Netweaver ABAP HA using cluster

[Former Member]

Hi,

I'm running ERP 6.0 EhP4 on DB2 HP-UX.

I have two nodes, planned to use HA switchover.

I have two scenarios:

Scenario 1:

- Install CIDBASCS on node A

- Install dialog instance + ERS on node B

HA scenario 1:

- If node A dies, CI+DB will go to node B

- If node B dies, nothing happen

-

---

Scenario 2:

- Install CI + ASCS on node A

- Install DB + ERS on node B

HA scenario 2:

- If node A dies, CI goes to node B

- If node B dies, DB goes to node A

What I want to ask is the pros and cons between scenario.

Please advice.

Regards,

[sunny_pahuja2 · Accepted solution]

Hi,

>

> I'm running ERP 6.0 EhP4 on DB2 HP-UX.

> I have two nodes, planned to use HA switchover.

>

> I have two scenarios:

> Scenario 1:

> - Install CIDBASCS on node A

> - Install dialog instance + ERS on node B

>

> HA scenario 1:

> - If node A dies, CI+DB will go to node B

> - If node B dies, nothing happen

>

> -

---

>

CI is not the SPOF. So, if node A dies then only DB and ASCS should move to node 2 not CI.

> Scenario 2:

> - Install CI + ASCS on node A

> - Install DB + ERS on node B

>

> HA scenario 2:

> - If node A dies, CI goes to node B

> - If node B dies, DB goes to node A

>

Again CI is not Single point of failure (SPOF)

You should install ASCS and DB on virtual instance/package so when node 1 fails it moves to node 2 or vice-versa. ERS you can also install separately.

But make sure that CI should not be moveable. Because if your node 1 goes down the DI on node 2 will work like CI. Because CI is no more SPOF.

Thanks

Sunny

[Former Member · Accepted solution]

Hi,

If CI is not a SPOF anymore, if CI die, will users currently logged on to that node be kicked out?

Thanks.

Regards,

[Former Member · Accepted solution]

If CI is not a SPOF anymore, if CI die, will users currently logged on to that node be kicked out?

Yes, users will be logged out (and batch jobs on that instance will be terminated).

Cheers Michael

[Former Member · Accepted solution]

Hi,

Then what's the use of Enqueue Replication Server?

Thanks.

Regards,

[Former Member · Accepted solution]

That's the point, as long as your database is not clustered (for example RAC) having a replicated enqeue service does not really increase availability. As i suggested especially in a HA environment it is important to keep the installation as simple as possible. Complexity is always leading to more failures.

Cheers Michael

[Former Member · Accepted solution]

Hi,

Can you explain as simple as possible means?

Do you suggest not to use RAC, or as little application server as possible?

Regards,

[Former Member · Accepted solution]

Unless you have a large system (>1000 concurrent users, large batch jobs etc) my suggestion would be:

Scenario 3:

- Install CIDBASCS Prd on node A

- Install Dev and Qas on node B

One single Service Guard package for the prd environnment.

Cheers Michael

[Former Member · Accepted solution]

Michael,

Oh I see..

That's a good scenario.

By the way, do you know what HP serviceguard extension for SAP offers?

Regards,

[Former Member · Accepted solution]

It supports starting/stopping of database, sap instances, saprouter, ERS whatever you imagine. Theoretically it is possible to just use plain Service Guard to cluster the system, but then you will have to supply the start / stop commands for the package.

If you have never buildt a cluster before, you might need support from a consultant or calculate additional time and build a sandbox environment first. Also a crucial question is the disc and network configuration.

All documentation can be found here: [HP Serviceguard Solutions - SGeSAP|http://h71028.www7.hp.com/enterprise/us/en/os/hpux11i-serviceguard-sgesap.html?jumpid=reg_r1002_usen]

Cheers Michael

[Former Member · Accepted solution]

... but then you will have to supply the start / stop commands for the package....

in form of operating system script?

can you point me what the script looks like?

i've already had the cluster configured.

but no script yet (no extension for SAP).

