<title>Architecture for a large, high-density, fast-access data repository</title>

Abstract

The ability to access very large data sets in seconds rather than hours or days is key to an efficient analysis procedure. We have developed a high performance robotic storage and retrieval system plus an innovative data management technique which provides this capability and is now in operation at user sites. Our EMASS (E-Systems Modular Automated Storage System) achieves this high performance with a combination of state-of-the-art technology. Helical-scan digital recorders, each having a sustained transfer rate of 1 5 megabytes/second and a burst transfer rate of 20 megabytes/second are used to store data, and many recorders can be accessed concurrently to achieve the desired transfer rate. Commercially available, outof-box D2 high density magnetic tape is used for archiving data, and this 19 mm tape is available in three cassette sizes having capacities of 25, 75 or 165 gigabytes, respectively. Cassettes are stored in modular robotic systems whose capacities can be expanded from a few terabytes to thousands of terabytes. EMASS software controls all aspects of data management including commanding the robots to move cassettes between their storage locations and the recorders. Performance is matched to each user' s needs by tailoring the number of recorders, the size and configuration of the robotic archive, and the number of robots for the user's environment.1 . AUTOMATED STORAGE MANAGEMENT EMASS provides storage management at two basic levels, volume serving and file serving. When the user has an existing file management system and only desires to automate the volume retrieval process, we provide volume serving. The request for a volume is sent to a volume server processor which is responsible for executing the commands required to locate, retrieve, and mount the proper tape volume in a tape transport where it is then read from or written to by the user's application. When the user desires the complete system, we supply file serving which incorporates volume serving as a part of the overall process. File serving supports user's requests for data on a file name basis. The user need only request the desired file, and the file serving application handles all activities associated with the file retrieval such as a volume mount request, copying a file from tape to disk, and data recovery processes.The EMASS product line addresses the high performance requirements of both file and volume serving for Concurrent, Convex, and Cray computing environments. EMASS accomplishes this through a combination of advancements in UNIX software applications, magnetic tape media, tape drives, and robotic technology. Figure 1 depicts the architecture in which an EMASS volume server supports main line computing environments. Our volume server software allows the client to control any combination of the three robotic archives shown in Figure 1 , and it also supports manual archiving. Figure 2 shows a high performance file storage server which can support many users. The file server software lets all users store or access their files via...