FIBRE CHANNEL and IP SAN INTEGRATION(4)

The maturity and mission-critical deployment of Fibre Channel (FC) in storage area networks (SANs) creates a unique class of multi-terabit networks with demanding throughput, latency, scalability, robustness, and availability requirements. This paper revie

2.3 Critical Factors in SAN DeploymentSAN deployments today range from small fabrics with less than 100 devices to large fabrics with several thousand devices. The following are factors critical to SAN design and deployment: High availability: The impact of down-time and lost of information to business is severe. High availability requirements are quantified to vary from several 9’s, to 99.999%, to no down time. Most highly available fabrics are based on dual-rail redundancy and highly available directors, switches, and gateways. Servers and storage devices may have redundant paths through one fabric or through separate redundant fabrics with no shared single point of failure. Directors and some switches are designed with high-availability features, including fully redundant and hot swappable field-replaceable units (FRUs) and hot software download and activation, meaning that operation may continue through a software upgrade. Robustness and stability: Some FC servers, associated host bus adapters (HBAs) and storage devices are extremely sensitive to frame loss and frame out of order delivery. Error recovery in the SCSI-FCP protocol is based on command and transaction level time-out and retry. Therefore, SCSI-FCP expects very low frame loss rate, since frame loss has significant performance impact. The design of SANs has to account for the following factors: o It is important to limit and reduce FC fabric size in terms of number of switching nodes. The goal is to limit the frequency of fabric initialization, FSPF route computation, and traffic for state notification and name services. o It is critical to ensure there is adequate aggregate bandwidth (fabric-wide and for individual links), to avoid severe and prolonged congestion. FC fabrics use a link-level, credit-based flow control, which is useful for handling short-term, bursty congestion. In FC, it is not common to use active queue management techniques (e.g., based on random early detection) to minimize queue build up. It is typical for a FC switch to discard frames that have been queued for a pre-determined time (e.g., 0.5 to 1.0 second), as part of the stale frame discard policy. As the deployment of multi-speed (1 Gbps, 2 Gbps, 4 Gbps, and 10 Gbps) ramps up, the design of the network and switching architecture becomes more challenging. As the size of network grows, comprehensive congestion management mechanisms become more critical and current link-level flow control may no longer be adequate. Performance: Most FC switches and directors specify best-case frame latency to be less than a few microseconds. But latency grows with loading and can result in effective bandwidth to be significantly less than nominal bandwidth. Measured frame latency at 70% link utilization [3] showed it was 5.2 to 6.5 microseconds for one vendor’s product and 2.6 to 2222.6 microseconds for another vendor’s 104