High-Availability building for Central System Installations

[pr_srinivas]

Hello Experts

We have  2 Installations of  [ABAP+JAVA] installed as Central System on a single host [ With out High Availability]. The Systems went live and are in Production from last 2 years.

Now we have procured new hardware and wanted build High-Availability for above 2 Instances and want to achieve  over IBM PowerHA (formerly HACMP - High Availability Cluster Multiprocessing).

i.e Database + Central Instance is running on single host which is live from 2 years & it is dual stack System.

Now to achieve High-Availability, over HACMP cluster We want to run DB on 1 host and CI on 1 host.

What should be my series of steps to make Central system running on host to Distributed system [ CI & DB on different host].

Thanks for all your suggestions & expert advises in advance.

Cheers,

PR

[Former Member · Accepted solution]

Hi,

to get a central instance installation into a distributed installation, you will have to perform a system copy to the new hosts. That means:

I would propose to put the new servers in a kind of "network panic room", where they can do no harm - to test the production procedure you will have to use the same hostnames (I suppose), start and stop them and do connection tests. If you install just in your central production network, this won't work.

For each system:

- use SAPInst to perform system copy for source host

- use SAPInst "distributed installation" to install central services on target host a

- use SAPInst "distributed installation" to install database instance on target host b (including RDBMS)

- integrate database and SAP resources into your IBM cluster configuration (not sure how exactly this is done with your software)

- test cluster functionality

If this worked fine for both test runs (ABAP + JAVA are separate SIDs, right?), then write down the runtimes of the copies and plan the maintenance window.

Good luck!

Peter

[pr_srinivas · Accepted solution]

HI...Thanks for your reply

it is a   Dual Stack - System. I mean it is having only one SID.

Rgds

PR

[Former Member · Accepted solution]

Hi,

then the procedure should apply too - but you just have to do it once.

But I would recommend to double-check if you really need a distributed configuration. There's no problem with that, but you get some consequences which should be outweighed by concrete advantages (e.g. need for strong db performance, therefore db has to be installed on separate host).

Regarding this context, kindly consider:

- there's setup effort to achieve distributed configuration (as described above)

- there's higher maintenance effort in this environment (every network issue touches your SAP <-> DB connection)

- there's higher risk for system downtimes (if one host is off, your SAP will be off too)

What about installing the system as central instance on host a, and put a dialog instance on the other host b? IF host b is down, SAP instance itself will continue to run. If host a is down, HACMP cluster software has to shutdown and unmount dialog instance, switch central instance to host b (including virtual hostname) and startup again.

What do you think?

Regards,

Peter

[pr_srinivas · Accepted solution]

HI Peter ...Sounds metculous advise with ZERO Risk.

Now i have my basic question in  mind

Best Option -1

----------------------------------------------------

DB+CI on 1 HostA  <-----> DI on 1 HostB

if HostA fails -  DI will be killed on HostB & DB+CI will move to HostB

in this scenario - The host names will be mostly physical  But IP's are virtual

& not virtual like actual clustering - Virtual host Name & Virtual IP

Q : Don't  we expect any issue bringing up Java Systems in this context,

      ABAP - very sure , we do some tweak & bring it up

      JAVA - Not sure..we should test thinkg [ You expect any issues

Option -2

---------------------

When we select Best Option -1

What we are loosing with standard Implementation of DB & CI on different hosts?

not able to write down in text  -

What is great with Risky Option -2 compare to Best Option -1

Is it just comlete utilization of HW resources or any thing more , which i am missing in sight.,

Pl comment

&

Thanks for all yoru Deep Dive views on breaking this.

Rgds

PR

[Former Member · Accepted solution]

Hi,

regarding Option 1 - you are right, you'll need virtual hostnames and IPs. If you don't have, you will get problems with JAVA stack at least (although ABAP adaption to a new hostname is possible, I don't think it's suitable for failover procedure).

If you use virtual hostnames and install everything you need in SAN (DB files, RDBMS, SAP kernel, interface directories, ..) to get it "switchable", then it should work fine.

If you install as described, you'll use full hardware capacity for Option 1 too (as dialog instance uses host B fully).

For option 2, the additional advantage is that you have maximum hardware resources for database (as you don't have to share host with SAP central services). So every advantage of distributed installation focuses on database operation and it's boundaries towards SAP.

For example you would be more independent in regards of operating system configuration or versions (if needed you could configure something completely different on db host regarding kernel parameters, os version or even - in certain boundaries - os vendor), also cluster concept (Oracle RAC) can be configured more individually.

Of course you can do more nasty things on infrastructure side, e.g. you could setup a "shared database host" where databases of multiple systems reside. As this would be SPOF for all connected systems, you should makes it HA and DR-proof by setting up multiple (at least 2) clustered hosts in 2 datacenters. You see, it always depends on your specific requirements - means that there are special scenarios where distributed installation is without alternative. But if you have the choice, please consider the side effects.

Regards,

Peter

[pr_srinivas · Accepted solution]

HI...

Peter - Your approach thought is to be good, instead so many risks in production environemnt, we are investing  very little more on Computing and memory utilization, but was ok for ERP Head.

Note 821904 - Separating SCS instances for ABAP and J2EE

The above note also found interesting.

Rgds

PR