Leana Golubchik, John C.-S. Lui, and Richard R. Muntz. Reducing I/O Demand in Video-on-Demand Storage Servers. In Proceedings of ACM SIGMETRICS, pages 25-36, 1995.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....reduce the bandwidth for handling requests for a hot video over di erent times. These approaches take advantage of the multicasting architecture (which allows the server to multicast a single video stream to many clients at the same time) to merge streams initiated at di erent times. Piggybacking [1, 14, 15, 19] is based on the fact that a small deviation of the play rate is not perceived by viewers. The server can deliver a video stream in two rates, normal and fast. The fast play rate is achieved via sampling techniques instead of increasing the server bandwidth. A stream X can run in the fast rate to ....

....of the optimal cost in the worst case. That is, with respect to the conventional competitive analysis, this is a 3 competitive on line scheduler. As a matter of fact, no competitive piggybacking scheduler has been known previously (though there are many heuristics proposed evaluated empirically [1, 14, 15, 19]) For skimming, Bar Noy and Ladner [4] are the rst to study worst case performance based on competitive (b) a) c) playback 1 2 time 0 1 2 time time 2 0 playback playback playback playback 1 0 client B client B client A client B client A bu er bu er bu er bu er bu er stream X ....

L. Golubchik, J. C. S. Lui, and R. R. Muntz. Reducing I/O demand in video-on-demand storage servers. In Proc. ACM Sigmetrics, pages 25-36, 1995.

....until they can be merged with other requests for the same clip. The merged streams then form one physical stream from the disk and consume only one set of buffers. Only on the network will the stream split at some point for delivery to the individual display station. adaptive piggybacking [GLM95] this approach adjusts the streams for the same video to go either slower or faster by a few percent, such that it is imperceptible to the viewer, and the streams eventually merge and form one physical stream from the disks. The focus of this study is on buffering. It ignores the other two ....

L. Golubchik, J. C.-S Lui, and R. R. Muntz. Reducing i/o demand in video-on-demand storage servers. In Proceedings of the ACM SIGMETRICS and Performance '95, pages 25--36, 1995. 23

....protocols under an environment where request rates may be highly variable, therefore VoD service has to be highly adaptive, optimizing its server bandwidth usage to the level of request traffic. Existing VoD schemes fall in two categories: closed loop schemes [11] 10] 6] 15] 2] 7] 8] [16] and openloop schemes [3] 26] 1] 18] 4] In most closed loop schemes, the server allocates channels and schedules transmission of video streams based on client requests using batching, or patching techniques. In batching, requests for the same video clip are delayed for a certain amount of ....

L. Golubchik, L. C.-S. Liu, and R.R. Muntz. Reducing (i/o) demand in video-on-demand storage servers. In Proc. of ACM SIGMETRICS, pages 25--36, 1995.

....of VoD request scheduling protocols under an environment where request rates may be highly variable. In this environment, VoD service has to be highly adaptive, optimizing its server bandwidth usage to the level of request trac. Existing VoD schemes fall in two categories: closed loop schemes [12, 11, 7, 19, 2, 8, 9, 20] and open loop schemes [3, 31, 1, 22, 4] In most closed loop schemes, the server allocates channels and schedules transmission of video streams based on client requests using batching, or patching techniques. In batching, requests for the same video clip are delayed for a certain amount of time ....

L. Golubchik, L. C.-S. Liu, and R.R. Muntz. Reducing (i/o) demand in video-on-demand storage servers. In Proc. of ACM SIGMETRICS, pages 25-36, 1995. 24

....provided: The naive approach of dedicating a separate unicast stream for each client request can quickly deplete server resources. Recent work has focussed on leveraging the multipoint data distribution facility of the Internet called multicast [20] to reduce the server and network load [18] [33], 1] 2] 76] 11] 38] 31] 39] 40] 42] 54] 9] 25] 32] 37] 46] 55] 64] 67] 26] 27] For example, a simple scheme is to aggregate client requests for the same media file that are separated by a small time interval into one group, and serve them by a single ....

