Virtualization solutions results in a more streamlined and efficient environment with better resource utilization and lower power usage.

Data centers become more complex daily. The need to improve asset utilization has led architects to create more dynamic environments, which has led to the need for increased automation for administration, provisioning and ongoing management. Implementing virtualization solutions results in a more streamlined and efficient environment with better resource utilization and lower power usage that improves key business metrics such as return on invested capital and operational costs.

Popular types of virtualization in the data center include:

- Server virtualization, which partitions a single physical server to multiple virtual server instances,

- Storage virtualization, which instantiates multiple virtual storage elements on a single physical storage system or aggregates multiple physical storage devices to form a single, logical storage volume, and

- Network virtualization, which can be achieved by partitioning the physical network into multiple virtual LANs (VLANs).

In addition to addressing flexible resource allocation and creating significant cost and power savings, these virtualization technologies create an abstraction layer that can provide easier management and isolate users from physical failures.

Today, all of these virtualization technologies work in silos. What is needed is to place the different virtual and physical resources on a single multi-purpose fabric. That fabric can be partitioned to form systems and application environments that consist of multiple virtual server, storage, and network resources connected in the desired topology.

Fabric virtualization consists of two key elements: I/O virtualization and fabric partitioning. I/O virtualization on the server side partitions the network/channel adapter to multiple virtual network interfaces (NICs) or storage interfaces (HBAs). Fabric partitioning allows the dynamic connection of different virtual I/O (end-points) to each other in the way which best addresses the specific application layout. Moreover, data centers can benefit from fabric virtualization regardless of whether they use physical or virtual server and storage elements.

Data Center Fabrics and Virtualization

Data center fabrics based on InfiniBand with advanced management software provide a virtual backplane for server, storage, and networks that can be re-shaped and re-allocated on demand to meet ever-changing business and application demand.

The InfiniBand fabric acts as the "glue" to unify the different virtualization technologies in the data center by enabling virtualization functionality including hardware-based virtual I/O, virtual local storage, fabric virtualization, application scale-out and management abstraction.

InfiniBand fabrics enable hardware-based virtual I/O where a single/redundant host channel adapter (HCA) and switch port can provide multiple hardware-enforced virtual NICs and virtual storage HBAs to each physical (or even individual) virtual server. Virtual I/O technology allows hardware-based prioritizing and provides the highest performance and isolation, unlike software-based virtual I/O solutions.

Some InfiniBand solutions today enable virtual local storage where remote NAS or SAN (NFS, FC, iSCSI, or InfiniBand) storage can appear as a local disk to the server, eliminating the need for local physical disks (and allowing diskless servers.) It also decouples the storage image from the physical server, simplifying the migration or fail-over of logical servers across different machines, to improve total cost of ownership.

Using InfiniBand, a 20 gigabits/second-based redundant fabric can be used for fabric virtualization with multiple data, storage, and cluster networks layers. Each layer can be fully isolated and used to dynamically allocate switching resources. The high-speed I/O router modules that are integrated in the switches enable seamless connectivity to standard Ethernet and Fibre Channel networks. Redundant fabrics can be formed so that fabric failures can be handled transparently.

For application scale-out, InfiniBand's software drivers offer low latency, RDMA and other protocols so users can form linearly-scaling clusters on top of commodity server platforms. This eliminates the need for high-end servers and mainframes and allow for standardizing on mainstream server platforms for the entire set of applications.