Virtualisation: Where are my savings? Three years ago you listened to vendors telling you how much we would all save from virtualisation. It seemed so plausible, but for most of us the savings never arrived. Why not? And what can we do about it? Good news: if you’re using VMware for virtualisation and Windows Server 2012, someone has finally worked out how to make …
"It seemed so plausible, but for most of us the savings never arrived. Why not?"
That heavily depends on the market. The article sounds as if you're targeting the whole industry, and if so it would be a clear sign of ignorance.
We're not saving eh? Ok, then let me spell something out for you.
In the old days you could only opt for either website hosting (getting your own private, but limited, webspace) and if you required more the only available options left were either hiring a server, or getting one of your own and make that available somehow.
The differences are even visible today. Take GoDaddy. They provide dedicated servers for you to use. Of course a decent one (8GB memory, 2x 1TB HDD, 10TB/month data traffic) will cost you approx. E 160,- / month at the time of writing.
GoDaddy also provides Virtual Private Servers and that shows a clear difference. Although the specifications are often lower you also pay a lot less. In this case a 'Deluxe' server would cost you approx. E 50,- / month. So you could easily opt to getting 3 servers instead of one.
And there are plenty of such examples. Leaseweb for example; dedicated servers starting off at E 30,- / month vs. cloud hosting starting around E 7,- / month.
So now you can get a full (virtual) server of your own to do as you like, while paying a whole lot less when you get yourself a physical server.
How is this not saving? How more obvious than this can it possibly get?
Sure, I know what's up. The article most likely talks about companies with in-house servers who more than often now get into contact with virtualization options but won't save a dime if they don't know how to use it. But that's not what the article told us, it basically addressed the whole market.
In the end it depends on the situation at hand if you'll be able to save up or not. The tool doesn't make the solution, it can only provide one. But that doesn't automatically make it the right solution for every problem.
On a like for like replacement, if your power costs are lower than the cost of the VMWare licenses, you're not likely to save anything by doing a like for like replacement.
But...
That assumes that your server estate is correctly sized. Which is usually not, most servers are oversize and VMWare knows that, so that you can fit 100 guests on the same VMWare host and save a bundle on power, plus the capex costs of all that unused hardware.
And the amount of tears saved by being able to stand up and down quickly servers for tests, quick one offs, etc has also its value. But don't forget that a fat desktop with VirtualBox can give you about the same flexibility in this area.
By my own experience, if you have less than 100 servers on your site, VMware is not likely to save you much, if any.
"And the amount of tears saved by being able to stand up and down quickly servers for tests, quick one offs, etc has also its value"
This is exactly the sort of thing that is valuable. The Reg in particular seems to have an oddly limited view of 'savings' - if it isn't on the bottom line of the capital costs for the initial project setup it doesn't seem to count.
Benefits of greater business "agility" or whatever this week's buzzword is are slightly harder to quantify but do amount to real cost savings, being able to run live standby / failover / DR systems is similarly beneficial - spending the same, or more, up front can still be a saving if you end up doing more, doing it faster, or doing it with greater robustness and lower risk.
All I've found is that instead of speccing a physical box (say, £6k?), with OS & sundry licensing (£1k, as the box wasn't particularly fruity) now I'm budgeting for core/RAM/hdd use (and they tell me there's never enough!) but the OS & hypervisor licensing is still a significant % of what the old hardware+license cost was, and I don't even have a shiny new box to show for it at the end of the day.
OK, hosting charges should be significantly less over the term, but there's precious little saving to shout about when bringing a solution together.
If you haven't saved any money with virtualization either your doing something really wrong or your workload is able to tap out the underlying hardware(very rare situation) so your not able to oversubscribe the systems. Even for small numbers of systems the savings (operational at the very least) is there. For small setups they do not need vSphere enterprise + (until vSphere 5.1 the only reason my org went with enterprise+ is because we were forced into it due to memory and/or CPU core license limits in lower versions - currently our main production is still on ESX 4.1, I have no immediate plans to change). Even something as simple as the vSphere essentials acceleration kit ($560 for 6 sockets) is a damn steal. I remember saying "SHUT UP AND TAKE MY MONEY" back when ESX was ~$3500/socket(no vMotion or anything like that for that price!). We saved TONS back then.
Even going back to my first production vmware deployment (e-commerce site) all the way back in 2004 (VMware GSX 2.0 if I remember right), saved the company big time by not having to have a bunch of extra hardware to launch a product in production (last minute fuckup by dev forced us to deploy new site on dedicated system). Company could not get enough hardware on site in time to launch in ~100 hours, so we took a couple VMware boxes(from the QA group) and threw the ~half dozen VMs on them that were required by the multi tier application. It was a fun project, even though it took about 60 hours to get the application ready for production. The QA group was crippled for about a week but the customer didn't sue us for not being able to launch on time, which I think was a fair trade off.
Going back to an Oracle DB consolidation I did in 2007, we had several distinct Oracle servers, and by leveraging vmware on a pair of single socket quad core processors I saved the company tons of cash by being able to consolidate multiple Oracle DBs on the same hardware (in one case it was on DAS storage, in another it used our existing SAN which our main production oracle DBs were running off of - on dedicated hardware). Oracle SE licensing is/was based on per socket, not per core, and by consolidating even two DB servers onto one we saved tons of $. And we kept the flexibility of being able to perform maintenance on different dbs without impacting the others(upgrades, downtime, whatever). VMware officially did not support single socket systems, and Oracle did not support running on top of VMware(at the time). Wasn't a big deal though. If there was an issue with Oracle I had the means to reproduce it on physical hardware, and blah blah blah..
The savings since memory prices have imploded and capacities have exploded have just gone through the roof since (my current dual socket Opteron systems are about to get their memory doubled to 384GB). Effectively doubling our VM capacity and we didn't have to add any servers, any licensing, even any storage. Proper capacity management at work!
Operating effectively in virtualization does in many ways require a slightly different way of operating, but if you can figure it out, and if your workloads are typical, then you will save regardless.
There's just no point in engaging in further conversation at this point so I won't waste more time than I already have.
one more tidbit, at one other company a few years ago I had a proposal on the table that would of consolidated roughly 25 server racks down to three with new hardware and virtualization. Two of those three racks was going to be the same storage array that was being used by the existing 25 racks. So really it was more like 23 racks down to one (yes a lot of that hardware was pretty old, and most of the racks were not fully populated due to weight or power constraints).
Project didn't get approved before I left the company, it was more of a conceptual thing rather than something that we were seriously planning for.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
