i am quite curious as to why you would run 5 or 6 virtual machines, one for DHCP, one for DNS, one for NFS ect ect ect instead of a single server that has all those server applications in a single install of linux. Doesn't all the overhead of the linux kernel running on 6 virtual machines give you worse performance?

thanks
-- Jezzirolk

- Security/Compliance requirements: - Ease of maintenance

"Doesn't all the overhead of the linux kernel running on 6 virtual machines give you worse performance?" -
if your servers specs make this a problem, you should upgrade your servers or run fewer machines per node. Also for KVM-based virtualization we have KSM and for OpenVZ/LXC this does not apply as there is a single kernel.

I would have to add the following reasons to run multiple machines running single services

-- Network Design testing
with the right hardware node you can create a complete office environment for testing network designs to ensure flow of traffic and QoS is optimized prior to implimentation

-- Platform stability testing
This allows you to test the latest patches and updates prior to implementing in production environment

-- Creation of Backup Medium
makes it much easier to create and maintain backups of entire servers

-- Hardware node testing
allows you to test the hardware fully to ensure that you have a stable kernel and stable hardware for the security of knowing your hardware is not just gonna fail as you have fully tested in development phase

I am sure there are other reasons why you should implement this type of topology. I have slowly started moving this way with my personal network which allows me to swap out hardware much easier as I don't have to take down all my services just to upgrade the RAM on my Web Server.

--

Virtualization may bring complexity to your setup rather than all service in one OS.
But one basic thing is isolation.
You may also refer basic benefits http://www.itworld.com/nls_windowsserver050411

Hi Jezzirolk,

I had the same question in the beginning. And in fact, I even started configuring all in one machine. But, as time went on, as others mentioned, I found it very difficult to maintain the server. Moreover, when you change/alter/upgrade physical server, transferring VMs comes very handy. You do not have to go through the tedious process of configuring a server from scratch (in your case you should re-configure all the services). Problem in one service affected all the others ie. downtime for maintenance of one service means downtime for all. I do not remember any particular example to point out straight away. If you are designing server for a very busy environment, you better isolate your servers. OpenVZ is one good solution, if you have low spec hardware.

Cheers

Here we have a three server architecture, web server, API server, database server. In our production environment these are separate physical machines, however for development and testing we don't need that level of hardware so by using VMs for dev and testing we can run these separate instances (configured the same as "production") on a single server saving money and electricity without compromising performance. Also means we need less rack space at the office

For a vast majority of applications even "entry level" server hardware is more than capable of running multiple VMs to allow a split of services.