From: the INQUIRER
The OS that dare not speak its name
By €uromole: éåí øáéòé 03 ãöîáø 2003, 10:39
EVERYONE IS TALKING about Windows clusters, Unix clusters and Linux cluster. But all we are saying, is that the 20 year-old architecture of clustered OpenVMS can teach these whippersnappers a thing or two.
At OpenVMS.org there's a report about an OpenVMS cluster which handles the major processing for the Greater Amsterdam Police and naturally is required 24 hours a day, seven days a week.
What is unusual about this cluster is that it has been running since April 1997 despite the addition of machines, upgrades to the operating system, upgrades to applications, a changeover to a SAN architecture and even the physical relocation of the machines to a new site some seven kilometers away. Yet, through all these changes the cluster has provided continuous IT services to the Amsterdam rozzers.
The current up-time for the cluster is about 2430 days, or more than six years, with the installation of the oldest hardware, an FDDI concentrator, only 1650 days ago, while the oldest computer system was installed about three years ago.
The planning and execution of these changes would have no doubt presented their own challenges. But it appears the most difficult part was explaining to managers why it was unnecessary to shut systems down, even during the physical relocation.
This report becomes even more interesting when we factor in the conclusions from a study by TechWise Research entitled "Total Cost of Ownership for Enterprise Class Clusters" that was released in January 2002. This study is slightly dated and so we hope that it is updated soon and will include data about Linux clusters.
The TechWise report was based on 93 interviews and looks at the costs of ownership over a five year period for machines from Compaq (as it was then), Hewlett-Packard, Sun and IBM, with approximately equal numbers of systems from each manufacturer.
In our experience, the report from TechWise is unusual in that it takes into account the actual costs of downtime as discovered in those interviews. It reached the startling conclusion that if downtime costs $50,000 per hour for a medium size enterprise system, then based on the average reported downtime, this is equivalent to 50% of the total cost of ownership over five years. If downtime rises to $100,000 per hour then on average it accounts for 80% of total costs. By contrast the acquisition is 26% in the first instance and just 10% in the second.
It seems that the financial advantage of Linux might be far less than people imagine especially if it accrues more than the average of 9.2 hours of downtime used in the above calculations, but that's a story for another day.
The upshot of the TechWise report was that when the average cost of downtime, $71,000 per hour, is included in the total cost of ownership over five years, the OpenVMS clusters are a clear winner. That's mainly due to their average annual downtime being 11.0 hours compared to the IBM clusters at 18.4 hours, HP clusters at 22.8 hours and Sun clusters at 28.7 hours.
Depending on the cluster configuration, the financial advantage of running OpenVMS was between $1.7 million and $6.8 million.
Your mileage may vary of course and these figures are somewhat dated, but they do give cause for thought -- both in relation to the clusters mentioned here and how Linux might stack up against them. It seems reasonable to suggest that if Linux clusters are used in environments where downtime is expensive, then it would only take about 25 hours downtime per year before the cost advantage of using Linux might be completely negated.
As to the cluster operated by our Dutch friends, if that operated in a commercial environment and had no downtime in five years, then compared to clusters from other manufacturers with their average downtime, the savings could easily exceed $10 million.
That's an amount that buys quite a few applications and certainly covers the costs of personnel.
These stark facts may also shake HP from construing VMS as Vendor Marketing Silence, and maybe make them think the neglected OS could be a Veritable Money Spinner. µ
The OS that dare not speak its name
By €uromole: éåí øáéòé 03 ãöîáø 2003, 10:39
EVERYONE IS TALKING about Windows clusters, Unix clusters and Linux cluster. But all we are saying, is that the 20 year-old architecture of clustered OpenVMS can teach these whippersnappers a thing or two.
At OpenVMS.org there's a report about an OpenVMS cluster which handles the major processing for the Greater Amsterdam Police and naturally is required 24 hours a day, seven days a week.
What is unusual about this cluster is that it has been running since April 1997 despite the addition of machines, upgrades to the operating system, upgrades to applications, a changeover to a SAN architecture and even the physical relocation of the machines to a new site some seven kilometers away. Yet, through all these changes the cluster has provided continuous IT services to the Amsterdam rozzers.
The current up-time for the cluster is about 2430 days, or more than six years, with the installation of the oldest hardware, an FDDI concentrator, only 1650 days ago, while the oldest computer system was installed about three years ago.
The planning and execution of these changes would have no doubt presented their own challenges. But it appears the most difficult part was explaining to managers why it was unnecessary to shut systems down, even during the physical relocation.
This report becomes even more interesting when we factor in the conclusions from a study by TechWise Research entitled "Total Cost of Ownership for Enterprise Class Clusters" that was released in January 2002. This study is slightly dated and so we hope that it is updated soon and will include data about Linux clusters.
The TechWise report was based on 93 interviews and looks at the costs of ownership over a five year period for machines from Compaq (as it was then), Hewlett-Packard, Sun and IBM, with approximately equal numbers of systems from each manufacturer.
In our experience, the report from TechWise is unusual in that it takes into account the actual costs of downtime as discovered in those interviews. It reached the startling conclusion that if downtime costs $50,000 per hour for a medium size enterprise system, then based on the average reported downtime, this is equivalent to 50% of the total cost of ownership over five years. If downtime rises to $100,000 per hour then on average it accounts for 80% of total costs. By contrast the acquisition is 26% in the first instance and just 10% in the second.
It seems that the financial advantage of Linux might be far less than people imagine especially if it accrues more than the average of 9.2 hours of downtime used in the above calculations, but that's a story for another day.
The upshot of the TechWise report was that when the average cost of downtime, $71,000 per hour, is included in the total cost of ownership over five years, the OpenVMS clusters are a clear winner. That's mainly due to their average annual downtime being 11.0 hours compared to the IBM clusters at 18.4 hours, HP clusters at 22.8 hours and Sun clusters at 28.7 hours.
Depending on the cluster configuration, the financial advantage of running OpenVMS was between $1.7 million and $6.8 million.
Your mileage may vary of course and these figures are somewhat dated, but they do give cause for thought -- both in relation to the clusters mentioned here and how Linux might stack up against them. It seems reasonable to suggest that if Linux clusters are used in environments where downtime is expensive, then it would only take about 25 hours downtime per year before the cost advantage of using Linux might be completely negated.
As to the cluster operated by our Dutch friends, if that operated in a commercial environment and had no downtime in five years, then compared to clusters from other manufacturers with their average downtime, the savings could easily exceed $10 million.
That's an amount that buys quite a few applications and certainly covers the costs of personnel.
These stark facts may also shake HP from construing VMS as Vendor Marketing Silence, and maybe make them think the neglected OS could be a Veritable Money Spinner. µ
Comment