InfiniBand is a new communications standard developed for today's high bandwidth servers and communications devices in applications from small wiring closets to large data centers. InfiniBand allows for higher throughput, lower latency, higher reliability and availability, greater ease of use, and lower cost via better sharing of expensive components.

Developed by a trade association encompassing nearly the entire computing and communications industry, demonstration products have been shown, and production products are expected in the middle of 2001. For more information and complete details on the InfiniBand Trade Association, click on the Infiniband Trade Association logo above, or go to www.infinibandta.org.

Click on the picture below for a larger view.

Banderacom envisions the deployment of IB in a three-phased approach over the next four to six years. Initially, the IB fabric will evolve with the introduction of host channel products residing within stand alone high-end server environments aggregated together with high performance low latency IB switch fabrics. These switch fabrics could possibly have integrated management resident inside the switch fabric or native with in a host environment. Early deployment of IB clustering will most likely be several 1X boxes connected together which will warrant external I/O subsystems to have the need for higher IB bandwidth capability. Banderacom believes the needs for 4X capable targets are important right from the start. Early adopters of target devices will utilize their current FC or RAID technology and route these protocols to IB utilizing standard PCI interfaces as the first step. As IB becomes pervasive, stand alone servers will be replaced with very large I/O chassis that will probably be connected with 4X capable or higher back planes with integrated switching (managed and unmanaged). There will be a need for very large infrastructure switches with 12X capability and very large port counts (20-40 ports). At this juncture, IB LAN target devices will emerge in addition to large storage subsystems needing IB links into the fabric with a minimum of 4X capable support. Also, mid-range stand alone servers will start utilizing 1-4X capable links as the price/performance of IB is in line with these devices. Eventually, all PCI in the high end will be replaced. The IB fabric standardization will be capable of providing connectivity for all I/O subsystems as costs decline and software infrastructure becomes prevalent and easy for IT managers to add and manage.