Skip to Content
0
Former Member
Mar 22, 2006 at 08:41 AM

How to handle big portal with many critical applications

40 Views

Hi,

We've got a big portal with quite a lot of usage which is going from an information portal to a work-desktop for many users. This means that a lot of mission critical applications will be integrated into the portal, mainly through SAP business packages and custom solutions (bsps,webdynpros etc.)

Even though a high availability environment is in place, there is some concerns that a single big portal will be too vunerable. Backup solutions are being designed, but for applications using business packages and webdynpro for java, it will be difficult to get a good alternative.

Since EP is based on java, it is very vunerable to application bugs, since everything is run in the same process. Even though we have 24 application server nodes, a single infinite loop in one of the applications, could effectively take down the entire environment. Some other J2EE servers, like Websphere, solve this by having each J2EE application run their own java processes. This is probably not possible in EP since portal applications are applications of a J2EE application named irj.

I see some possible alternatives:

1. One big portal which is tuned for max uptime containg both applications and information content

2. Two portals, one for information content and one for applications. Reduces load on the application one since it is only called when needed, but increases maintenance and applications can still affect each other

3. One information portal, and one small portal for each application. Maintenance nightmare, plus HA is difficult to guaranteed for each application due to size of cluster

4. One portal for all content. Applications are logically mapped against two application servers in this cluster, by use of a separate dns name/alias and external loadbalancing rules. Let's say we have 6 servers and 4 applications; A-D.

Server1: Information portal + Running App A,D

Server2: Information portal + Running App A,D

Server3: Information portal + Running App B

Server4: Information portal + Running App B

Server5: Information portal + Running App C

Server6: Information portal + Running App C

We have one major alias portal.bouvet.no, and several other portalapp1.bouvet.no, portalapp2.bouvet.no and portalapp3.bouvet.no(or maybe one for each application if we really want to waste IP-addresses) . If say server 2 goes down, we can change the load balancing for portalapp2.bouvet.no to include server3 instead of. Since applications A and D are allready installed on the entire cluster, the affect is minimal.

As you might see, I believe alternative 4 is very interesting.

Our question is if anyone else is in this situation and have some thoughts/feedback on it (anything at all).

Dagfinn