| R. Sandberg, D. Goldberg, S. Kleiman, D. Walsh, and B. Lyon. Design and implementation of the sun network lesystem. 1985. |
....We believe that our system is the rst to combine these characteristics; however, many previous storage systems have provided one or more of them. In this section we outline the related contributions from these projects. 2. 1 File Systems Early distributed le systems such as AFS [32] NFS [29], and Sprite [24] allowed multiple le servers to cooperatively export a partitioned namespace. A second generation, including Amoeba [23] Coda [31] Echo [8] Ficus [25] Frangipani [35] and Harp [21] replicated les across servers to improve availability. This server only replication is ....
R. Sandberg, D. Goldberg, S. Kleiman, D. Walsh, and B. Lyon. Design and implementation of the sun network lesystem. 1985.
....was originally designed as an environment for single process jobs. As a result, most distributed le systems have little or no provision for coordinating shared le access by multiple cooperating processes. In fact, the semantics of concurrent access are sometimes left unde ned. For example, NFS [41] may produce inconsistent results when a le is write shared by a number of processes. Those le systems that do provide Unix writewrite and read write sharing semantics among concurrently executing processes on di erent nodes implement this sharing through a costly cache coherence protocol [33] ....
R. Sandberg, D. Goldberg, S. Kleiman, D. Walsh, and B. Lyon, \Design and implementation of the Sun network lesystem". In Proc. Summer USENIX Technical Conf., pp. 119-130, Jun 1985.
....little structure. Furthermore, it is inherently tied to a speci c processor architecture and does not tolerate failures, making it an unsuitable API for distribution outside of tightly managed installations. Files, which are clearly one of the most popular abstractions found in distributed systems [2, 8, 36, 37, 50, 56, 63, 67, 79, 85], represent a higher level of abstraction and o er a much larger granularity than memory. But, they also provide little structure, thus raising the issue of how to manage concurrent modi cation of the same le without relying on some form of external synchronization. As a result, les are an ....
Russel Sandberg, David Goldberg, Steve Kleiman, Dan Walsh, and Bob Lyon. Design and implementation of the Sun network lesystem. In Proceedings of 1985 Summer USENIX Conference, pages 119-130, June 1985.
....3.1.1 Storage Bricks As previously mentioned, each storage node in the cluster maintains an independent, self contained storage brick . These bricks contain single node hash tables that can be accessed via an RPC like interface, similar in nature to networked le systems such as Sun s NFS [67]. Signi cant design and engineering e ort went into the implementation of these bricks, since we wanted to be able to run bricks under di erent operating systems, thread models, and network primitives and communications media. Figure 4 illustrates the architecture of one of these bricks. ....
....active service framework [1] All of this research is complementary to our planned research; we can borrow relevant techniques and mechanisms, and make our SDDS implementations available to the cluster platforms. There have been volumes of research results in networked and distributed le systems [6, 72, 77, 23, 66, 12, 67]. Many of the techniques used in distributed le systems such as striping, structuring of les to induce long sequential disk operations, and the organization of serverless le systems are directly applicable to our research. File systems, however, are much more complex and support far richer ....
R. Sandberg, D. Goldberg, S. Kleiman, D. Walsh, , and B. Lyon. Design and implementation of the sun network lesystem. In Proceedings of the USENIX 1985 Summer Conference, El Cerrito, CA, USA, June 1985.
Online articles have much greater impact More about CiteSeer.IST Add search form to your site Submit documents Feedback
CiteSeer.IST - Copyright Penn State and NEC