HTML Clients

NGodusky

Member
As a new JDE Customers, we are planning to implement HTML Clients. Our configuration will be: AS400 as the DB Server with IIS as the Web Server. I would like to know how the response time of the HTML Client compares to the response time of the Terminal Server or FAT Client.
 
Typically, your WEB server or TS server is next to (i.e. on the same switch)as your Enterprise server. So network traffic is less of an issue than a FAT client. So i/o to the db between the two thin clients are basically the same. Therefore, aside from network considerations, the main difference between the WEB and
TS servers is (besides dollars):

1.) Amount of clients support per box: Comparable boxes, WEB servers can typically support more clients.
2.) Amount of Strain put on the Enterprise server: Becuase WEB servers
cannot run any BSFNs, WEB servers put a greater strain on the Enterprise server, most of the time an apps server needs to be introduced.
3.) Functionality: TS clients get a fully functional OneWorld client and "can" run jobs locally. Web clients get a stripped down OneWorld client and cannot run jobs locally.

So to your question:

If configured correctly, I have seen TSE clients function a bit more quickly in interactive jobs than WEB clients, but the difference is nominal. But,note, you cannot expect 100 web users/1 enterprise compared to 100 tseusers/1 enterprise to be equal. This many web users would require an offload of bsfns. In my experience it would seem both thin-clients "function" more quickly than FAT clients,strictly because of the difference in "hops" between client and db and the sizing of the actual client.

Hope this helps,

mo
 
Mo,

regarding your last statement "In my experience it would seem both thin-clients "function" more quickly than FAT clients,strictly because of the difference in "hops" between client and db and the sizing of the actual client."

This is a surprising statement to me. Are you referring to UBEs or interactive applications? Assuming your LAN has an adequate backbone and switching (GigaBit seems the norm nowadays for new installs) and clients have 100MB connections I would not expect latency/bandwidth on a LAN to be an issue, particularly for interactive apps. For UBEs several sites report some UBEs run faster on FAT clients than they do on their Logic Servers.

I am speaking without the benefit of actual experience here, but it seems counter intuitive to me that thin clients have quicker application response times than fat. There's just more layers to go through than the typical Client/Server approach. I can't picture this unless the fat clients do not meet minimum hardware requirements and/or they are running on a 10/100MB LAN.

Anyone else want to chime in on this? I'd really like to hear from mixed fat and thin client environments since we are considering going that route next year.

Regards,

Larry Jones
[email protected]
OneWorld XE, SP 15.1
HPUX 11, Oracle SE 8.1.6
Mfg, Distribution, Financials
 
If you focus on the last part of the statement:

"...and the sizing of the actual client."

it's not hard to imagine that 5 thin TS clients on a 4x800Mhz server would seem more responsive than a Pentium 166 on the other side of the building, a switch or two away.

Of course I'm going to the extreme to make a point.

I asked this same question to many people about UBE's because it seemed contrary to common sense that everyone submitting UBE's on the server at the same time would be faster than just running them on your (fairly fast) workstation. This "sizing" issue is the only remotely legit response I ever received.

I would add that, from a developer's eye, because of the complexity and limitations of the CNC and toolset infrastructure, OneWorld is extremely fat in terms of network traffic, so a TS would get a significant "head start" being very "close" to the enterprise/db server as was pointed out.

---------------------------------
OneWorld Xe SP15
Clustered Windows 2000 + SQL 2000
 
I'm in a lab environment, not in real world production, but for what it is
worth:

There are 18 standard application test scripts we use, (such as sales order
entry, purchase order entry, supply demand inquiry, etc.). Across all the
applications we test, the average response time for the win32 clients,
(which include TSE & Citrix), was generally between .25 & .75 seconds. For
HTML client it was usually between .5 and 1.0 seconds.