And do you have information about the service fee?

Thanks.

Regards,

[Former Member · Accepted solution]

In the Service Guard courses you used to learn to cluster an xclock, so instead of starting the xlock you start the SAP system. The problem is, that it is becoming more and more complicated to start a SAP system (startsrv, saphostctrl etc etc).

I said theoretically you can build a package without SGeSAP, but i still suggest to use SGeSAP. It won't help that in case of failure you notice that your script had an error preventing a proper failover. To do all the stuff on your own would require very good hp-ux and SAP basis knowledge.

And do you have information about the service fee?

I cannot give you exact numbers, you should talk to your HP sales guy. In my opinion it is worth the money, HA always has its price.

Cheers Michael

[Former Member · Accepted solution]

Michael,

Example is this plot:

Without HP Extension for SAP.

Node 1 down:

Cluster software automatically move the mount point to node 2.

SAP Administrator manually start the db and sap there.

Case solved.

With HP Extension for SAP.

You pay $10,000 for annual license.

Node 1 down:

Cluster software automatically move the mount point to node 2.

HA software start the db and sap there.

Case solved.

HA software just replace the manual task by system admin.

Is that it? And worth the price?

Thanks.

Regards,

[Former Member · Accepted solution]

Well i don't know how it is in Japan but usually having a SAP basis admin available 7x24 costs way more than 10'000$ per year.

But i see your point we also have a kind of low cost cluster on linux with no Service Guard at all, its all a bunch of self written scripts. You have to decide that yourself, it all depends on how large your system is and if an automatic failover is needed. If the manual failover is ok, then you might very well save that money.

Cheers Michael

[Former Member · Accepted solution]

Hi Michael,

Here in my country resource is cheap.

So I think I know now the strategy for my HA scenario.

Thanks for your help answering my queries.

Regards,

[former_member189546]

Hi,

Oonly ha system directly supported by sap is cluster on Windows.

For ha on unix best to follow advice of hardware supplier

regards,

John Feely

[former_member188883]

Hi Bobby,

There are 2 approaches to have HA setup done for SAP.

1) Active- Active : Where both the server nodes are being utilised. SPOF components like DB services, ASCS services and ERS are distributed across both the server nodes.

For this you can have following setup done.

Node A:

DB + CI : DB services will be under failover configuration. CI will be static

Node B:

ASCS+ Diag: ASCD services will be under failover configuration. Diag will be static.

ERS : can reside either on Node A or Node B.

2) Active - Passive: Here all the SPOF component like DB services, ASCS services and ERS lies on single node. Another node has static dialog instance only.

For this you can have the following setup done.

Node A:

DBCIERS+ASCS: DB and ASCS will be under failover configuration.

Node B:

Static Dialog instance resides on this node.

Hope this will help you in planning the HA setup. Best approach would be to discuss your scenarios with hardware vendor and chalk out the failover scenarios with them.

Regards,

Deepak Kori

[Former Member]

The configuration depends on your sizing too. I like configurations, where the Dev and Qas system reside on the failover side. In case node A fails Dev and Qas are stopped on node B and the productive system is started. On newer system we have installed more memory on node B so the Dev and Qas don't necessarily need to be stopped when a failover happens. The SAP extension for Service Guard supports these kind of setup.

So this would be scenario 3:

Scenario 3:

- Install CIDBASCS Prd on node A

- Install Dev and Qas on node B

HA scenario 3:

- If node A dies, Prd go to node B

- If node B dies, Dev and Qas go down (theoretcally they could also be clustered and switch to node A)

Cons against scenario 1+2:

- you have to size the HW twice as big, because you have only half power when a node fails.

- load balancing can be very complicated, often only one of the servers is loaded, while the other is idling

Scenario 2:

- will experience twice as many failures, because you have single points of failure on both nodes in parallel

- overall performance could be lower, because SAP and database are separated

Of course my scenario 3 is best used on smaller to medium sized systems, because the prod has to fit on one single server. However you can still add applications servers if you like.

Cheers Michael