High-speed Storage Area Networks Using a Fibre Channel Arbitrated Loop Interconnect

Fibre Channel Arbitrated Loops (FC-AL) ooer a new approach to realizing high-speed storage interconnection networks. In this paper, we describe a FC-AL topology and FC-AL protocols for storage networks. In particular, we describe channel arbitration, signaling and transmission protocols, and the Fibre Channel mapping protocol for SCSI. A simulation model of FC-AL storage networks is described. Performance results derived from the model are presented and are used to investigate FC-AL storage network performance. Storage area networks (SAN) interconnect computer hosts with storage subsystems through an interconnection network. A SAN requiring a high-performance storage interconnect might consist of several servers connected to several storage subsystems with each subsystem having as many as 72 disks. Each subsystem could have as much as a terabyte storage capacity. A typical application of this system would be to provide servers with 1 high-speed access to disk les. Telecommunications billing and online database transaction processing are speciic examples of le access applications. These applications are characterized by small data transfers of between 2 and 8 kilobytes (KB). The large le caches employed in these applications result in disk access patterns that are essentially random over a large range of disk addresses. Applications of this type require low response times, typically 30 to 50ms, and high IO throughput rates, typically 1000 to 2000 IO/s per host. Generally, these applications do not beneet from subsystem optimization techniques such as read-ahead. Performance is enhanced by the use of concurrent threads for IO processing and load-balancing by the subsystems. A high-performance storage interconnect is required to achieve the low response times and high IO rates of these applications. Other high-performance SAN applications such as stream video require real-time storage access and have very diierent performance requirements. IO response times must not exceed a speciied value; for example, 300 to 500ms are common response time limits for many real-time applications. Real-time SAN applications are characterized by large data transfers. Disk access is primarily for sequential blocks. Performance is improved by striping data over several disks; that is, data sets are partitioned and stored on multiple disks. Then IO requests accessing large amounts of data, typical of these applications, are simultaneously processed by the disks on which the data reside. A video-on-demand application that provides Motion Picture Expert Group version 2 video streams requires a throughput rate of about one MByte/s per stream. Other video applications using motion JPEG images require throughput …