"... In PAGE 3: ...6 Performance Measurements We measured the frame rate of CPU [Stam 2003], GPU [Harris 2004], and Cell implementations in order to validate our analysis. The results can be seen in Table2 . Note that this is not intended as a benchmark, since the codes are not necessarily optimized.... ..."

"... In PAGE 12: ... The latter can also be demonstrated by loss rate of the UDP-based RTP packets for video streaming in the tested systems. The data in Table4 are collected in two client machines from which details of 6 to 8 sample streams are recorded. For a streaming session, the Accuracy is de ned as the ratio of the average to theoretical value of inter-frame delays (see Table 3) while SD/Mean is the ratio of standard deviation to the average.... ..."

"... In PAGE 16: ...15 of header information each. Table2 presents the theoretical expected rates on eachlayer... In PAGE 16: ... The GES performs the functions of SONET line terminating equipment #28LTE#29, where the section and line overheads of the SONET frames are terminated locally at the GESs and the SONET SPE bytes are transported over the satellite #5B7#5D. Therefore, the throughput shown in Table2 for SONET OC-3c and SONET OC-12c should be achievable over the satellite network. 6 Experimental Results All the experimental scenarios were carried out with the default MTU size for Classical IP over ATM networks, which is 9180 bytes #5B5#5D.... In PAGE 23: ...1 ms 39 ms 534 ms Table 4: Table showing the average throughput, standard deviation, and round trip time on OC-3c connections over a LAN, WAN, and ACTS environments in ten trials. 7 Analysis of results and TCP performance pitfalls The results obtained from the experiments where no congestion conditions were present are a bit less than the theoretical ones shown in Table2 , for all three environments. This is due to the protocol implementations under different operating system kernels in the transmitting and receiving hosts, as well as due to the congestion algorithms implemented in the TCP protocol.... In PAGE 27: ... In our simulation models, the network #28IP#29 and physical #28SONET#29 layers are not included since they are of little signi#0Ccance to the outcome of the results. The impact of these layers is captured by accounting their information overhead; the rates displayed in Table2 were used. Also, due to the long run-time of the ATM segmentation and reassembly process in the simulated satellite environment, an ATM model was not used either in the simulation models of scenarios 1 and 2.... ..."

"... In PAGE 29: ... We circumvented this randomness by including an artificial forwarding delay, which allowed time for the ACK frame to be transmitted before forwarding a new data packet. Table3 shows the analysis with- out any forwarding delay. Without a forwarding delay, the backoff random variable is sampled in the range [0 .... In PAGE 33: .... The extra random delays were removed from ns2. Once we made the adjustments listed above, all three simulators produced identi- cal results, as can be seen in Figure 3. Analytical predictions similar to Table3 with basicRate changed to 11 Mbps prove that the results are indeed correct. 3.... ..."

"... In PAGE 13: ... On the rst row, streaming is done without competition from other programs (low contention) while the number of concurrent video streams are high (high contention) for experiments on the second row. Under low contention, both systems (Fig 5a and 5b) demonstrated timely processing of almost all the frames, as shown by their relatively low variance of inter-frame delay ( Table2 ). Note that some variance are inevitable in dealing... In PAGE 14: ... Upon getting control over CPU, it will try to process all the frames that are overdue within the quantum assigned by the OS (10ms in Solaris). Besides high variance, the average inter-frame delay is also large for VDBMS under high contention ( Table2 ). Note the theoretical inter-frame delay for the sample video is 1=23:97 = 41:72ms.... In PAGE 15: ... The original VDBMS obviously keeps the largest number of concurrent streaming sessions (Fig 6a). However, the seemingly high throughput of VDBMS is just a result of lack of QoS control: all video jobs were admitted and it took much longer time to nish each job ( Table2 ). To avoid an unfair comparison between VDBMS and QuaSAQ, a VDBMS enhanced with QoS APIs is introduced.... ..."

"... In PAGE 4: ...296 Hz 356 Hz 18 Hz 712 Hz 355 Hz 87 Hz 49 Hz 10242 9 Hz 0.038 Hz 89 Hz 5 Hz 178 Hz 99 Hz 21 Hz 18 Hz Table 2: Theoretical peak frames per second (The bandwidth-bound values from Table1 ) and actual measured frames per second. None of the measured times exceed the predicted theoretical peaks, validating the finding that the algorithm is bandwidth bound.... ..."

"... In PAGE 14: ... We assume that the decoder has available some approximation ^ f i of frame i, so the next frame in the sequence can be constructed as ^ f i+1 = ^ f i + d #01f i : Wehave implemented a prototype video compression system based on the ideas de- scribed above and achieved promising initial results. The results of an experiment in which we compressed 30 frames of the standard Miss America video sequence #28the images were #0Crst rescaled to 256 #02 256 pixels#29 are presented in Table2 . The experimentwas performed on a 66 MHz 80486 computer running the OS#2F2 operating system, and the entire video 4 We note that because the wavelet transform is linear, it does not matter from a theoretical standpoint if we form W#28#01f i #29byW#28f i+1 #29 , W#28f i #29orbyW#28f i+1 , fi#29.... ..."

"... In PAGE 5: ...5 Mb/s, a value of 10 for 2 Mb/s and 5 for 1 Mb/s, as shown in Table 1. Since it is possible to exactly compute the coding gain of the slower transmission rates with respect to the 11 Mb/s one, we can verify that the observed threshold values are consis- tent with the signal modulation and coding characteristics, summarized in Table1 , where the theoretical coding gain as- sociated with each transmission rate is shown together with the measured coding gain, computed with the 1.7 conversion factor obtained above.... ..."