Yeonseung Ryu

A hybrid buffer cache management scheme for VOD server

In this paper, we present a novel buffer cache management scheme for video-on-demand servers called hybrid buffer cache management (HBM). We carefully argue that multimedia workload may exhibit characteristics other than sequential access pattern. If the access pattern is not sequential, the legacy buffer cache schemes that are grounded on the assumption that the applications sequentially access the multimedia file may not work properly. The HBM automatically detects the reference patterns of each file and intelligently switches between different buffer cache management schemes on per-file basis. According to our experiment, the HBM exhibits better buffer cache hit ratio than interval based caching or LRU, especially when the workload exhibits not only sequential but also looping reference property.

Handling sporadic tasks in multimedia file system

Handling mixed workload in streaming server becomes important issue as integrated file system gets momentum as the choice for next generation file system. In this article, we present novel approach of handling sporadically arriving non-playback related disk request while minimizing its interference with the timely retrieval of data blocks for ongoing playbacks. The main idea of our approach is to extend the length of period in period based disk scheduling in multi-media streaming and subsequently a certain fraction of each period can be set aside for handling unexpected I/O request. We develop the analytical model to establish the relationship between the length of period, request arrival rate, request service time, and P(jitter). This model is used to precisely compute the length of period and the respective buffer size. We present the result of simulation based experiment.

Intelligent Buffer Cache Management in Multimedia Data Retrieval

In this paper, we present an intelligent buffer cache management algorithm in multimedia data retrieval called Adaptive Buffer cache Management (ABM). The proposed ABM scheme automatically detects the reference pattern of each file and intelligently switches between different buffer cache management schemes on per-file basis. According to our simulation based experiment, the ABM scheme yields better buffer cache miss ratio than the legacy buffer cache schemes such as LRU or interval based scheme. The ABM scheme manifests itself when the workload exhibits not only sequential but also looping reference patterns.

ABM: Looping Reference-Aware Cache Management Scheme for Media-on-Demand Server

Legacy buffer cache management schemes for multimedia server are grounded at the assumption that the application sequentially accesses the multimedia file. However, user access pattern may not be sequential in some circumstances, for example, in distance learning application, where the user may exploit the VCR-like function(rewind and play) of the system and accesses the particular segments of video repeatedly in the middle of sequential playback. Such a looping reference can cause a significant performance degradation of interval-basedcac hing algorithms. Andth us an appropriate buffer cache management scheme is required in order to deliver desirable performance even under the workload that exhibits looping reference behavior. We propose Adaptive Buffer cache Management(ABM) scheme which intelligently adapts to the file access characteristics. For each opened file, ABM applies either the LRU replacement or the interval-based caching depending on the Looping Reference Indicator, which indicates that how strong temporally localized access pattern is. According to our experiment, ABM exhibits better buffer cache miss ratio than interval-basedcac hing or LRU, especially when the workload exhibits not only sequential but also looping reference property.

Considering Non-real-time Traffic in Real-Time Packet Scheduling

Real-time packet schedulers based on EDF policy have been extensively studied to support end-to-end bounded delay for real-time traffic. Since non-real-time traffic could hurt the schedulability of real-time traffic, we need a real-time packet scheduling method considering non-real-time traffic. In this paper, we propose a new admission control algorithm for EDF-based packet scheduling scheme. Proposed scheme consists of two packet schedulers: an EDF scheduler servicing real-time traffic and a Periodic Server (PS) scheduler servicing non-real-time traffic periodically. Proposed scheme provides good performance to non-real-time flows while guaranteeing the schedulability of real-time flows.

Providing Multimedia Traffic with Predictability

Real-time packet schedulers have been studied extensively to support end-to-end bounded delay for multimedia traffic. Even though real-time flows are guaranteed by admission control scheme in the packet scheduler, their schedulability can be violated by arrival of non-real-time traffic. In this paper, we propose a new packet scheduler scheme to provide real-time traffic with predictability. Proposed scheme also provides non-real-time traffic with fast service time. We present the schedulability condition considering two packet schedulers and give the admission control algorithm.

Using admission control algorithm in resource reservation scheme for enhancing predictability

Real-time packet schedulers have been studied extensively to support end-to-end bounded delay for multimedia traffics. Since non real-time traffic could hurt the predictability of real-time traffic that are already guaranteed, we need a packet scheduler mechanism to support both real-time traffic and non real-time traffic. In this paper, we propose a new admission control algorithm considering real-time trafic and non real-time trafic in an integrated way. In proposed scheme, there are two packet schedulers: EDF scheduler for real-time traffic and a periodic scheduler for non real-time traffic. Proposed scheme provides non real-time traffic with fast service time while not violating schedulability of realtime traffics that are already guaranteed.

Considering User's Access Pattern in Multimedia File Systems

Legacy buffer cache management schemes for multimedia server are grounded at the assumption that the application sequentially accesses the multimedia file. However, user access pattern may not be sequential in some circumstances, for example, in distance learning application, where the user may exploit the VCR-like function(rewind and play) of the system and accesses the particular segments of video repeatedly in the middle of sequential playback. Such a looping reference can cause a significant performance degradation of interval-based caching algorithms. And thus an appropriate buffer cache management scheme is required in order to deliver desirable performance even under the workload that exhibits looping reference behavior. We propose Adaptive Buffer cache Management(ABM) scheme which intelligently adapts to the file access characteristics. For each opened file, ABM applies either the LRU replacement or the interval-based caching depending on the Looping Reference Indicator, which indicates that how strong temporally localized access pattern is. According to our experiment, ABM exhibits better buffer cache miss ratio than interval-based caching or LRU, especially when the workload exhibits not only sequential but also looping reference property.