Oracle 10g – grids at tenth of the price
KISS and tell
A twin processor Lintel server for $5000 is the only rational building block for enterprise computing, Oracle's Larry Ellison said at a recent conference,
writes Peter Abrahams of Bloor Research. It is faster and nearly ten times cheaper than the equivalent power on a mainframe or UNIX server. This statement is at the heart of his drive for grid computing in the latest Oracle release 10g.
The problem is that you have to put a lot of them together in a grid to deliver the requisite power. They have to have the right OS, middleware and application software installed. Then they must be configured and connected together. If the load changes significantly, either because of a long term growth in transaction volumes, or by a sudden on-demand surge, they must be extended and reconfigured to cope. If there is a hardware failure changes must be made to reallocate the load.
All this is expensive and time-consuming and compounded when the software has to be maintained, upgraded or have emergency patches applied to multiple systems in a co-ordinated and error-free manner.
The total cost of the human effort required to keep these systems going, let alone the inevitable cost of human error, far outweighed the reduced hardware cost. IBM and Sun have been successfully using this true story for many years.
The seemingly obvious answer is for IT to take its own medicine and automate these manual, largely clerical, tasks. Surprisingly the traditional system management vendors did not reach that conclusion or provide the software.
Oracle, with RAC, had done some of this for its database for a couple of years. And if Larry's vision of a grid was to become reality it was essential to automate as much as possible of the manual system management.
Oracle has therefore looked at all aspects of system management. It conducted a survey and found that 26 per cent of sysadmins' time was spent in provisioning, 6 per cent loading data, and 55 per cent with monitoring and on-going maintenance.
In 10g, Oracle has put a lot of effort into this area with Enterprise Manager and believes it has reduced effort to a tenth of that which would otherwise be required.
How has it done this?
First, by greatly simplifying the installation of all Oracle software; any installation process now asks a minimum of two and a maximum of five questions and these can be pre-recorded to enable totally silent install.
The system stores information about all software, release levels and patches and will either automatically download patches and install them or lead the system programmer through this process so they can decide when to do the actual install.
All upgrades can now be done on-line without having to take the system down. New levels of the software are designed so that they can run in parallel with previous versions.
New servers can be cloned automatically from existing servers.
The system is monitored for performance and security issues. Policies can be defined that specify what to do when thresholds are reached, including adding extra servers on demand and releasing capacity that is no longer required.
Oracle has extended its database server and application server to provide more detailed monitoring information as well as taking advantage of the Enterprise Manager functions. Although EM works best when used with Oracle products it is not limited to supporting these products but can be extended to support anything in its eco-system.
It really does appear that Oracle has done enough to make the grid built out of commodity servers a practical and economic reality. As a by-product it has made the management of smaller systems easier as well.
This functionality is good news for existing Oracle users but it also makes Oracle more attractive to the rest of the IT world.
Sponsored: Benefits from the lessons learned in HPC