How NonStop keeps running
NonStop's availability comes from a few ideas working together: independent processors that share nothing, a message-based operating system, process pairs that keep a backup ready to take over, components that stop at the first sign of an error, mirrored disks, and TMF, which makes transactions all-or-nothing.
Shared nothing
A NonStop node contains several processors (CPUs). Each has its own memory and its own copy of the operating system; they share no memory and talk only by messages over a redundant system interconnect. Because nothing is shared, one processor can fail without corrupting the others. Nodes can be joined into a network of many nodes with the Expand networking product.
Process pairs
Critical system and application processes run as process pairs: a primary process in one CPU and a backup in another. The primary sends checkpoints of its state to the backup. If the primary's CPU fails, the backup takes over from the last checkpoint and the work continues.
Fail fast
A NonStop component that detects an internal error stops immediately rather than carrying on with possibly bad data. That sounds harsh, but with a backup ready, stopping fast is safer than continuing wrongly.
Mirrored disks and dual paths
Disk volumes are usually mirrored: every write goes to two disks, so either can fail. Disks, controllers and network connections are reached by more than one path.
TMF: all or nothing
The Transaction Management Facility (TMF) protects database changes. Changes made inside a transaction are recorded in audit trails; at the end the transaction is either committed or backed out completely, even across several processors and disks. After a failure, TMF can back out incomplete transactions and recover files, much as logging and recovery do for Db2 and CICS on z/OS.
What is the process that takes over when a primary process's CPU fails? (one word)
Show a hint
The other half of the pair.
Show the solution
The backup process.
Common mistakes
The backup does not run the work in parallel; it keeps the primary's state from checkpoints and takes over from there.
The other CPUs continue, and process pairs move to their backups.
Incomplete transactions are backed out by TMF; check its audit trails and status before trying to repair data by hand.
What you will see at work
- Operators watch CPU status, process pair status and disk mirror status; a 'down' half of a mirror is repaired online.
- Application designers place a pair's primary and backup in different CPUs and keep checkpoints small.
- The same reasoning about single points of failure is used when designing z/OS sysplex availability.
Key terms
Check your understanding.
Take this lesson's quiz and save your progress. Free.