Roger Haskin. Tiger Shark --- A scalable file system for multimedia. IBM Journal of Research and Development, 42(2):185--197, 1998.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....Many of these scheduling algorithms have also been implemented in commercial and open source operating systems. Streaming media servers implemented at the user level are an example of managing disk resources at the application level [20] while file systems such as SGI XFS [6] and IBM TigerShark [5] implement schedulers that support guaranteed rate I O. 6 Concluding Remarks In this paper, we examined two architectural alternatives, namely native OS support and a middleware, for supporting multimedia applications. Specifically, we examined whether extensions to OS functionality are ....

R. Haskin. Tiger Shark--A Scalable File System for Multimedia. IBM Journal of Research and Development, 42(2):185-- 197, March 1998.

....Many of these scheduling algorithms have also been implemented in commercial and open source operating systems. Streaming media servers implemented at the user level are an example of managing disk resources at the application level [28] while file systems such as SGI XFS [12] and IBM TigerShark [11] implement schedulers that support guaranteed rate I O. As explained in Section 1, despite these numerous research efforts, there has been no systematic study of the tradeoffs of the two approaches; our current work is a step in this direction. 6 Concluding Remarks In this paper, we examined ....

R. Haskin. Tiger Shark--A Scalable File System for Multimedia. IBM Journal of Research and Development, 42(2):185-- 197, March 1998.

....is to migrate data off a node when it leaves the cluster [13] assuming that the node is still accessible but is busy with a sequential job. There are many cluster file systems presented in the literature. GPFS [9] a cluster parallel file system based on Tiger Shark multimedia file system [10], achieves an extreme scalability through its shared disk architecture. All nodes in the cluster have equal access to all disks, across which files are striped. Like other file systems for distributed memory computers, GPFS uses client buffering with a distributed locking mechanism to maintain ....

R. L. Haskin, "Tiger Shark-A Scalable File System for Multimedia," IBM Journal of Research and Development, 42(2), pp. 185-197, March 1998.

....data and employs mechanisms that exploit these semantics to cater to different types of applications, while the latter is typically oblivious of semantics of files and provides a simple best effort service to all applications. Whereas several integrated file systems were designed in the mid 1990s [4], 5] 6] 11] research in integrated file systems has stagnated in the recent past. However, the development and deployment of several new technologies for emerging web based multimedia applications has ne This research was supported in part by an NSF CAREER award CCR 9984030, NSF grants ....

.... architecture employing separate servers [10] While these benefits come at the expense of increased file system complexity, a number of conventional file systems are nevertheless moving towards an integrated architecture by providing native support for data types such as audio and video [4], 5] Most existing file systems, whether integrated or partitioned, are based on the traditional server attached disk architecture. The emergence of storage area networks will require us to rethink the policies and mechanisms employed by existing file systems. As explained in Section I, in a ....

R. Haskin. Tiger Shark--A Scalable File System for Multimedia. IBM Journal of Research and Development, 42(2):185--197, March 1998.

....balancing characteristics. disk 1 disk 2 disk J A 1 C 2 B j A 2 C 3 B 1 A i C 1 B j 1 Object B= B 1 ,B 2 , B j Object A= A 1 ,A 2 , A i Object C= C 1 ,C 2 , C k Figure 1: Objects striped across disks. In the recent past, a great deal of CM server designs, e.g. as in [2, 10, 14], have focused on wide data striping techniques, where each object is striped across all the disks of the system, i.e. disk 1 through disk J, as illustrated in Figure 1. The potential load imbalance is largely due to the skews in data access patterns which, without data striping, could result ....

.... and the communications network infrastructure dictate an upper bound on the number of disks over which that object can be striped, beyond which replication of objects is needed to increase the number of simultaneous users [10] e.g. to the best of our knowledge, in implementations described in [10, 14] striping is performed over (at most) a few tens of homogeneous disks only 1 . Note that, delivery of relatively short CM objects is of use to many applications, including digital libraries and news on demand systems. 3. Due to the continuity constraints, some form of synchronization in ....

[Article contains additional citation context not shown here]

R.L. Haskin. Tiger Shark : A Scalable File System for Multimedia. Technical report, IBM Research, 1996.

