Throughput-Competitive Admission Control for Continuous Media Databases
Minos N. Garofalakis
Yannis Ioannidis
Banu Özden
Avi Silberschatz
Date published: 
Published In: 
17th Int’l ACM PODS Symposium, Seattle, WA, June 1998, pp. 79-88
Conference Article

Multimedia applications require a guaranteed level of service for accessing Continuous Media (CM) data, such as video and audio. To obtain such guarantees, the database server whom the data is residing must employ an admission control schomo to limit the number of clients that can be served concurrently. We investigate the problem of on-line admission control where the decision on whether to accept or reject a request must bc made without any knowledge about future requests. Employing competitive analysis techniques, we address the problem in its most general form with the following key contributions: (1) we prove a tight upper bound on the competitive ratio of the conventional Work-Conserving (WC) policy, showing that it is within a factor (1+Δ)/(1-ρ) of the optimal clairvoyant strategy that knows the entire request sequence in advance, where A is the ratio of the maximum to minimum request length (that is, time duration), and p is the maximum fraction of the server’s bandwidth that a request can demand; (2) we prove a lower bound of Ω[(log Δ)/(1-ρ)] on the competitive ratio of any deterministic or randomized admission control scheme, demonstrating an exponential gap between greedy and optimal on-line solutions; (3) we propose simple deterministic schemes based on the idea of bandwidth prepartitioning that guarantee competitive ratios within a small constant factor of log Δ (i.e., they are near-optimal) for sufficiently large server bandwidth; (4) we introduce a novel admission control policy that partitions the server bandwidth based on the expected popularities of different request lengths and present a set of preliminary experimental results that demonstrate the benefits of our policy compared to WC. We believe that our results offer new insights to other optimization problems that arise in CM data management, including data placement and load balancing in distributed CM databases.

Related files: 

MaDgIK 2009-2016