This is for latest release & SP for Xe. Results for earlier versions and
some of the early service packs was much worse. We saw a lot of
performance improvements after SP14. This is also for a pretty well tuned
system, with some of the fun little pieces, like silent post enabled. You
can find some of the tuning recommendations at:
http://www.ibm.com/servers/eserver/iseries/service/erp/jdesupport.htm

Also, these were all based on WebSphere as the java application server.
For the iSeries tests we used the HTTP server that comes with it. For
xSeries tests we used IIS as the HTTP server and WebSphere as the web
application server. I think you have to use WebSphere as the java
application server with Xe, not definite on that, but you can use other
HTTP servers.

There are some marketing flyers at the website below that give a little
more info on some of the tests we've run. let me know if you'd like more
details.

http://www.developer.ibm.com/erp/jdedwards/index.html

Best regards,

Rob Jump
Sizing Specialist
IBM/J.D. Edwards International Competency Center
303-334-1054
[email protected]
 
Typically, your WEB server or TS server is next to (i.e. on the same switch)
as your Enterprise server. So network traffic is less of an issue than a
FAT client.

Aside from network considerations, the main difference between the WEB and
TS servers is (besides dollars):

1.) Amount of clients support per box: Comparable boxes, WEB servers can
typically support more clients.
2.) Amount of Strain put on the Enterprise server: Becuase WEB servers
cannot run any BSFNs, WEB servers put a greater strain on the Enterprise
server, most of the time an apps server needs to be introduced.
3.) Functionality: TS clients get a fully functional OneWorld client and
"can" run jobs locally. Web clients get a stripped down OneWorld client and
cannot run jobs locally.

So to your question:

If configured correctly, I have seen TSE clients function a bit more quickly
in interactive jobs than WEB clients, but the difference is nominal. But,
note, you cannot expect 100 web users/1 enterprise compared to 100 tse
users/1 enterprise to be equal. This many web users would require an
offload of bsfns. I would say both function more quickly than FAT clients,
strictly because of the "hops" between client and db and the sizing of the
actual client.

Hope this helps,

mo
 
We have noticed that thin clients can be faster than fat clients. We've
performed no formal tests, but I've experienced it *many* times. Why? Who
knows---who has the time to figure out these things?

In our configuration, all of our users are on thin clients--only the
developer workstations have fat client capability.

Regards,

Wynn Osborne
Flair Corp

B7332, SP11.3
AS/400 V4R4
NT 4.0 Citrix
 
I think you will find, that in the real world 10/100MB Lan is the norm. With
most client PCs sharing a 10MG Connection to the server, in a WAN scenario
this can get worse. Citrix in the other hand is generally connected via a
dedicated 100MB link.

As far as UBEs are concerned, they are a major mover of data from the server
to the client if run locally, UBEs do not create a lot of CPU usage - look
at an NT Task Manager, but they do a lot of Database reads (in the form of
SQL queries). UBEs are therefore best run on the Enterprise Server or a
Logic Server, as this reduces network traffic.



OW733.3 Xe SP 14.2
Enterprise Server - Intel NT + Oracle 8.0.6
Client - Citrix TSE + 4 NT PC's for development
 
well folks,

every time i've encountered thick and thin clients, the thin ones seem to
move way faster, no mater what speed network or specs of the thick client.
 
Re: RE: HTML Clients

Hmmm,

this may truly be an environmental issue. Rob Jump's laboratory test results are what I expected to see in a modern network environment for interactive applications.

When we were first implementing OneWorld (1999, B733 base), I did a number of client response time / performance tests on interactive and batch (UBE) applications. Among the configurations tested were two nearly identical PCs - main difference between the two was the speed of the NIC 10MB on one vs 100MB on another. Test results showed substantially reduced response time and faster UBE performance on the PC with the faster NIC (10% to 60% faster). This confirms what many of you are saying, that OneWorld performance IS very dependant on your network configuration. However, if your LAN is what I consider to be "modern" (100/1000MB or higher), then the location of the application object on the LAN relative to the database becomes less of a factor and other variables assume a weightier role.

So the response to the original poster seems to me to be "it depends on your environment" - typical I.T. answer :)

Larry Jones
[email protected]
OneWorld XE, SP 15.1
HPUX 11, Oracle SE 8.1.6
Mfg, Distribution, Financials
 
Back
Top