[link|http://www.linuxuser.co.uk/articles/issue15/lu15-smoothiron.html|Here]. It's the story of an ISP which replaced a server farm with an IBM zSeries with Linux. Maybe not quite what you're looking for, but...

Telia Net saw that by using this solution it could save time and money. "The (virtual) machines are completely independent," Laking points out. "This is why consolidating multiple physical servers on to Linux on zSeries can be so attractive to many businesses. If you take the typical paradigm of one app per server that you get with NT and indeed with many Unix servers, you can just map that directly on to a virtual server environment. Each virtual server running Linux can be running exactly the same as it would on a physical server."

[...]

Since IBM s zSeries 64-bit z/Architecture machines only became available at the end of last year, the machine currently in place is a 31-bit addressing, S/390 server with 12 processors. SuSE is the flavour of Linux that was chosen. "There was no particular reason to choose SuSE. Red Hat was another good option, but we had to choose one," Wulff Riedl says diplomatically. Telia Net has around 1500 virtual servers running on the hardware at present, which already represents a significant money saving as compared with a non-virtual implementation. "We have made a cost calculation and we think that a virtual server running on this mainframe system is about 10 per cent of the cost of running on the different platforms," says Wulff Riedl. "So we are able to make a cost reduction of around 75-90 per cent."

This claim is backed up by Laking's findings: "Even at the 25-30 physical server point it starts to make sense to virtualise on the mainframe, but the moment you start to talk about 100 servers or more then the savings really do become substantial. So that figure of 10 per cent I find completely believable."

Indeed, the cost savings attributed to the mainframe deployment come from many different areas. Doug Neilson, a systems consultant for IBM enterprise systems, explains that savings can come from unlikely sources: "A lot of the savings are going to be on infrastructure, space, power and cooling - it's ironic that what used to be labelled as the mainframe dinosaur now uses dramatically less power in cooling than a typical server farm. If you think about a typical Intel server consuming, lets say a Kilowatt, well the biggest z900 mainframe consumes around 15KW and is capable of functionally replacing a couple of thousand NT servers, so the cost savings in power can be dramatic. Just on electricity power savings the mainframe cost can be justified."

"One of the major savings is in manpower," says Laking. "You do not need as many people to manage this environment because of the ability to virtualise and to build infrastructure using a mixture of Linux system management tools and VM tools. As a combination you can pick and choose and what you end up with once you've done all the automation is something that is much, much easier to manage." Updates to the system are easily shared, one update can be performed and the whole system can be updated from this one copy, if required. As Neilson explains, systems management is also made easier: "It can be a Linux application, so it can be a graphical interface or the customer can drop down to the next level to talk to the VM OS if they have the skills and use tools on the VM environment. But mainly it's an automated software process, so much so that they can create a series of empty servers and have them ready to go."


(Emphasis added.)

HTH.

Cheers,
Scott.