their passwords open in the code ( configuring two variable to get the users username and password and use these variables as JCO username and password )
Thanks Ubergeeks,
Guy

Hi Guy
You will have to use Single Sign On to achieve this. In the destination you have defined to connect to R/3 , there is an option to 'useSSO' instead of userid and password. This will ensure that calls to R/3 will be with the userid that has logged into WAS. You wont need to pass any passwords because a login ticket is generated from WAS and passed on to R/3. The userid is derived from this ticket.
For this to happen you will have to maintain a trust relation ship between R/3 and your WAS ,there is detailed documentation of this in help files. Configuration is very straight forward and is easy to perform
Regards
Pran

Scalability in BlazeDS - What is the maximum number of concurrent users possible with BlazeDS?

Hi all,
is there a limit on the number of concurrent users that can connect via BLAZEDS? I see many places that it is only in 100s and not in 1000s.
is that true? Anybody has doen performance testing on that?
Thanks
Kiran

What is practical? Probably between 1 and 6 streams can be done with reasonable use of bandwidth. However, keep in mind that you will likely want to use the lowest quality setting for cameras so they are demanding the least amount of bandwidth.
Using the Connect 8 numbers, Low quality webcam video with the room at LAN settings should have an aprox average bandwidth consumption of 184 kbps. From there it is just simple math. 6 live streams at that setting will be 1.1 Mbps. Alternatively, at High quality webcam video and with the room at LAN, each stream will average aprox 501 kbps. 6 streams will average 3 Mbps.

IFS and 3000 concurrent users ?

We have tested iFS in a small scale and it works fine, but it requires a lot of resources. We now have a case with a client who wants to run iFS for 3000 concurrent users. Are there any whitepapers etc where scalability of this scale is discussed ? I have read about the Siemens-reference, do they have any numbers ?

Comments below (after your questions)
Q1) To increase the maximum heap size in iFS 1.0.8.3 on Solaris 7, we modify the -mx heap size in the ifsstart script (command below). Is this the correct location to modify this setting?
A1) No. In the iFS 1.0.8 Installation Guide, on page 1-5, it says "JVM maximum heap sizes can be configured by editing the following file:
$ORACLE_HOME/ifs/settings/IfsProtocols.def."
There you can specify the MX setting independently for each protocol.
Q2) There is an additional command in ifsstart to restart JWS. By default, this -mx heap size is smaller than the other. What is the reason behind this?
A2) It's not exactly clear, but on page 1-4 of the iFS 1.0.8 Installation Guide, it says:
Note: The Java Web Server (JWS) is actually made up of up to three processes, each of which is set to have a maximum heap size of either 64M (minimum requirements) or 128M (recommended settings). In almost all cases, only one of these processes will ever come close to consuming this maximum amount. The memory usage of the other two processes is negligible.
I believe that the "mx" is lower because it doesn't need much memory. (But I don't have a 1.0.8 install in front of me).
Q3) If you have the memory avaiable, should each of these heap sizes be set to the maximum the JVM can handle (2GB)? Drawbacks?
A3) If you have enough memory, and you are starting the iFS protocols using ifsstart, then each of the MX settings in the IfsProtocols.def file should be set as high as 2G (for each protocol you're using).
Q4) $IFS_JWS_HOME/bin/jserv -jre -vmargs -mx512m > $ORACLE_HOME/ifs/log/JWS.log 2>&1 &
A4) I think this is right.
null

Hign vol txns and concurrent users

          I have a application in hand which will be reqd to support 100000 txns per day and about 1000 concurrent user
          . Can BEA WebLogic deliver that. Do you have a live project example of this volume done in Weblogic. What will be the scalability
          It would be nice if U can help me on this


In article <[email protected]>, [email protected] says...
          >
          > I have a application in hand which will be reqd to support 100000 txns per day and about 1000 concurrent user
          > Can BEA WebLogic deliver that. Do you have a live project example of this volume done in Weblogic. What will be the scalability
          > It would be nice if U can help me on this
          >
          DT,
          Do call a BEA sales rep. They have the documentation and access to
          engineers to discuss this with you in person. Rob, Gene et al are not
          equipped (or authorized, I think) to disseminate this sort of
          information in this forum.
          BTW 1M tx a day is a lot as is 1000 concurrent users. I'd be more
          focused on networking and database hardware then app server.
          Jesse

Load balancing and clustering for SQL SERVER ANALYSIS SERVER instances to access large number of concurrent users for bi site

HI
how to do analysis server clustering
we have sharepoint bi site and we published many dashboards pointers ,filters kpis
and at a time some 200 to 300 concurrent users access these dashboards pages
so for maintain this requirement how we apply clustering for ssas intances
and also it has load balancing
adil