....of each copy are staggered evenly across the channels. This scheme requires many channels to achieve low client start up delays. For example, a two hour movie would require 24 channels (each at the media playback rate) to guarantee a maximum client waiting time of 5 minutes. Adaptive Piggybacking [33], 2] allows the media playback rate of on going streams to be altered such that different streams can merge when they reach the same position in the media object. For example, consider arrival of a new request for a media file for which there exists an on going stream. In response to this new ....

L. Golubchik, J. C. S. Lui, and R. Muntz. Reducing I/O Demand in Video-on-Demand Storage Servers. In Proc. ACM Sigmetrics '95, Ottawa, Canada, May 1995.

....and clients simply join such multicast channels. Since a server is not interactively responding to request arrivals, clients may have to wait before service could start. Both closed loop and open loop approaches have been well studied, including the early batching [13, 15] piggybacking [23, 24, 2, 30], and stream tapping patching [10, 27, 9, 22, 41, 8] techniques, as well as the more recent stream merging [17, 18, 19, 7, 14, 33, 6, 44] and broadcasting protocols [43, 3, 28, 16, 21, 29, 36, 37, 38, 40, 39, 26, 42] Two particular techniques stream merging and periodic broadcasting have been ....

L. Golubchik, J. C. S. Liu, and R. R. Muntz. Reducing I/O demand in video-on-demand storage servers. In Proceedings of SIGMETRICS, 1995.

....and clients simply join such multicast channels. Since a server is not interactively responding to request arrivals, clients may have to wait before service could start. Both closed loop and open loop approaches have been well studied, including the early batching [13, 15] piggybacking [23, 24, 2, 30], and stream tapping patching [10, 27, 9, 22, 41, 8] techniques, as well as the more recent stream merging [17, 18, 19, 7, 14, 33, 6, 44] and broadcasting protocols [43, 3, 28, 16, 21, 29, 36, 37, 38, 40, 39, 26, 42] Two particular techniques stream merging and periodic broadcasting have been ....

L. Golubchik, J. C. S. Liu, and R. R. Muntz. Reducing I/O demand in video-on-demand storage servers. In Proceedings of SIGMETRICS, 1995.

....protocols under an environment where request rates may be highly variable, therefore VoD service has to be highly adaptive, optimizing its server bandwidth usage to the level of request traffic. Existing VoD schemes fall in two categories: closed loop schemes [11] 10] 6] 15] 2] 7] 8] [16] and openloop schemes [3] 26] 1] 18] 4] In most closed loop schemes, the server allocates channels and schedules transmission of video streams based on client requests using batching, or patching techniques. In batching, requests for the same video clip are delayed for a certain amount of ....

L. Golubchik, L. C.-S. Liu, and R.R. Muntz. Reducing (i/o) demand in video-on-demand storage servers. In Proc. of ACM SIGMETRICS, pages 25--36, 1995.

....6 Piggy backing Piggy backing wil zeggen dat 2 batches die niet al.. te ver na elkaar de zelfde film beginnen, naar elkaar toe worden geduwd door de eerdere batch iets langzamer de film te laten kijken, en de latere iets sneller. Zolang dit 2 a 3 procent blijft, merken de gebruikers dit niet [9]; met een extra filter is zelfs 8 procent haalbaar. 7 Chaining Chaining [16, 11] wil zeggen dat in een multicast een client niet al..leen een stream ontvangt, maar deze ook zelf weer doorstuurt, na deze gecachet [11] of gebufferd [16] te hebben. Met het cachen wordt de server load verminderd. Met ....

Leana Golubchik, John C.S. Lui, and Richard Muntz. Reducing i/o demand in video-on-demand storage servers.

....that multiplexes all the resources available at a server the storage space, the disk bandwidth, and the buffer cache among multiple application classes and data types. Since techniques for designing file systems that are optimized for a single application class are well understood [9, 13, 22, 23], partitioned file systems are easy to design and implement. The design of integrated file systems, on the other hand, is challenging due to the heterogeneous performance requirements of data types and applications. However, since such file systems share all resources among all application ....

