More than one ATP Server

[Former Member]

Hello guys and gals,

Is it possible to define more than one ATP server for one system?

Like setting rdisp/atp_server = host1,host2

???

Thanks,

Martin.

[Former Member]

Hi,

When reading the profile parameter documentation, it seems possible.

I don't really see why this would be useful. Maybe if the ATP load is huge ?

Regards,

Olivier

[Former Member]

Not for the load, more for the redudancy.

[Former Member]

Hi,

>Not for the load, more for the redudancy.

I don't see the point : I always use the CI server as the ATP server.

If the CI is down, the system is down. So installing only one ATP server on the CI does not add a SPOF.

(single point of failure).

My system with the ATP server has the central instance on an MSCS cluster. It is so HA.

Regards,

Olivier

[Former Member]

Allright, here's the story.

We're running ECC6.0 on AIX, ABAP Stack only.

We are clustered using AIX's HACMP.

We have four LPARs (Logical Partitions, or machines)

Two are clustered and two are standalone supplemental application servers.

Among the two that are clustered, one node is running Oracle (not RAC) and replicated enqueue (ERS), and the other node is running Central Services Instance for ABAP (ASCS) and a primary applicative instance (formerly CI).

The two nodes are mutually redundant.

The CI is actually our ATP server.

If one of the nodes fails, obviously the a take-over occurs on the other node. However, not to overload the node that's taking over the other one, the ASCS instance is restarted but NOT the CI.

The ASCS being the enqueue and message server, the system can run perfectly without the "CI" running as long as at least one supplemental app. server is running.

BUT! This will NOT ensure that the ATP is redundant.

I'm thinking of adding a third instance on the "CI" HACMP node exclusively used for ATP. So on that CI node, there will be ASCS, CI and ATP. Three different instances.

HACMP would be configured to provide redundancy for ASCS and ATP and still not for CI. This would ensure that ATP is always there and respect the rule of not overloading the HACMP node that'll be taking over.

This raises many other questions, like: what work processes do I need for ATP? Can it be a dialog-only instance? Of course it will not placed in any logon-groups...

Martin.

[Former Member]

Hi again,

Thanks for the detailed explanation which gives a good reason for your question.

Your problem is that in your setup the CI is just the first DI.

I think that for an ATP only instance, dialog processes and a big enough import/export buffer are enough.

You'll have to experiment though.

Why not just setup ATP server on, 2 at least, dialog instances. This should not become a SPOF.

Thanks for pointing me to the HA problem of the ATP server : I will have this problem when I will update my R/3 4.7 MSCS cluster to ECC 6.0...

Regards,

Olivier

[Former Member]

Why not just setup ATP server on, 2 at least, dialog instances. This should not become a SPOF.

Hahaha, this is my question in the first post. 😜