Dromedary Date Nut Bread

power or continuity '? Visions

One of the things that the IT world and why I welcome less waste and conceptual' when companies' advice (or alleged guru of the field) manage to persuade companies to acquire enormous computing power without thinking of the consequences of their actions. Machines with four or eight-way, with a capacity 'bloated memory, combined with some' RAID disk to prevent loss of data and redundant power supplies ... um ... sorry, but if the car dies, your business what of? Assuming you also have the best support contracts, managed to have a firm also has four hours (read: half-day) on your work ... Thank goodness there 'the Windows Solitaire.

Recently I see many of these reality ', which require a high capability' preparation. I am not sure of the people who are frightened of numbers or transaction processing (application, not database) of which the company needs. Pero 'I have often wondered why the' nobody (or hardly anyone) ever speaks of the model "load balancing". Perhaps not everyone knows that the clusters are not only important to ensure the continuity 'of a service, but can also be used to increase the capacity' of processing system in its complexity '.

I take as an example a project that I developed (in the laboratory, but it 's quite capable of operating in the real world) for a test. The problem was you can make a large amount of processing' format documents PostScript to PDF conversion.

For the development test environment I used virtual machines that ran a Windows Server 2003 Standard Edition (yes, the standard edition, so it is not 'cost-prohibitive). The first Windows 2003 server I have configured the hardening also now making application, in order to have a virtual machine to be replicated on all the others. On completion of the installation I sealed the system with sysprep (other system component) and I copied the disk image to a DVD. I launched the first virtual machine, completed its configuration, and then put the domain (in this case I had already 'two DC running). The three other nodes I simply copied my image from the DVD and start the virtual machine from that disk. In less than two hours I had 4 servers perfectly ready to be configured in the cluster.

For the clustering load balancing I used the Network Load Balancing Microsoft Windows 2003 Server, I created the cluster and added the 4 nodes. The service was of course the shared file and printer sharing service. Then I started to send the files to the cluster and verified which were analyzed by the various server nodes. Of course I could not expect tremendous performance (the laboratory was running on a uniprocessor machine) but 'changing hardware architecture (possibly via a server Xeon Dual Core processors, or even games) the advantage was obvious.

The solution presented would be possible to develop a more 'large amount' of data with a single machine (reducing the waiting time of the output) to save on hardware (nearly 20%) of the proposal made by a 'On the other hand (a quad machine with a roar of memory and disk configurations unintelligible). The advantage of a true solution for load balancing (over for hardware that brings the same volume calculation) and 'continuity' of service. Suppose (as e 'in the case of the project described) that the PDF is crucial for the business' company and that a lack of delivery of the PDF you make a loss not only temporary but also image. What happens if the system stops? Between servicing, telephone calls from people who rail against the administrator of the system, the service call and prayers "is not the disk controller" spend between 2 to 4 hours before the problem is solved (already ', because 'found the courier traffic due to an accident, and if our SysAdmin opens the car assurance goes to hell). The damage ' fact: there are two hours to 20, closing date of the delivery of PDF, the next day there is 'an important event in which the company had invested money and can not' have the brochures from the printers. The system is solved, are the 19, the system and 'just broken, taking charge of the first file ... is damaged, it must recontact the chart to put them back in the queue of processes ... pass 20, the system 'still in the middle' way to finish the conversion ...

definitely the solution all on one "does not work, especially because 'that are real benefits in these applications. If she died a single virtual machine (dump system, crash the application), other nodes would however, continued to operate, and the delay would have been really low, without compromising the image.

Business Continuity and High reliability '? Let's talk about ...