SASW@MIT-MC.ARPA ("Steven A. Swernofsky") (12/07/85)
MSG: *MSG 4790 Date: 12/06/85 11:38:31 From: LYALL at MIT-XX.ARPA Re: Seminar Received: from MIT-XX.ARPA by MIT-MC.ARPA 6 Dec 85 11:38:18 EST Date: Fri 6 Dec 85 11:34:59-EST From: Neena Lyall <LYALL@MIT-XX.ARPA> Subject: Seminar To: *mac@MIT-MC.ARPA cc: lyall@MIT-XX.ARPA Message-ID: <12164973450.26.LYALL@MIT-XX.ARPA> Wednesday, December 11, 1985 2:15 Refreshments 2:30 Lecture Room: NE43-512A A DISTRIBUTED FILE SYSTEM FOR A NETWORK OF WORKSTATIONS Paul Leach and Nathaniel Mishkin (Apollo Computer) The Apollo workstation supports a distributed file system based on the concept of a "single-level store." Files are "objects" identified by 64 bit IDs that are unique across all Apollo file systems. A process accesses an object's data by mapping the object into the process's address space and making normal memory references. The distinction between local and remote objects is known only to the pager, which loads objects' pages on demand. We will discuss several aspects of this architecture that we feel are crucial to its success: a distributed lock manager, caching of object data, a network-wide registry of users, and a distributed, replicated naming database that is used to locate nodes based on name. Conventional stream I/O (open, close, read, write, etc.) is provided by "Streams." Streams can be used to access objects in the single-level store and other objects that support the concept of stream I/O (e.g. tape drives, terminal) I/O lines. Streams is object-oriented and extensible---the behavior of Streams depends on the type of the object. Users can define new types of objects and write managers that implement the stream operations for the new type. We will discuss the the principles behind Streams' extensibility and describe how we see it being used as the mechanism for supporting file system connections in a heterogeneous environment. The Apollo distributed file system has been implemented on a 680x0 microprocessor-based workstation. Workstations are connected by a 12MB token ring; rings can be connected by gateways to form an internetwork. All workstations in the internet have equal access to the distributed file system. The internet at Apollo Computer has over 1000 workstations. Host: Professor Richard E. Zippel
SASW@MC.LCS.MIT.EDU ("Steven A. Swernofsky") (02/22/86)
Date: 20 Feb 1986 1204-EST From: ALR at XX.LCS.MIT.EDU Subject: SEMINAR SEMINAR DATE: MONDAY, FEBRUARY 24, 1986 TIME: REFRESHMENTS AT 3:00 PM TALK AT 3:15 PM PLACE: ROOM NE43-512A THE ANDREW FILE SYSTEM: PRINCIPLES AND DESIGN M. SATYANARAYANAN Carnegie-Mellon University ABSTRACT Andrew is a computing and communication system consisting of thousands of personal computer workstations linked by a location-transparent distributed file system. The design of Andrew is a synthesis of the best features of personal computing and timesharing. It combines the rich user-machine interface and predictable performance characteristic of personal computing with the ease of information sharing typical of timesharing systems. This system is being built at Carnegie-Mellon University in collaboration with IBM. In the talk, I will focus on the design and rationale of the distributed file system for Andrew. While scale has been the dominant design influence, careful attention has also been paid to the goals of location transparency, user mobility and compatibility with existing operating system interfaces. Security is an important design consideration and the mechanisms for it do not assume that the workstations or the network are secure. Caching of entire files at workstations is a key element in this design. In the latter part of the talk, I will discuss our usage experience with an initial prototype and the conclusions which led to the current implementation. A configuration of about 150 workstations is in use by more than 500 users at the present time. These numbers are expected to grow by an order of magnitude over the next two years. HOST: PROF. LISKOV