Future of High-Availability Solutions

Introduction

High-Availability solutions have evolved significantly over the past couple of years; many of today’s solutions combine mirroring or replication technologies with intelligent management software to enable geographic separation of cluster nodes and storage to provide disaster recovery within the High-Availability framework. But what will tomorrow’s High-Availability solutions look like?

Solution Architectures

There are probably three realistic architectures for high-availability which will exist once consolidation in the market place is complete;

Distributed Computing

Data is distributed among many machines. Changes are normally made on one machine and roll out over time to all machines. This model is in use today, in a simplistic form, by software vendors who use �mirrors’ for downloads like tucows or in distributed computing projects like the �search for extraterrestrial life’ (SETI).

This model allows fairly low powered machines to provide a high computing capability and more importantly a system resilient to large scale failure. However, conflicts can occur if the same �object’ is changed in two different places at the same time. This is normally overcome by running a master, for writes, and multiple slaves, for read operations. Problems can occur if the master is unavailable for any length of time.

Parallel Computing

Transactions are completed simultaneously on different servers at the same time. However, in contrast to the distributed computing model, the servers perform a co-ordinated locking arrangement, so that no conflict can occur. Today only a handful of vendors, like Oracle with their RAC solution, have begun to provide some of the components needed for true parallel operations.

The theoretical benefits of parallel computing include improved scalability and performance. Depending on the specific solution, the servers don’t need to be co-located or share the same disks but if performance is also important then the links between the machines must be very low latency, normally only available with co-location and specialised hardware.

There are a number of ways to achieve parallel computing operations; at the application level, at the operating system level and through middle-ware.

On-Demand Computing

Only one instance of an application processes all transactions but other instances are available to take over if a fault should occur.

This can be achieved using co-located hot-standby servers or disparate mirrored data. The mirrored solution is only practicable where performance is a not priority.

So What is Grid Computing?

The constant marketing push for new buzz words and news items has perpetuated a view that true grid computing for the enterprise is available, or will shortly be available. Distributed applications, like SETI, are already running in a form of grid like architecture but the enterprise needs databases and other applications where synchronised data must be provided to be useful. Even the director of SETI rejects the grid label, preferring the term “public resource computing”.

The term “grid computing” was originally coined by Carl Kesselman and Dr Ian Foster of America’s Argonne National Laboratory in 1998, they drew an analogy between the supply of computing power and the supply of electricity, which is delivered when and where you need without needing to worry about where it came from.

In 2000 Dr Foster refined his definition with Steve Tuecke to describe the essence of grid computing. They came up with three essential components;

1) coordinates resources that are not subject to centralised control …

(A Grid integrates and coordinates resources and users that live within different control domains—for example, the user’s desktop vs. central computing; different administrative units of the same company; or different companies; and addresses the issues of security, policy, payment, membership, and so forth that arise in these settings. Otherwise, we are dealing with a local management system.)

2) … using standard, open, general-purpose protocols and interfaces …

(A Grid is built from multi-purpose protocols and interfaces that address such fundamental issues as authentication, authorisation, resource