1 person found this helpful
You have most of the big stuff. One thing you didnt mention was using vmotion, assuming you are using vmotion, to prevent downtime during server HW maintenance and ESX OS patching.
You mention Ratio - and on DL585 I feel comfortable running as many as 30-40 VM's - but you what it really should be called is increased cpu resource utilization.
Even if you use a DL380 - i use two of them - you can put more than 20 Systems on both of them. So it will demand on you hardware what ratio you will get.
I was using a 20:1 ratio, per the intel white paper, for the low hanging fruit,
My plan was to start with that, add in some of the lower ratios, until I hit about 80 percent and try to average it out to show a good number.
1 person found this helpful
Don't forget that consolidation is more than virtualising - it is also worth looking for multiple app instances that can be consolidated.
Im my experience IIS and SQL especially tend to be deployed with many apps and consolidating these into fewer instances can be a cost saver if you can get around the "it's my server" mentality.
Also, I seem to recall HP have a product that ties into your desktop virtualisation thoughts but I can't remember the name of it, only that at a HP roadmap session a presenter got very excited about it (this was about 12 months ago). It is apparently a popular solution with offshore call centres.
Power consumption reduction(someone care to walk me
through this in detail?)
My personal opinion is that reduction in power consumption is over-rated.
Based on the maximum power consumption figures at http://www.dell.com/downloads/global/corporate/environ/PE_2850.pdf as an example, the average 2U server would use somewhere around 400W of power. (it's actually less than that but this will give you a figure that is in favour of virtualising) Leaving room for a KVM switch you would fit around 20 such servers in a full height rack, resulting in a total power consumption of 8kW for a full rack. If you are paying $0.05 per kWh, that works out to just over $3500 per year for a fully loaded rack.
Even with a 20:1 consolidation ratio, you are only saving $3300 per year in power and paying more than that for the ESX license.
This doesn't take into account the associated savings you would see from airconditioning, physical real-estate etc, but you can see that the power saving argument is pretty flaky.
I'm also looking at virtualizing the desktops, but not
in the usual way. Thin clients RDP session into a vm.
I need to know how the performance will look on this
if anyone has tried it.
I haven't tried it on a large scale myself (only 2 XP VMs) but apparently it doesn't scale well at all. Have a look here for more details:
A couple of points you didn't mention were business continuity and reductions in overtime.
Making use of VMotion, if a physical host requires hardware maintenance, a BIOS upgrade or anything else that would usually require an outage, you can VMotion the VMs to another host and shutdown the host you need to work on without any disruption to the end users. That's your business continuity argument. This can all be done during working hours, which means less overtime being worked.
Yes, but when you're looking a couple hundred servers, that 3200 dollars per rack adds up. Add in cooling, which nearly doubles that number, and you're looking at around 5000 dollars per rack, per year.
Then look at UPS. 200 servers will take even a large UPS installation to red in no time flat. Meanwhile, that same UPS already in place would get us probably a day or more at a 20:1.
No one single point is the killer to my way of thinking, it's 10k here and 10k there, and suddenly you're a million dollars ahead.
A colleague of mine did some rough calculations using the max loads of the servers we moved from to the max loads on the servers we moved to. Based on that and the amount of electricty it would take to cool off the heat produced by the servers he calculated a savings of around $1,000 per year per server virtualized.
I have no idea how accurate that is, as I didn't check it out.
You are right, power savings are huge, and you should be factoring in some $$$ saved due to less cooling being needed.
We have 5 4-way boxes with single core CPUs. 20-1 with power savings of $3200 per ESX server = $16,000 per year. We also have 5 newer 4-way dual core and it looks like we will be able to hit 40VMs per ESX server, so $6400 per ESX server = $32,000 year. We calculate ROI on projects for 3 years, so $144,000 saved over 3 years, not including any reduced cooling savings, I would say that is significant. A lot of these little costs add up to a lot of money.
The point I was making is that, when compared to the cost of an ESX/VC agent license, the savings aren't huge.
Sure, them are some savings to be had, but when these are compared to the other benefits (savings in hardware purchasing, reduction in downtime etc.) the savings in power pale into insignificance.
Yes, but when you're looking a couple hundred
servers, that 3200 dollars per rack adds up.
You're also looking at a couple of hundred ESX licenses which reduces this quite a bit. That was the point I was making.
cooling, which nearly doubles that number, and you're
looking at around 5000 dollars per rack, per year.
Then look at UPS. 200 servers will take even a large
UPS installation to red in no time flat. Meanwhile,
that same UPS already in place would get us probably
a day or more at a 20:1.
That's a definite plus on the side of virtualization.
No one single point is the killer to my way of
thinking, it's 10k here and 10k there, and suddenly
you're a million dollars ahead.
Correct but for a full ROI you also need to take into account the associated costs, not just the benefits. In the case of VMware, with licensing, training etc. this is not an insignificant part of the equation.
Of course costs would be on the sheet also, TOTAL COST OF OWNERSHIP. TCO. Which is part of the ROI. You cannot calculate one without the other. At least not when you are presenting the migration. That is the spec I'm researching, having already completed citrix for the same migration, now I am presenting ESX.
Let me give you a for instance. I am assuming that energy costs alone will balance out the ESX licenses. What do we have left?
Increased time on the UPS.
Redundancy, and ease of maintenance.
Less need to build a new data center every 4 years or so.
Less switch ports=less switches(expensive)
Easy upgrades to servers, when you need more horsepower, simply give it more. When you upgrade physical servers, you no longer have to rebuild your server and apps for the new architecture.
And those are just a few positives.
Switches? Network and SAN Fabric. HBA's. It's tough to sell the soft-cost stuff (our time).
Yes switches, in our case it will be fibre, but also ethernet, gig ports aren't cheap when you go to 3 or 4 hundred(dual ports on several) and definately HBA's.
They never see time as a cost, they think since they are already paying us.....