HP Nonstop TMF - In Detail

The HP NonStop TMF transaction manager is the foundation for building fault-tolerant and disaster-tolerant applications on HP NonStop servers.

For applications that are TMF-enabled, TMF guarantees the consistency and the correctness of the Enscribe and NonStop SQL databases maintained by the applications.

It accomplishes this ability by ensuring that either all of the operations included in the application’s transactions are applied to the database, or none are. Transactions are durable, surviving any fault in a NonStop server, by safe-storing all changes to the database in a mirrored, diskbased audit trail. The TMF audit trail is the heart of TMF.

An application is not informed of the completion of a transaction until its changes have been safe-stored in the audit trail. After a system recovers from a failure, the audit trail’s contents are used to roll back any incomplete transactions. The audit trail can be used to restore files that have been corrupted or that have been accidentally deleted. TMF allows databases to be reorganized online.

Backups can be written to physical or virtual tape without application downtime. Archived audit trails can be accessed to restore corrupted or deleted files while the system is fully operational.

Because transaction durability is guaranteed by the audit trail, the actual data in the databases does not have to be written to disk immediately. Instead, updated data blocks can be batched and written to disk at the convenience of the operating system. Thus, application processing capacity and response times are significantly enhanced by TMF. TMF provides an interface which allows it to support the management of transactions distributed across a heterogeneous network of other systems.

As will be described in upcoming sections, the TMF audit trail plays a key role in many of the HP NonStop data replication facilities. It provides a log of all database changes to be replicated to other systems.

HP NonStop Remote Database Facility (RDF)

HP NonStop Remote Database Facility (RDF)

HP NonStop RDF extends the NonStop server’s fault tolerance to disaster recovery.

Using the audit trail generated by NonStop TMF and operating-system level services, RDF instantly replicates audited database changes to identical target databases on one or more target systems.

Thus, in the event of a failure of a primary system, the application can be recovered to a backup system within minutes and with minimal data loss. RDF will back out any incomplete transactions on the target system when the target system takes over processing.

RDF can be configured to protect individual audited files or tables, the contents of one or more subvolumes, or entire disk volumes. It can replicate changes to one or more target systems.

Alternatively, it can replicate changes from many source databases to a single target database to provide a single backup for multiple production systems. RDF is useful for upgrading hardware and software with no application downtime. The backup system is taken offline and upgraded. Application processing is then redirected to the upgraded backup system, and the other system is upgraded. Though applications on the target system cannot open databases for write access that are being replicated by RDF, they can read the target databases. On the target system, other applications can be running, replicating their database changes to other target systems. If the backup system has to take over processing, the backup applications open the application databases for write access before continuing application processing. There is a small time latency between when transactions are executed on the source system and when they have been successfully replicated to the target system.

Thus, transactions in the replication pipeline may be lost due to a primary system failure. If this loss is unacceptable, RDF implements a Zero Lost Transaction (ZLT) feature. With ZLT, the audit trail disk mirrors are split; and one mirror is located up to 100 kilometers from the production system. No transaction completes until it has been successfully written to the remote mirror. The remote mirror is available to a backup system if the backup system needs to take over processing with no lost transactions.

Nonstop Business Continuity Product Suite : An Introduction

Product Suite Consists of

1. HP NONSTOP TMF
2. HP NONSTOP Auto TMF
3. HP Nonstop RDF
4. HP Shadowbase for Nonstop and other servers
5. HP Nonstop SQL Data Definition Language Replicator(SDR)
6. HP Nonstop AutoSYNC

HP NonStop AutoTMF

HP NonStop AutoTMF

Many legacy applications have been written without using the HP NonStop TMF transaction manager.
In such instances, HP NonStop AutoTMF supports existing application unaudited Enscribe files, allowing applications to be TMF-protected without the need for coding changes or recompiling.

AutoTMF permits such applications to take advantage of the TMF transaction protection of audited databases and to benefit from the performance improvements delivered by TMF.

AutoTMF can be configured to audit only one or any subset of applications or files. Once a file is placed under AutoTMF protection, AutoTMF will dynamically determine effective and efficient transaction boundaries within the application.

These delineated transactions are then written to the database under the TMF transaction manager. By instrumenting an application with AutoTMF, changes to unaudited files are written to a TMF audit trail.

This is a requirement for replicating the data in unaudited files to other databases.

Few Nonstop Fundamentals - V

CHECKPOINT

A checkpoint is a Nonstop specific Term relevant only to Nonstop Process Pairs. The purpose of a checkpoint is to ensure that the two process have identical process state so that if there is a single point of failure, the backup process can take over the completion of processing  without the requesting process needing to do any exception handling.

The purpose of checkpoints are to make primary process failures transparent.

Checkpoints occur on critical region boundaries to ensure all-or-nothing processing process semantics.

The primary process sends checkpoints on critical region boundaries to the backup and waits for acknowledgement before proceeding to the next critical region step.

Its upto Application Architects to determine what the critical regions of a process are.

Few Nonstop Fundamentals - IV

Massively Parallel Processing(MPP) VS 

Symmetical Multi-Processor(SMP)

MPP is a System Architecture that presumes each processor has its own RAM and that other processors can't access it. 

Each processor is presumed to operate as an autonomous system that has mechanism for coordinating work with other processors - typically message bus.

SMP is a system Architecture that presumes two or more processors access the same shared RAM.

To do so, Synchronization mechanisms, (typically semaphones) are used to coordinate access between processes running in competing CPUs.

Few Nonstop Fundamentals - III

IPM

IPM means Inter Process Message, is a type of message that is sent across a message bus to tie the processors of an MPP system together.

In the context of Nonstop, this is a message that is sent between any two processes within a given Nonstop Node or across an EXPAND network on Nonstop Nodes.

These messages are unsolicited requests that are sent to the $RECEIVE message queue of a specified process by another Nonstop Process.

On Nonstop X, InfiniBand is the bus fabric.

RDMA

RDMA means Remote Direct Memory Access, is a capability specific to ServerNet and Infiniband architectures where by the memory of a given CPU can be directly referenced by a process in another CPU without the CPU that owns the memory being involved in the I/O operation.

There is no context Switching, Interrupt or Trap Handling needed to service the I/O.

Everything happens in user mode during the process execution time slice for maximum efficiency.

Few Nonstop Fundamentals - II

Multi Core Processors

Multi Core processors are a more recent evolution of CPU architectures where by sophisticated chip logic and compilers allow processing to be broken into threads of critical regions and submitted to multiple cores simultaneously.

Instead of processor needing to run at ever faster clock speeds, cross-section computing power can be aggregated across multiple processors to achieve the same effect.

TMF

TMF Means Transaction Monitor Facility.
Transaction monitor is a mechanism that  can ensure ACID properties  of a given database update or transactional all-or-none execution consistency respectively. In Nonstop Systems, this transaction monitor is achieve through TMF subsystem.

Its fully integrated with IPM and file system.

Few Nonstop Fundamentals - I

Nonstop Process Pair

A Nonstop Process Pair refers to
       - a primary process residing in a particular CPU
       - and a designated backup CPU residing in another CPU.

They share the same process name and but have a difference Process ID(CPU:PIN).

They can be configured to be amneisa backup processes(i.e they know nothing about the state of other process) or

they communicate with each Other to preserve state using checkpoints.