Abstract
Server performance is a crucial factor for many distributed, clientserver based multimedia applications. Popular forms of multimedia data, such as audio-visual data, must be delivered in continuous, high-volume streams. Hence, requirements for multimedia server architecture are complex. Parallel architectures have been devised to overcome the limitations of traditional servers; however, they can be expensive due to custom hardware and are thus not widely deployed. This paper pursues a different approach: by isolating streaming devices along with shared buffer memory in autonomous streaming modules, we can improve the scalability and performance of a traditional server architecture. In contrast to parallel servers, streaming modules can be implemented with inexpensive massmarket devices. Moreover, these modules are transparent to existing hardware and require only minor software modification. Simulations were carried out to evaluate the proposed architecture. The results suggest that shared local buffering can provide enough bandwidth to accommodate concurrent access from streaming devices, if internal read and write buffers are provided. A simple round-robin arbitration scheme with limited transfer length is effective in allocating memory bandwidth fairly to these buffers, if both buffer size and transfer length limit are matched to streaming device characteristics. The simulations show that a single streaming module can support as many as four times the number of data streams handled by a conventional server.
| Original language | English |
|---|---|
| Pages (from-to) | 79-89 |
| Number of pages | 10 |
| Journal | International Journal of Computers and Applications |
| Volume | 23 |
| Issue number | 2 |
| Publication status | Published - 2001 |