L. Golubchik, J. C. S. Lui, and R. R. Muntz. Reducing I/O Demand in Video-On-Demand Storage Servers. In Proceedings of SIGMETRICS '95, Ottawa, Canada, May 1995.

....this environment, VoD service has to be highly adaptive, optimizing its server bandwidth usage to the level of request traffic. Current state of the art, however, has not produced an efficient adaptive protocol for this environment. Existing VoD schemes fall in two categories: closed loop schemes [6, 13, 2, 7, 8, 14] and open loop schemes 1 [3, 25, 1, 16, 4] In most closed loop schemes, the server allocates channels and schedules transmission of video streams based on client requests using batching, or patching techniques. In batching, requests for the same video clip are delayed for a certain amount of ....

L. Golubchik, L. C.-S. Liu, and R.R. Muntz. Reducing (i/o) demand in video-on-demand storage servers. In Proc. of ACM SIGMETRICS, pages 25--36, 1995.

....require additional resources, it may result in longer waiting time, which is unacceptable in some circumstances. Fox example, when performing the VCR functions, clients usually expect to get the response immediately, and even a few seconds waiting seems to be undesirable. adaptive piggybacking [7]: in this approach streams for the same video are adjusted to go slower or faster by a few percent, such that it is imperceptible to the viewer, and the streams eventually merge andform one physical stream from the disks. Batching and adaptive piggybacking are orthogonal to buffer sharing, and we ....

L. Golubchik, J. Lui, and R. Muntz. Reducing I/O Demand in Video-On-Demand Storage Servers. In Proceedings of the ACM SIGMETRICS, pages 25--36, 1995.

....by the disk layout and scheduling to serve multiple streams simultaneously, improved request scheduling can reduce customer waiting times. This in turn results in better service and or lower customer loss. Several previous papers have investigated request scheduling policies for multimedia servers [11, 15, 7, 5, 2, 1]. One difficulty in evaluating the algorithms is that the characteristics of multimedia server requests are currently unknown. Only a few prototype systems are in existence. Workloads on these systems have not been reported and may not be representative of future workloads anyway. Nevertheless, ....

GOLUBCHIK, L., LUI, J. C., AND MUNTZ, R. Reducing I/O demand in video-on-demand storage servers. In Proceedings of the 1995 ACM SIGMETRICS Conference on Measurement and Modeling of Computer Systems (Ottawa, Ontario, Canada, May 1995), pp. 25--36.

....sequence header and in the header of the first GOP. This information can be used to allocate memory for the frame buffer during the initial signalling phase. 6 Striping and Disk Scheduling An extensive amount of research has been done on the design of multimedia storage systems (see, for example, [17, 18, 23, 24, 34, 39, 12]) In a typical VOD system, each compressed movie is segmented into blocks, which are distributed over multiple storage devices (or disks) This process, known as striping, is used to increase the disk retrieval throughput. Striping is being used in many storage systems, such as disk arrays and ....

L. Golubchik, J. Lui, and R. Muntz. Reducing I/O demand in video-on-demand storage servers. In Proccedings of the ACM Sigmetrics/Performance '95 Conference, pages 25--36, 1995.

....architecture that multiplexes all the resources available at a server the storage space, the disk bandwidth, and the buffer cache among multiple application classes and data types. Since techniques for designing file systems that are optimized for a single application class are well understood [13, 9, 22, 23], partitioned file systems are easy to design and implement. The design of integrated file systems, on the other hand, is challenging due to the heterogeneous performance requirements of data types and applications. However, since such file systems share all resources among all application ....

L. Golubchik, J. C. S. Lui, and R. R. Muntz. Reducing I/O Demand in Video-On-Demand Storage Servers. In Proceedings of SIGMETRICS '95, Ottawa, Canada, May 1995.

....or very close by an SS. The SCF SCA components relieve the SS of the timing analysis of stream data and effectively provides a rate control mechanism between the SSs and the clients. For example, the SCA can be used to adjust the rate of each stream in order to reduce the I O demand in the SSs[2]. It can also provide timing information for the SS to achieve initial synchronisations between related streams for playing back, though it is preferred that stringent synchronisation is performed near the sink by a stream group agent method[3] In this case, the SCA can be used to adjust ....

