Self-similar traffic sources: modeling and real-time resource allocation

Communication networks have to rely on efficient resource allocation schemes to share the network resources (bandwidth, buffer size, etc.) among users offering different types of traffic (eg, voice, video and data). Existing schemes based on self-similar traffic models assume that the network traffic is Gaussian and exhibits long-term memory characteristics only. Certain classes of network traffic (eg, MPEG video traces) are however, non-Gaussian and long-range-dependent. In such cases, resource allocation based on simplified assumptions will be either excessive or fail to provide the specified guarantees on the quality of service (QoS). In an earlier work, we had presented an efficient resource allocation scheme for traffic sources having: (i) Gaussian as well as non-Gaussian (log-normal) distributions; and (ii) exhibiting short-term and/or long-term memory characteristics. In this paper, we assess the real-time performance of our as well as several existing schemes using a Texas Instruments TMS320C6701 DSP. The results show that: (i) although our algorithm has a higher computational load, real-time implementation is still feasible; and (ii) the increased computational load is justified since the proposed algorithm is more reliable in providing QoS guarantees than existing simplified schemes