Bandwidth required to run JAVA

andy_smith

Well Known Member
Hi List,

Does anyone have any guide to the bandwidth required to run HTML and
JAVA interfaces. Java is intended for LAN as oppose to WAN but if the
WAN link were high enough bandwidth and the user count low could it
support Java ? I.e. for 15 Java users would a 256 Kb line be adequate ?

Apparently there are some guidelines on the average bandwidth required
for HTML, Java an FAT client but I cant seem to find anything.

Is anyone running the Java interface in production - on what type of
link ?

Xe, NT, SQL7


Andy Smith
Technical Consultant

WHITEHOUSE
Consultants
Office: 01564-711-1037
Direct: 01564-711-1539
Mobile: 07949 603770







Andy Smith
Whitehouse Consultants
Win2K SQLServer7 Xe
 
A number of performance tests have been performed with the HTML client versus Java/Win32 and citrix models. There are 2 definitive figures one should be looking at, however, to ensure that you truly get a picture of how OneWorld will perform on a WAN.

The first is bandwidth - very easy - the more traffic flowing across the network - the more bandwidth is required.

However - more importantly is latency and the number of TURNS. This figure is more important than bandwidth usually since it cannot be improved upon easily.

An example is as follows. Let us suggest that someone performs a Sales Order Entry in OneWorld - and that it creates 500Kbytes of network traffic. If we have a 4Mb/s connection - then in theory all traffic can flow in 1 second - hence response time for the application is 1 second. If we have a 500kb/s connection - it'll take 8 seconds right ? Not true. This is true ONLY for the time it takes for the TRAFFIC to flow across the network. But OneWorld is Event Driven - as you click between events, logic is activated, and requests are being made to the server using SQL - lots of chattiness.

Each time a conversation goes across the WAN - the application will hit LATENCY. This can be measured using a PING - and on the internet can be anywhere between 100ms and 1 second. Lets look at the previous example. In our example sales order entry - there could be as many as 500 turns (SQL conversations). If we have 100ms latency - then our reponse time is 500x100ms = 50 seconds of response time. Add the 500kb/s bandwidth and we have almost 1 minute of network response. That is pretty typical of most thick client server applications.

The Thin Client alternatives dramatically change how this traffic looks - they usually attack the chattiness rather than the bandwidth.

I found a presentation from last year that compared the different network traffic bandwidths for a Sales Order Entry test :

Client Type Bytes C>S Bytes S>C #Turns
----------- --------- --------- -------
Win32(FAT) 214825 685961 1065
Win32(BSFN) 108232 235983 275
HTML 23777 109360 10
Java 349992 330529 404
TSE(RDP) 40694 65928 ---
Citrix(ICA) 39332 50774 ---
Java(Citrix) 34546 61014 ---

Some explanation is needed. First of all - it is important to note that the FAT client types are the Win32 clients - one is 2 tier (FAT) and the other is with Business functions mapped. Note that just by mapping business functions - one will get almost 5 times better performance across a WAN.

It is difficult to note down the number of turns for a Citrix/RDP client since the clients use a streaming technology and will always adapt for the type of networks used. One can imagine that citrix is like Video and will skip frames to catch up.

Note that these tests were performed in early 2000 - before the more interactive HTML client - and it is assumed that the new HTML client will cause more turns.

Hope that helps

Jon Steel
Xe Upgrade Specialist
http://www.erpsourcing.com

ERP Sourcing
http://www.erpsourcing.com
[email protected]
 
Back
Top