You've hit on my reason for doing the superfarm project!!
If you follow IDC's evaluation guidelines for TCO, 23% of the cost of a solution is hardware, 37% is facilities and operations, and the remainder is people. It's not uncommmon to have only one or two VM systems programmer in a whole organization. I've seldom seen an Unix-based organization with a staff of less than 10.
Think also about application licensing -- at $495 a seat, MS Office is pretty expensive. What if you only had to buy ONE copy and share it? What if all your files were in ONE place, not scattered around on floppies or odd servers? The point about the mobility of your desktop is a fantastic one -- if your cube is under construction, just move to a working one -- voila, your environment is just the same as it was at "home". All your data is there. All your tools are there.
In terms of hardware cost alone, the IBM mainframes have substantially decreased in cost. Comparable Sun and HP systems are in the same ballpark, eg a 9672 with a base VM license and a useful amount of disk space is on the order of $200K, about the same as a couple Sun E4500s with Sun Cluster and some useful disks. The HP hardware tends to be proportionally more expensive than the Suns due to HP's propensity for overbuilding their hardware. If you take the total cost of ownership into account (eg, number of people needed to manage a large farm, management software and tools, operational costs and spares, setup fees, power and environmentals, etc), then the IBM solution is unbelievably cheap. The cost per WWW server in this arena is minimal -- most of the management tools are included in the VM OS, you don't need a lot of the others, the operational costs and spare kits remain the same as you increase the number of virtual servers, you don't have to pay someone to rack a new box when you have a new customer, and the power/environmentals are exactly the same if you have one system or 10,000). It's not the price of the hardware that is the telling part of Linux/VM -- it's the cost of operations and ownership that makes it a powerful idea. I design large data centers for a living -- on the scale of Exodus and Intel. BIG centers. These companies are spending in some cases 20 million for systems and up to another 10 million for management software. The space, power, etc costs them big bucks -- the whole project is more than a billion US dollars. I simulated that entire project on Linux/VM in 10x10 square feet using a $200K box. (Yes, that's me in Scott's article). Time from startup to 40,000 images -- 11.4 minutes. Time to deploy a new WWW server when I needed one -- a little under 90 seconds to create a new system, configure it, and fire it up. Time to build and install Apache: 7 minutes from source. Time to install a preconfigured RPM package of Apache on a new instance: 11 seconds. That's why this kind of configuration is interesting -- the ability to rapidly deploy and manage large configurations of servers reliably and cheaply. I could do dedicated WWW hosting for $5 a month with this setup. I don't need thousands of dollars of rack space, routers, cabling, etc -- it's all inside, and blindingly fast. For small companies, this probably isn't a solution unless they already have a S/390 box. The focus is really more on large farm systems or on companies with very fast ramp ups that need to depoly systems rapidly. It works. It scales. It's mainstream. It doesn't cost a fortune. It has 24x7 no excuses reliability. It handles disk I/O at a phenominal rate. It handles network I/O at a phenomenal rate. In contrast to the Sun E10K, the VM solution scales to many more servers with much finer control than is present in the Sun partitioning scheme. You can have a maximum of 24 processor boards in an E10K, and you can only divide partitions on processor board boundaries. No such limit exists in VM. You also need to reboot the system to repartition -- VM stays up when you add a new system (after all, it's just another user). One of the things Scott's article didn't mention was the ability to bring up bochs and Intel-based systems in these Linux instances. I've run MS Exchange for Intel on a NT system running on bochs on Linux/390. It's godawful slow at the moment, but the fact remains: it worked. Compaq, Dell, etc are just starting to reinvent the reliability and serviceability features of the S/390 hardware. Which would you rather bet your business on -- 30 year old *proven* technology, or 3 month old "next gen" stuff? Think carefully about this one -- it could mean the difference between success and being bankrupt.
Slashdot Mirror
You've hit on my reason for doing the superfarm project!!
If you follow IDC's evaluation guidelines for TCO, 23% of the cost of a solution is hardware, 37% is facilities and operations, and the remainder is people. It's not uncommmon to have only one or two VM systems programmer in a whole organization. I've seldom seen an Unix-based organization with a staff of less than 10.
Think also about application licensing -- at $495 a seat, MS Office is pretty expensive. What if you only had to buy ONE copy and share it? What if all your files were in ONE place, not scattered around on floppies or odd servers? The point about the mobility of your desktop is a fantastic one -- if your cube is under construction, just move to a working one -- voila, your environment is just the same as it was at "home". All your data is there. All your tools are there.
It's just so...obvious to do this.
In terms of hardware cost alone, the IBM mainframes have substantially decreased in cost. Comparable Sun and HP systems are in the same ballpark, eg a 9672 with a base VM license and a useful amount of disk space is on the order of $200K, about the same as a couple Sun E4500s with Sun Cluster and some useful disks. The HP hardware tends to be proportionally more expensive than the Suns due to HP's propensity for overbuilding their hardware. If you take the total cost of ownership into account (eg, number of people needed to manage a large farm, management software and tools, operational costs and spares, setup fees, power and environmentals, etc), then the IBM solution is unbelievably cheap. The cost per WWW server in this arena is minimal -- most of the management tools are included in the VM OS, you don't need a lot of the others, the operational costs and spare kits remain the same as you increase the number of virtual servers, you don't have to pay someone to rack a new box when you have a new customer, and the power/environmentals are exactly the same if you have one system or 10,000). It's not the price of the hardware that is the telling part of Linux/VM -- it's the cost of operations and ownership that makes it a powerful idea. I design large data centers for a living -- on the scale of Exodus and Intel. BIG centers. These companies are spending in some cases 20 million for systems and up to another 10 million for management software. The space, power, etc costs them big bucks -- the whole project is more than a billion US dollars. I simulated that entire project on Linux/VM in 10x10 square feet using a $200K box. (Yes, that's me in Scott's article). Time from startup to 40,000 images -- 11.4 minutes. Time to deploy a new WWW server when I needed one -- a little under 90 seconds to create a new system, configure it, and fire it up. Time to build and install Apache: 7 minutes from source. Time to install a preconfigured RPM package of Apache on a new instance: 11 seconds. That's why this kind of configuration is interesting -- the ability to rapidly deploy and manage large configurations of servers reliably and cheaply. I could do dedicated WWW hosting for $5 a month with this setup. I don't need thousands of dollars of rack space, routers, cabling, etc -- it's all inside, and blindingly fast. For small companies, this probably isn't a solution unless they already have a S/390 box. The focus is really more on large farm systems or on companies with very fast ramp ups that need to depoly systems rapidly. It works. It scales. It's mainstream. It doesn't cost a fortune. It has 24x7 no excuses reliability. It handles disk I/O at a phenominal rate. It handles network I/O at a phenomenal rate. In contrast to the Sun E10K, the VM solution scales to many more servers with much finer control than is present in the Sun partitioning scheme. You can have a maximum of 24 processor boards in an E10K, and you can only divide partitions on processor board boundaries. No such limit exists in VM. You also need to reboot the system to repartition -- VM stays up when you add a new system (after all, it's just another user). One of the things Scott's article didn't mention was the ability to bring up bochs and Intel-based systems in these Linux instances. I've run MS Exchange for Intel on a NT system running on bochs on Linux/390. It's godawful slow at the moment, but the fact remains: it worked. Compaq, Dell, etc are just starting to reinvent the reliability and serviceability features of the S/390 hardware. Which would you rather bet your business on -- 30 year old *proven* technology, or 3 month old "next gen" stuff? Think carefully about this one -- it could mean the difference between success and being bankrupt.