Hi Adil,
According to your description, you want to implement SQL Server Analysis Services Clusster and Analysis Services Load Balancing, right?
In SSAS, we can design highly available solutions in Analysis Services by using: network load balancing (NLB) technologies, failover clustering technologies, or both. For the details about implemention, please refer to the links below.
http://msdn.microsoft.com/en-us/library/dn736073.aspx
http://sqlsrvanalysissrvcs.codeplex.com/wikipage?title=Analysis%20Services%20Load%20Balancing
http://technet.microsoft.com/en-us/library/bb500217(v=sql.105).aspx
Regards,
Charlie Liao
TechNet Community Support

How EM invokes the concurrent users and list of parameters scripts in EM ?

Hi,
How to check if EM is invoking the concurrent users and list of parameters scripts for a db instance target or not ?
And how EM invokes these in EM ?
Thanks in Advance.

Not sure what you mean with this question, could you be a bit more specific?
Regards
Rob
http://oemgc.wordpress.com

EP Performance Tunning and concurrent user login problems

Hi all.
We have a EP , integrated with Windows AD and SAP R3.
In real situation, there will be 1000 users login in EP in 0.5 hours. And their operation time will be 6-8 hours, connect to R3 or other applications.
Is EP can support such concurrent user login scenario ?
if EP fails, is there some performance monitor tool or performance tunning issue ?

Hi
EP supports this logon load, however it all depends on sizing and architecture! Also the content of the STARTPAGE can be critical (make it "light").
If the portal fails, then often it is only one J2EE server process, which fails - so by having more server-processes will be an easy/minimum form of high availability. A "real" HA solution must be considered if your business will not work without the portal og SAPGUI connections can be used as workaround/fallback in such situations.
BR
Tom Bo

Restrict concurrent user and/or session from a client

Hi All:
We have an application with database Oracle 10g.
I want to add a new validation to restrict concurrent user and/or session from a client. (we have almost 60 client firms using the software to enter daily trasnactions).
All users from all clients are connecting to the database using a common functional ID.
What I did was:
1) Add a column 'user_logged_in' in the master table for client and update it as Y when user from that client logged on to the system,
2) Insert the application logon details (we can figure out the client details from this) into a global temp table,
3) Create a logoff trigger to update the 'user_logged_in'flag in client master table by using values from global temp table when session logged off and
4) Restrict the users from same client if the flag is 'Y'
But the problem in this case is logoff trigger will not be executed in case if the session got killed or terminated abnormally.
Is there any other good solution for this scenario?
Thanks
Robin

>
But the problem in this case is logoff trigger will not be executed in case if the session got killed or terminated abnormally.
Is there any other good solution for this scenario?
>
A better question might be - is there any other 'worse' solution than that?
>
2) Insert the application logon details (we can figure out the client details from this) into a global temp table,
>
Meaning - if that session goes down you have NO logon details since the data isn't persistent. Is that really useful?
There is nothing you can do to account for someone just pulling the plug on: the server, the network connection, the bridge, the router, etc. And if a user is connected, starts a long-running process and then gets disconnected the process is likely to keep running in the background until Oracle needs to talk to the user again and then realizes that they are gone.
Meanwhile the user is trying to logon again but can't because the system thinks they are already logged on.
>
All users from all clients are connecting to the database using a common functional ID.
>
Doesn't that kind of make it harder, not easier, to track who is really connecting and using your application?

Concurrent user and hardware

can any one tell me pleas ..
how many concurrent user can access on oracle at the same time??
what is the minimum computer hardware specifications to install oracle??

user12831849 wrote:
can any one tell me pleas ..
how many concurrent user can access on oracle at the same time??
what is the minimum computer hardware specifications to install oracle??Hi and welcome to the forum
Look at the documentation :
http://download.oracle.com/docs/cd/B19306_01/install.102/b15660/pre_install.htm#i1011296

Named and Concurrent users

Dear Friends,
Please help me to find named number of users in PeopleSoft specific to module.Below are the modules i am using and i wanted to know named and concurrent users using these modules
Human Resources
Benefits Administration
Pension Administration
Thanks
Uday.

