Assigning Block Size Based on Speculation for Parallel Downloading

With the development of mirror sites, parallel downloading, which can decrease download latency, is becoming more and more popular. Dynamic parallel downloading can be adapted according to the fluctuation of network bandwidths, and has a better performance than static parallel downloading. To reduce the number of requesting document blocks, we propose a scheme in which block size is assigned based on speculation. It deals with the unbalance of assignments during downloading and decreases download latency effectively. Further more, we have implemented a parallel downloading prototype, which makes use of this scheme. Through experimental tests, this scheme shows its advantages of both download latency and the number of block assignments.