How many servers do I need? Every one know this – or no one knows this

I was planning on installing a product on z/OS and was going through the documentation.  It is hard to see things that are not there, but I was surprised to see nothing about initial planning and how to set up the product.  I looked at other products – and they were also missing this information. It feels like this information is so obvious that every one knows – or the person responsible for the installation instructions only had experience of installing one image and documenting it.  (Tick the box, job done).

There are two reason for configuring more than one instance of a product

1. You want multiple copies of the same configuration
2. You want a different configuration.

This is obvious but it took me half an hour to realise this.  I’ll cover the implications of these decisions below

You want multiple copies of the same configuration

There are many reasons for this:

1. You have to support more than one LPAR.
2. You want to have more than one instance on an LPAR for availability, scalability and performance.  For example with a z/OS queue manager; it can support up to 10,000 channels, and log at about 100MB a second.  If you want to do more than this you need a second queue manager.
3. You want availability by having instances running on multiple LPARs, so if one LPAR is shutdown  work can continue to flow to the other LPARs.

You want a different configuration

The main reason for this is isolation:

1. You have different environments for example production and test.
2. You have different customers for example you support, bank1 and bank2.  You want isolation so if bank1 fills up the disk space, bank2 is not affected.
3. You want bank1 and bank2 systems to use different certificate authorities, so bank1 end users cannot connect to bank2 systems.
4. You have different performance criteria – you configure bank1 systems to have higher priority in LWM than bank2 systems.
5. If you have bank1 and bank2 sharing a system, and you want to restart it, you have to get agreement from both the end users.  Getting agreement for a date from one bank is easier than getting agreement for a date from multiple banks.

Implications of “you want multiple copies of the same configuration”

Ideally you want to replicate a system with very little work – a so called cookie-cutter approach.

Configuration files

Create self contained configuration files.  For example with Liberty Web Server on z/OS each server has its own server.xml file.  Create a file called keystore.xml and include that in the server.xml file.  The server.xml configuration file may look like

<server>
<include location="${server.config.dir}/keystore.xml"/> 
<include location="${server.config.dir}/mq.xml"/> 
<include location="${server.config.dir}/saf.xml"/>
instance specific data
</server

With JCL you can use

//OUTPUT1 INCLUDE MEMBER=....

to include common configuration.

TCPIP ports

You are likely to use the same port number across different LPARs.  You can define a port as a SHAREPORT, and have multiple applications listening on the same port number on the same LPAR.

Started task userid

Have the instances start with the same userid, so they have the same  access to resources.

Liberty profilePrefix

Have the instances use the same profile prefix, default BBGZDFLT.  This is defined using RDEFINE APPLY BBGZDFLT…

Define who can access the instance using SAF EJBROLE for example giving different groups of people access to the  BBGZDFLT.zos.connect.access.roles.zosConnectAccess profile.

Isolated instances

If you want to isolate instances, then use the above list and make sure you use different values.

Future proof your definitions.

An advantage of “define something once in an include file, and reuse it” is that if you change the contents, for example /usr/lpp/mqm/V9R1M1/java/lib, you change it once, and restart the servers.

If the servers each have a unique configuration file, you have to make the same change in each file.   This can be good – for example you want to change this server this week, and that server next week.

You could also have an alias /usr/mq pointing to /usr/lpp/mqm/V9R1M1.   When you want to change the version of MQ, change the alias and restart the servers.  To undo the change, change the alias to the old version and restart.  This is much easier than changing the individual files ( how many change records would you have to raise?).

Liberty Specials.

With Liberty you have have one JCL procedure, and start multiple servers.  This environment is a mixture of multiple copies of the same configuration, and isolated instances.

Each server has its own server.xml file, but uses the same JCL file, so common userid etc.

You can start a liberty server

s baqstrt,parms='stockManager',jobname=smanager

And have WLM classify this under SMANAGER.

The default Liberty profile is stored under /var/zosconnect.  If you want to have a copy of this running on two LPARs you will need  to have the profiles store in different places.  You could have

/var/zonsconnect/LPAR1/…

but if you need to move the server to a different LPAR this might point to the wrong directory.

You could change the procedure so you pass in the location of the profile.

s baqstrt,parms=’stockManager’,jobname=smanager,profile=’/var/zosconnect1/instance1′