Weighing Pros and Cons of implementing Physical Server Vs Virtualized Server

So, this is a project that I've inherited from my predecessor.

I'm in charge of installing and maintaining a HP Blade Server in a data center that isn't in the same building as I work in. The server will be in a data center that is offsite and located a few miles away.

I've started the kick-off paperwork to start off the project along with a few meetings however the data center has turned around and strongly recommend that we go virtual instead of using the physical server that we have already bought. Now, I've always been on the data center side of things and realize that pros of going virtual.

However I now work for a bunch of developers and have to have their needs in mind if I were go recommend going virtua, along with making sure that I've though of everything thru if I were to flip the switch. The server will have Windows 2008 R2 fo the OS, however it will be hosting Oracle on Windows along with a custom C# application.

One of the questions that I have is that will the deverlopers have to do anything special or change the way they code if the server is virtualized? How about installing, tuning and performance of Oracle on a virutalized server? How about setting specific hardware from VMWare for the custom C# app and Oracle?

For cost, will there be a cost for using VMWare to host Windows 2008? Are there any other types of costs that I should be thinking of here?

Backups? Are these done on from VMWare or a separate system?

These are some of the questions that come to mind when starting to research this project. Any other input would be awesome as well.

thanks

Find more posts tagged with

Comments

darkerosxx

These are all good questions that you will want answers to, but I think your first action should be to gather requirements for your server environment on your side. For example, is there a reason this hardware was picked, did it meet specific requirements? Do the developers have specific requirements they can share pertaining to their developing environment, processing power, memory needs, storage, etc.? Compare these vs what's being offered.

In the end, will there be a gain for the company to forego the equipment already purchased and go virtual that will eclipse what has already been spent?

ccnxjr

You can potentially use your existing hardware as an alpha/mock-up environment that your developers could use locally, before pushing a VM or code to the cloud.
I believe AWS allows you to import/export VMs from different formats, qemu (RAW), VMDK or VHD.
That can potentially help "right size" your costs once you push a VM to "the cloud".
I suppose that also depends on how many VMs you'll be running, how often will they change.
Or will the application running in the VMs will be changing more often than having to spin up a VM.

That sorta also rests on the development methodology, and how load is distributed.
Is there some sharding or load balancing taking place (and where in the stack is load balancing happening! application vs network vs both).

I'm assuming by "virtualization" you mean using a service such as AWS/Rackspace/Google Compute ?
It's not unreasonable for you build your own VMs and host them in a datacenter.

I've no experience with those types of IaaS , however a few burning questions I would have would be patch policy.
How often do they apply OS patches/updates? Is that going to be your responsibility or theirs ?

Also, moving to "The cloud" what networks are provided that interconnect with your VM instance? Do you have any control over it?
(or maybe choice ISP doesn't really matter for the type of applications being hosted on there, as long as it falls within some SLA)

So many design things!

If you have your own hardware, you can certainly play around with design , until you run out of capacity.
With IaaS you can start off with lower Capex, and scale out depending on your usage.
However, if your not sure on exactly how your going to scale out then this cost needs to be factored into prototyping or sandboxing.

dave330i

Oracle has a very anti-virtualization licensing policy, so you'll need to worry about that.

JockVSJock

dave330i wrote: »

Oracle has a very anti-virtualization licensing policy, so you'll need to worry about that.

You confirmed my suspicions...and...I found the following pdf

http://www.vmware.com/files/pdf/partners/oracle/Oracle_Databases_on_VMware_-_Best_Practices_Guide.pdf

JockVSJock

I've tried to Google around and find articles on say, cost associated with converting from physical server to virtual server.

I only found this one:

Making the choice between virtual and physical servers - TechRepublic

It seems like high transaction servers, like database or mail servers should still rely on phyiscal servers. However the article doesn't give the why behind this.

So I'm leaning heavily towards virutalization because this won't be a high transaction server, however there will be an Oracle Database involved.

A big concern is the cost. Will my department be charged for utilizing the ESX servers? I'm not so worried about the cost of the OS, because I'm pretty sure we have a site license.

ccnxjr

Not sure about the others, but there's also this:

NTP server was not designed to run inside of a virtual machine. It requires a high resolution system clock, with response times to clock interrupts that are serviced with a high level of accuracy. NTP client is ok to run in some virtualization solutions.

KnownOsIssues < Support < NTP

That is, of course, if you plan on running your own NTP server.

If you do have the manpower, and not wholly dependent on vendor support you may also want to consider KVM/qemu.
Seems to play well with Windows Servers as guests.

http://www.linux-kvm.org/page/Guest_Support_Status

There are quite a few consultants/3rd parties that offer support if needed.

dave330i

A typical physical server uses ~10% of physical resources. If you deploy physically, you're limited to 16 servers in (assuming half height blades) and no redundancy. If you virtualize it, you can deploy 120 servers (8*15) and have 1 blade as redundancy (assuming you have the storage). Obviously this is a very crude calculations. The actual consolidation depends on the actual workload, HW purchased, oversubscription policy, etc.