One of the key features that have spun up in recent time is the
concept of high availability. The desire to have a system which will never go
down. There are bigger risks and greater costs associated with it. One of the
major consideration people miss out is the infrastructure. An infrastructure
built up with very high availability should be aware of certain things in the
organization :
1.
Highly skilled admins ( Linux / Windows / Solaris / AIX )
- High
- High
- High Disk Space for any coredumps that might be created for any app error
- High performance and a back up system for applying fix packs
One of the major reasons sometimes HA is not pursued by
organizations and is not a success is because of technologies sitting beneath
the stack. Lets say if I want to have ITIM (Tivoli identity manager ) as HA ,
other than making all components of it to be HA, like LDAP , TDS , DB2 , I should
be focused on WAS or Weblogic on which the application is deployed. A broken
WAS will result in breach of SLA or HA. Then
Tivoli Identity Manager is full of Java
classes, which has memory leaks. The trouble increased when there is custom
code in Workflows or person account using java statements and printing out
outputs to system console. A not well tuned application code can be one of the
biggest challenges in HA.
Once these challenges are faced, we have to make sure that
each component in the application is performance tuned to perform the work it
requires to do. The term tuning essentially means “best setting for the system”
and not best setting in general. A 32 bit machine should not have a JVM of
4096kB , regardless.
Once all these steps are performed , then there is a need to
understand how the things will shape up in real world. Here comes the biggest
problem , the data. Here we have three challenges , data integrity , data
warehouse , and data store. The speed at which data can be accessed , modified
and use should not be at the cost of data security or HA , and otherwise.
Let us see in coming days how HA , Data security and Data
Accessing should be tackled.