L. Golubchik, J. C. S. Lui, and R. Muntz. Reducing i/o demand in video-on-demand storage servers. In SIGMETRICS'95, Ottawa, Ontario, Canada, 1995. ACM.

....during the display, or both. Section 3 quantifies trade off associated with these techniques using a simulation study. Brief conclusions are provided in Section 4. 1. 1 Related Work A number of studies have investigated techniques to ensure a continuous display of stream based presentations [AH91, RVR92, TPBG93, GR93, RW94, BGMJ94, GLM95, MKT95, DS95]. These studies conceptualize a presentation as a file that is read sequentially at a pre specified rate. Alternatively, a structured presentation consists of a collection of objects that are referenced and displayed based on a pre specified schedule. A single object might be referenced at ....

L. Golubchik, J.C.S. Lui, and R. Muntz. Reducing i/o demand in video-on demand storage servers. In Proceedings of ACM Sigmetrics, May 1995.

....utilization a system has, the lower is its response time for servicing requests. There are several ways to batch requests into a single I O stream. Due to space limitations we do not discuss batching policies here and in the remainder of the paper assume that the batching by timeout policy (see [8]) is used, 8 In either case we assume that when frames are inserted, the additional frames are some interpolation of existing frames (not simply duplicates) Similarly, when a frame is deleted, the preceding and succeeding frames are altered to reduce the abruptness of the change (e.g. each ....

....for the fairly small interarrival time of 0:5 minutes, to 20:92 , for the fairly large interarrival time of 10 minutes. The results are qualitatively similar, when batching is used in conjunction with adaptive piggybacking 15 ; however, due to lack of space we do not illustrate them here (see [8]) 15 Clearly, as the batching interval increases, the range of workloads over which these policies exhibit significantly different behavior decreases. Although a greater reduction in I O demand is achieved by the greedy policy, as compared to the odd even policy, allowing more than a single ....

Leana Golubchik, John C.-S. Lui, and Richard R. Muntz. Reducing I/O Demand in Video-On-Demand Storage Servers. Technical Report CSD-940037, UCLA, October 1994.

....2. buffering [11, 10, 3] i.e. closing the temporal gaps between successive requests through the use of buffer space, i.e. holding data read for a leading stream and servicing trailing requests out of the buffer rather than by issuing another I O stream, and 3. adaptive piggybacking [8, 7, 12, 1], i.e. adjusting consumption rates of requests in progress (for the same object) until their corresponding I O streams can be merged into one 1 . Though there may be many ways of handling client requests, an optimal way is one that minimizes some objective function 2 . Once the basic model ....

....members of a sync class is one such property. So is a weighted average which takes into account priorities of different clients (if a priority classification is defined) 1 Due to lack of space we will not discuss the mechanism for adjusting consumption rates but refer the interested reader to [8]. 2 It is well known in operations research [9] that all maximization problems are expressible as minimization problems. As already mentioned, our goal is to reduce the overall load on the storage subsystem, and hence the motivation to group users requests into sync classes. That is, all ....

[Article contains additional citation context not shown here]

L. Golubchik, J. C.-S. Lui, and R. R. Muntz. Reducing I/O Demand in Video-On-Demand Storage Servers. In Proceedings of the ACM SIGMETRICS and Performance, pages 25--36, May 1995.

No context found.

Leana Golubchik, John C.-S. Lui, and Richard R. Muntz. Reducing I/O Demand in Video-on-Demand Storage Servers. In Proceedings of ACM SIGMETRICS, pages 25-36, 1995.

No context found.

L. Golubchik, J.C.S. Lui, and R. Muntz. Reducing I/O demand in Video-On-Demand storage servers. In Proc. 1995.

No context found.

L. Golubchik, J. C. S. Lui, and R. Muntz, \Reducing I/O demand in video-on-demand storage servers", Proc. ACM SIGMETRICS, pp.25-36, Ottawa, Canada, May 1995.

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