Properties of the dynamic infrastructure, that the one should keep in mind:

• utilized servers

• multiple virtual machines per server

• high bandwidth links that connect virtual storage between servers

However, the one should consider following point while planning data center infrastructure:

• latency

• performance

• cost-effective scaling

• high-availability

• virtualization

• unified management

• rapid development of new resources

Development of the technology is motivated by the desire to reduce cost while supporting an increasingly dynamic data center.

Architecture should allow to use all the computational and stable power in any way imaginable. Resource allocation happens dynamically, when user is required more space/computation he/she gets it. When user do not need it anymore, the resources are added to the common “pool” and can be used by request.

Ideal system would provide unlimited resources available anytime, to all users with infinite bandwidth and zero latency. However, in real world each port in the network is having upper bandwidth limit, non zero latency. Number of ports is limited as well for the each level of subscription. Good network design would try to minimize the impact of those factors in order to create transparent network for data processing environment.

Network switches are using quite complicated system, and its functionality is controlled within a data center network. Location of this intelligence can be in one of three locations in the network: at network core switches, towards the network edge, or at the attached servers. First option when this control plane functionality is at the network core switches offers economic benefits while decreasing throughput and scalability. Second option provides bigger throughput and scalability. Also it enables better interaction between network and servers. Drawback of this option is the cost of implementation, due to necessity of every switch to have embedded intelligence. Third option offers scalability level between previous two and high throughput for some applications.