Hi Bart,
Difference between named users and concurrent users:
Named user means when you buy BOXI you buy a certain amount of named user licenses and then you give each of these named users a login. It`s a cheaper type of license but you have to know exactly who your users are, and so is pretty useless if your userbase is constantly changing or your exposing Infoview over the web.
Concurrent user licenses are now only available in BOXI Edge - the equiv in BOXI Pro\Premium is a CPU license, which means you can let as many people login as your hardware will support. Concurrent user licenses are useful when you don`t know who your userbase is - you can just give a generic login for people to use. Or give people specific logins and not worry that they might not be using the system and using up a valuable named user license.
for more information follow the below links:
http://www.forumtopics.com/busobj/viewtopic.php?p=476977
http://www.forumtopics.com/busobj/printview.php?t=125096&start=0&sid=095882f7c0d636e7bca723e5930c21d3
Hope this helps.
Regards,
Deepti Bajpai

High txn vol and concurrent users

I have a application in hand which will be reqd to support 100000 txns per day and about 500 concurrent user
. Can BEA WebLogic deliver that. Do I need any tools like Tuxedo etc. Do you have a live project example of this volume done in Weblogic.
It would be nice if U can help me on this

100000 txs per day is about 1-2 tx per second. (unless my math is
off). That should be no problem for WLS.
I would suggest that you talk to your sales representative or
[email protected] for customer references.
-- Rob
DT wrote:
>
I have a application in hand which will be reqd to support 100000 txns per day and about 500 concurrent user
Can BEA WebLogic deliver that. Do I need any tools like Tuxedo etc. Do you have a live project example of this volume done in Weblogic.
It would be nice if U can help me on this

High volume txns and concurrent users

I have a application in hand which will be reqd to support 100000 txns per day and about 500 concurrent user
. Can BEA WebLogic deliver that. Do you have a live project example of this volume done in Weblogic.
It would be nice if U can help me on this

"DT" <[email protected]> wrote:
>
I have a application in hand which will be reqd to support 100000 txns per day and about 500 concurrent userwhen you say concurrent user, do you mean to say
1. concurrent HTTP sessions or
2. max connections(TCP/IP) to the server at any point of time..
can you provide with more info regarding the requirements...
1. How big is the application?
2. What type of application?
3. MAX HTTP Sessions at any point of time
4. MAX Connections at any point of time
5. HOW fast is the application...(jsp/ejb/db or jsp/ejb to external interfaces like messaging systems...etc...)
. Can BEA WebLogic deliver that.TTBOMK and assuming jsp/ejb/db...yes,..the point is to determine the h/w required to support the requirement.
Do you have a live project example of this volume done in Weblogic.
It would be nice if U can help me on this

What are the AM Tuning parameters need to be set for 500 concurrent users

Hi All,
We are using JDEVADF_11.1.1.5.0.
What are the exact AM parameters(Pooling and Scalability tab) to be set when we planned for 500+ concurrent users?
Along with this do we need to set any parameters at weblogic server level, if any please give inputs.
Any input's would be appreciated.
Thanks in advance.
Regards,
Dinesh

We can?t give you exact numbers as there are many factors influencing the performance. In hte end you have to make a load test to find the right set of parameters. A good starting point is Andrejus's blog Andrejus Baranovskis's Blog: Stress Testing Oracle ADF BC Applications - Internal Connections and Andrejus Baranovskis's Blog: Stress Testing Oracle ADF BC Applications - Do Connection Pooling and TXN Disconnect Level
Timo

Need the most performance and scalable way to "trigger" JavaEE application

Hi,
Need the most performance and scalable way to "trigger" JavaEE application code running on WLS from PLSQL code.
There are some thirdparty JavaEE application running on WLS that we need to invoke from PLSQL code. I have looked at couple of options:
Option 1)
Use oracle db callouts(http call etc.) but it seems the connection to WLS will not be maintained across database-sessions and so every call from different plsql database-session will make fresh TCP/IP connection. Usually fresh socket connection setup is expensive and can become a bottleneck.
Option 2)
Use AQ in the database from PLSQL and have a MDB in WLS to invoke the java code. The problem here is how do I simulate the Request/Reply pattern using the AQ API from PLSQL. The PLSQL code needs to wait until the JavaEE code completes as it needs some result back from JavaEE app. AQ also does not have true temporary queues so I am not sure how it will work with concurrent user/sessions.
Option 3)
Use Java in the database, we have tried this but for our usecase it is too expensive resource wise and is not meeting our performance needs.
Performance is the triggering mechanism is of highest interest to us.
Any suggestions/pointers is greatly appreciated.
Thanks,
Prantor
Edited by: Prantor on Nov 8, 2010 10:37 PM

Perhaps you would benefit from viewing the problem differently.
In some/many/most cases the application server (WLS) invokes procedure residing in the DB server.
Place the control for the desired results in the application code running on the WLS.

BlazeDS and scalability (i.e., concurrent users)

Similar Messages

Maybe you are looking for