Just as Exadata can be used for running multiple Databases in a single rack, I am looking for notes / experiences running multiple Applications Instances on Exalogic. Typically Application Consolidation run as
a. Multiple EBusiness Suite Installs (e.g. one install for APAC, another for EMEA and a third for Americas) --- not Global Single Instance, but 3 separate Apps Stacks
b. Multiple Peoplesoft Stacks running (e.g. one for each region)
c. Multiple Applications using same stack (e.g. one for Peoplesoft HR and another for Peoplesoft Financials)
Yes that is very much feasible. In some sense Exalogic is a set of standard Linux-Intel nodes interconnected by inifiniband. So we dont have seperate installation guides for each of the apps you mention on Exalogic. Regular installation guides work as-is and regular application partitioning strategies apply. For example for EBS you can go for a single node install (everything on one node) to a multi-node install with Web Tier, Cocurrent Mgr and Database on different nodes.
Re mulitple instances, the one thing to note is that Exlaogic comes with a shared storage, so seperating the software bits for each installation would be required. This is typical for any NAS based storage.
Also we recommend that the database be on Exadata. I am assuming since these instances are a seperate stack, so is the database. You will need a database strategy - we suggest Exadata (of course).
I can share our experience while testing JD Edwards EnterpriseOne on Exalogic. As you might imagine, the eagerness among Oracle software engineering teams to test on Exalogic is quite high, but the number of actual Exalogic systems available for testing is finite. When we requested access to an Exalogic system for our JD Edwards EnterpriseOne testing we were allotted 3 nodes from an 8-node (quarter rack) system.
On those 3 Exalogic nodes we were able to observe:
- JD Edwards EnterpriseOne ran without errors or modification. As promised, Exalogic looked just like Oracle Linux on x86-64.
- We easily scaled to 900 simulated interactive users with lots of CPU and memory to spare. We're not calling it a "benchmark" because we're sure we could go higher.
- We observed the fastest average response time (0.16 seconds) of any platform we've ever tested.
That was on 3 nodes. What was happening simultaneously on the other 5 nodes in the quarter rack system? We have no idea. That's the point. The nodes are isolated, so our brethren of other Oracle applications who were pounding away with their own tests on their own nodes didn't affect our testing at all.
I realize that is anecdotal rather than empirical evidence, but our experience was good.