Hyper-V Program Manager
The other day I was reading this article: http://blogs.zdnet.com/BTL/?p=4324 which proposes that virtualization is going to cause troubles for hardware vendors. It is certainly a logical conclusion to come to, I mean if you can run more operating systems on a single piece of hardware that means less money for the hardware vendors, right?
Well, this may be the case but I'm not quite convinced. The reason for my skepticism is because of two trends I've noticed:
Anyway, that's just my take on the situation.
Yup, I pretty much agree with you too Ben. Another point I'd like to put out there is Host Clustering -- as Virtual Machines assume more critical tasks, people turn to host clustering for higher uptimes. So 1x beefy box eventually becomes 2x, and so on. Anyone else got thoughts on this?
Cost savings are a huge driver for virtualization. Everyone has lightly used servers that are good candidates. Maybe they should make it up on consulting for virtualization.
I'm with Ben. I think that article makes a big leap.
Many organizations are also looking to purchase new technology to help them with virtualization - translation - Intel VT and AMD-V. They aren't going to get that by using their old server equipment. That means, new hardware investment.
Test and Dev = use old equipment. Serious server consolidation = buy new equipment.
Good points all.
I personally don't care if hardware sales would suffer. It's the same thing with gasoline vs electric cars... if electric cars drive gasoline companies bankrupt, good riddance. I'm still happy in my electric car. Of course the difference here is we'll always need hardware, just perhaps less of it.
Ben you make some valid points and I think as we see the demand for virtualization rise over the next year, the hardware vendors may start to change their game plans in regards to what they offer customers.
Dugie, in response to your question about Host Clustering, I see that as more of a temporary solution to increase uptime. Another alternative to clustering might be the implementation of high availability software. Unlike clustering high availability software can be installed with little to no downtime. Clustering could potentially knock out a system for hours or days at a time depending on your IT depeartment.
Does anyone have any thoughts on HA software?
Here's my math. Please comment on any errors.
I'm running a (hypothetical) computer center of, say, 100 servers. I do my analysis for server consolidation and find I have a bunch of machines running "less than 10% busy". Being conservative, I buy two new servers to consolidate 10 old ones. The new servers need more memory but don't need more CPU horsepower (5 x 10% still leaves plenty of headroom).
Clearly the hardware vendor is happy to sell me the two new servers, but the extra memory doesn't make up for the revenue of the eight servers not sold. And now I get HA "for free" by running inactive fail-over clusters as virtual machines.
Hardware vendors are already starting to see the pinch where over time they will ship less but bigger machines. Hence you see them attempting to sell customers on Blade PC'S that are installed in the datacenter and on Hosting desktops on Virtual machines in the datacenter all those drive higher margin server sells over lower margin desktops. I agree that most customer purchase new servers for virtualization projects. Some customers are still in the scale out mindset and do not want to consider large systems 8 core and up.