....architecture that separates mechanisms from policies. systems and video on demand servers) integrated file systems have received significant attention recently. Several research projects such as Symphony [36] Fellini [20] and Nemesis [32] as well as commercial efforts such as IBM s Tiger Shark [13] and SGI s XFS [16] have resulted in storage servers employing the integrated architecture. 5 CONCLUDING REMARKS Emerging multimedia applications differ from conventional distributed applications in the type of data they store, transmit, and process; and also in the requirements they impose on ....

R. Haskin. Tiger Shark--A Scalable File System for Multimedia. IBM Journal of Research and Development, 42(2):185--197, March 1998.

....the load of the network. This is done by accurately timing one way transmission delays, which can only be accomplished if both the receiver and sender have synchronized clocks. The correctness of timestamp based protocols, used in [6] rely on the synchronization of clocks over the network. Haskin [3] uses synchronized clocks to implement a global deadline I O scheduling policy over a distributed file system. Modern computers typically have two clocks, a hardware clock and a system clock implemented in software. These clocks derive their reference to external time through a quartz oscillator. ....

R. Haskin. Tiger shark-- a scalable file system for multimedia. IBM Journal of Research and Development, 42(2):18--27, 1998.

.... disk storage system itself rather than that of the clients, for example, HP s AutoRAID system [11] There have been several real time storage server projects such as SUNY Stony Brook s SBVS [12] Microsoft s Tiger server [14] Starlight s StarWorks [13] and IBM s Tiger Shark parallel file system [15]. All the above systems took the more traditional network file system architecture rather than the NASD architecture. Some of these enhanced their scalability by deploying a clustered system architecture, but all data transfers had to go through the file servers. Power management by reducing disk ....

Haskin R.L., " Tiger Shark-a scalable file system for multimedia," IBM Journal of Research and Development, vol.42, no.2, p. 185-97, March 1998.

....consideration then is the placement of objects on the nodes of the CM server. As in traditional database systems, data placement on a distributed storage sub system directly affects the load balancing characteristics of that system. In the recent past, a great deal of research work, e.g. as in [2, 20, 3, 8, 10], has focused on wide data striping as a data placement technique for designing continuous media servers. By wide data 1 Here by data placement we mean decision of which object or fraction of an object to place on which disk or disk cluster, i.e. this does not refer to data placement issues ....

....between the degree of striping and the degree of replication is key to the design of a scalable distributed CM server. This is the topic of our paper. 1. 1 Related Work Recently much research has been done on design of continuous media, and specifically video ondemand, storage servers, e.g. as in [2, 20, 3, 8, 10], to a name a few. Much of this work falls into several broad categories 3 , which include: 1) small scale servers, where in most cases all disks are connected to a single node; 2) medium scale LAN based servers, and (3) medium scale (either distributed or not) servers, which employ high speed ....

R.L. Haskin. Tiger Shark : A Scalable File System for Multimedia. Technical report, IBM Research, 1996.

.... disk storage system itself rather than that of the clients, for example, HP s AutoRAID system [11] There have been several real time storage server projects such as SUNY Stony Brook s SBVS [12] Microsoft s Tiger [14] server Starlight s StarWorks [13] and IBM s Tiger Shark parallel file system [15]. All the above systems took the more traditional network file system architecture rather than the NASD architecture. Some of these enhanced their scalability by deploying a clustering system architecture, but all data transfers have to go through the file servers. Phoenix is heavily influenced ....

Haskin, R.L., "Tiger Shark-a scalable file system for multimedia," IBM Journal of Research and Development, vol.42, no.2, p. 185-97, March 1998.

No context found.

Roger Haskin. Tiger Shark --- A scalable file system for multimedia. IBM Journal of Research and Development, 42(2):185--197, 1998.

No context found.

R. Haskin. Tiger shark--a scalable file system for multimedia. IBM Journal of Research and Development, 42(2):185--197, March 1998.

No context found.

Roger L. Haskin. Tiger Shark - a scalable file system for multimedia. IBM Journal of Research and Development, 42(2):185--197, March 1998.

No context found.

R. L. Haskin. Tiger shark -- A scalable file system for multimedia. IBM Journal of Research and Development, 42(2), 1998.

No context found.

R. Haskin. Tiger Shark--A Scalable File System for Multimedia. IBM Journal of Research and Development, 42(2):185-- 197, March 1998.

No context found.

R. Haskin, "Tiger shark--a scalable file system for multimedia," IBM Journal of Research and Development, vol. 42, pp. 185--197, March 1998.

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