HA in Windows + SQL Server - in NW 7.0

[Former Member]

Hi all,

I have to install an ABAP-only SAP ERP 2005 system using High Availability in Windows 2003 + SQL Server 2005 for my customer.

There's no way else except using MSCS so SAP can support me in case of problem?

After reading SAP installation guide (ERP 2005 SR2 ABAP on Windows SQL Server), I find 2 one-cluster options:

- Option 1:

+ Machine 1: 1st MSCS node: ASCS, DB instance, and CI.

+ Machine 2: Additional MSCS node: ASCS, DB instance, and DI.

- Option 2:

+ Machine 1: 1st MSCS node: ASCS, DB instance.

+ Machine 2: Additional MSCS node: ASCS, DB instance

+ Machine 3: 1 CI node.

+ Machine 4: 1 or more Dialog instance.

Unfortunately, the installation guide says that CI is a non-SPOF and my customer wants a MSCS with full clustered so the CI have to be clustered also. Customer also wants to use 3 machines only, they don't want to use 4 machines to save cost.

At first thougth, I think the infrastructure is:

- Machine 1: 1st MSCS node: ASCS, DB instance

- Machine 2: 1 CI node

- Machine 3: additional MSCS node: ASCS, DB instance, and CI.

Machine 1 & machine 2 make the primary application server instance, machine 3 is for case of failover.

I don't know my infrastructure is right or not with 2 CIs. What I want is to make a full clustered solution with 3 machines.

Someone with experience please help me.

Thanks a lots,

Toan Do

[Former Member]

Hi!

What exactly is 'fully clustered' for you? It looks like your customer is aiming for the 'classical' HA-Setup which is putting the CI (with the workprocesses) in a switchover group as opposed to the Netweaver7 HA, which only puts the SPOFs (ASCS, SCS) in switchover groups. When you use an ASCS your 'CI' turns into a primary application server that does not run any services anymore that are not available in any secondary application server. (If you have Java installed there are some things in the FS of the primary application server and there is the SDM, which is only available here, but these are not seen as SPOFs).

So the first question you need to find an answer for (with your customer) is, if they require an instance with workprocesses to be in a switchover group or if they (for whatever reason) do not wish to run more than one instance for the users. Depending on the answers you get we might be able to provide more inputs.

Regards,

Jörg

[Former Member]

you need to have AT LEAST 2 Application servers (e.g. one Central Instance and

one Dialog Instance) installed on 2 DIFFERENT machines (in your case MSCS

Node A and B).

When you use loadbalancing (which in ABAP goes over message server - ASCS

instance) , it would connect you to app. server which is available and has the best

performance.

If you install more App.server then 2 (e.g. additional DI) on additional different

machines - it is even better, in terms of (high) availability and as well scalabiliy.

But 2 are minimum.

SAP offers you possibility to install CI and DI on MSCS nodes, so that you can

reduce TCO. In this case you have to do a proper sizing, as it can happen that on

one MSCS you run an appserver and (clusterd) database.

Bu you have also possiblity to install app servers outside of MSCS. Any

combination is possible. What is important is - at least 2 app- servers on 2 different

machines

Pls do Ack with Points if problem is cleared.

Regards

Manjunath

[Former Member]

Hi Manjunath,

Thanks for your answer, but I understand what you mention; I got it from document.

But here's my question:

*. Given that I use option with 2 machines (minimal servers):

+ Machine 1: 1st MSCS node: ASCS, DB instance, and CI.

+ Machine 2: Additional MSCS node: ASCS, DB instance, and DI.

. What I'm unclear is that:

+ When machine 1 dies, DI in machine 2 is stopped and then CI in machine 2 is started to make the SAP system available? Because I do not test the MSCS yet I wonder where the CI in machine 2 comes from?*

Regards,

Toan Do

[former_member185954]

Hello Toan,

With my customer , we have IBM servers p570 LPARs with HACMP, we have something similar to option 1.

We have the CI and DB on one server (machine1) and that is clustered with (machine2) running with DI.

In the event of a failure, the system automatically stops the DI on machine 2 and machine 1 is failed over to machine2 meaning CI and DB are started back up on machine2.

Regards,

Siddhesh

[Former Member]

Hi Siddhesh,

Thanks for your shared info.

So you have 2 machines:

+ Machine 1: 1st MSCS node: ASCS, DB instance, and CI.

+ Machine 2: Additional MSCS node: ASCS, DB instance, and DI.

You mean that when machine 1 dies, DI in machine 2 is stopped and then CI in machine 2 is started to make the SAP system available?

Because I do not test the MSCS yet, but I wonder where the CI in machine 2 comes from?

Regards,

Toan Do