Virtualization is a technology which allows one physical server to run many virtual servers. By virtual server, I mean a software container that contains a full operating system and set of software, just like a traditional server would. Each virtual server “believes” that it is running on physical hardware, and thus any software you would run normally on real physical machines will work without modification in a virtual server. Furthermore, each virtual server appears to the outside world – to clients which connect to them – to be traditional physical machines with dedicated hardware. The operating system of the physical server provides hardware interfaces to the virtual servers via simulation in softwre: CPUs, hard drives, network interfaces, displays, CD roms, etc.. The physical server translates I/O to these virtual devices to the I/O to the requisite physical devices, if necessary. The resources allocated to virtual machines can be changed nearly on-demand (disk, memory, network interfaces); most operating systems need merely to be rebooted in order to see the new resources. Limiting factors in how many virtual machines an individual virtualization server can run are simply how much memory, disk space and network bandwidth the virtualization server has available to it, and to how many resources each virtual machine consumes. A very basic virtualization server could host eight to twenty virtualized machines.

Furthermore, most implementations of virtualization technology allow a cluster of physical virtualization servers to seamlessly exchange running virtual machines.

Virtualization is a paradigmatic shift in server resource provisioning for several reasons. First, once an organization has built a cluster of virtualization servers, new server provisioning can happen without an additional capital outlay, without having to have machines shipped, unpacked, tested, racked and provisioned with networking, and with substantially less power consumption. Thus the barrier to entry for new services and servers, and the time to bring a server up from identification of a need to a working system can be dramatically be reduced. Operating costs for datacenters also become dramatically reduced. Secondly, the on-demand provisioning characteristic of virtualization lowers the risk of implementation – if a virtualized server doesn’t work out, we simply delete it. If we didn’t allocate enough RAM or disk to it, we can add more; or if we started with too much RAM or disk, we can reduce it There is no sunk cost (aside from staff time), and no depreciating, aging physical hardware to dispose of . Thirdly, once a cluster of virtualization servers has been built, organizations can experience improved uptime due to being able to move virtual servers from physical server to physical server without impact to the users; this allows us to lower the cost of hardware maintenance and outages.

Among the drawbacks of virtualization are that it can be quite expensive to set up a virtualization cluster. Virtualization machines should be very powerful, with a large amount of computational resources and RAM. They should be SAN backed to enable quick recovery from hardware faults and to enable seamless moving of virtual servers. Managing virtualization servers and virtualized machines requires a new and somewhat different and deeper set of skills than managing individual physical servers, because the environment that virtual servers run in is significantly more complicated, technology-wise, than is that which physical servers run in. Thus we can have fault conditions and problems that we’ve never experienced before. Furthermore, not every service can be virtualized: services which need direct hardware access or very high resource allocation are not good candidates for virtualization. High performance database servers and graphics render farms are good examples of this.