"... In PAGE 2: ... the ability to reconstruct natural sounding speech at the receiving side. End-to-end delays have a signifi- cant impact on the quality of interactive voice trans- missions, as shown in Table1 [18]. Variation in the network delay experienced by individual packets, i.... In PAGE 3: ... The first packet in a voice stream is artificially delayed at the receiver for the period of the control time in order to buffer sufficient packets to provide for continuous playback in the presence of jitter. Note however, that the the control time cannot be arbitrar- ily large due to constraints on the end-to-end delay (see Table1 ). Since voice data consists of an alternat- ing series of talkspurts and silence periods and since talkspurts are generally isolated from each other by relatively long silence periods [3], voice protocols typ- ically impose the control time on the first packet of each talkspurt.... In PAGE 5: ... We conduct experiments with the simulation model and provide answers to the following questions: How much control time is needed for S-ARQ to ensure a high probability of successful retransmis- sions? Note that the control time at the PVR results in increased end-to-end delays for all packets. How- ever, since voice transmission is sensitive to end- to-end delay (see Table1 ) the control time cannot be increased arbitrarily. ... ..."

"... In PAGE 31: ... The available bandwidths between Palo Alto and NSP#2, NSP#1 and VBNS are taken to be 20, 5 and 1 Mbits/sec. For the trans- mission of a message from Palo Alto to Washington DC, the single and multipaths that achieve minimum end-to-end delay are given in Table3 . Note that the single paths with minimum end-to-end delay migrate via VBNS, NSP#1 and NSP#2 as the message size is increased.... ..."

"... In PAGE 5: ... In this experiment, the input was a set of English utterances. The utterances were paraphrased back into English via the interlingua ( Table1 ) and translated into Italian (Table 2). The data used to train the DA classifiers consisted of 3350 SDUs annotated with IF representations.... In PAGE 5: ... The table shows the average of grades assigned by three graders. The row in Table1 labeled SR Hypotheses shows the grades when the speech recognizer output is compared directly to human transcripts. ... ..."

"... In PAGE 5: ... In this experiment, the input was a set of English utterances. The utterances were paraphrased back into English via the interlingua ( Table1 ) and translated into Italian (Table 2). The data used to train the DA classifiers consisted of 3350 SDUs annotated with IF representations.... In PAGE 5: ... The table shows the average of grades assigned by three graders. The row in Table1 labeled SR Hypotheses shows the grades when the speech recognizer output is compared directly to human transcripts. ... ..."

"... In PAGE 7: ...5 - 80Mbps) using 256 bytes packet size. Table2 shows the results of the model fitting done for a set of sending bit rates scenarios for 256 bytes packet size. Results were obtained in a similar manner to those in Table 2.... In PAGE 7: ...aried sending bit rates (0.5 - 80Mbps) using 256 bytes packet size. Table 2 shows the results of the model fitting done for a set of sending bit rates scenarios for 256 bytes packet size. Results were obtained in a similar manner to those in Table2 . Form Table 2 it can be seen that ARMA model orders, for both packet delay and IPG series, decreases rapidly for sending bit rates higher than 10Mbps.... In PAGE 7: ... Results were obtained in a similar manner to those in Table 2. Form Table2 it can be seen that ARMA model orders, for both packet delay and IPG series, decreases rapidly for sending bit rates higher than 10Mbps. This can be expected from the PACF distribution observed on Figure 5.... In PAGE 7: ... Also it can be seen that IPG ARMA models offer smaller negative log-maximum likelihood estimator than packet delay for scenarios with sending bit rate greater than 10 Mbps. By comparing results obtained from Table2 to the ones observed on Table 2, it can be concluded that the packet delay model becomes an AR process at smaller sending data rates for the 256 bytes packet size experiment than for the 64 bytes packet size one. Also, in general ARMA/ARIMA packet delay and IPG models show lower orders for the 256 bytes packet size cases than their corresponding 64 bytes packet size cases.... ..."

"... In PAGE 1: ... We report averages over five runs after a warm-up run. Table1 shows the least-squares linear fit of the curves of Figure 1, along with the latencies estimated from Table 6 of [1] using the scaling model. The throughput for single 60 KB AAL5 packets predicted by the scaling model is 132, 348, or 384 Mbps for copy, emulated copy, or emulated share semantics, respec- tively.... In PAGE 1: ... In [2], we discuss the hardware support required for emulated copy in multiple-packet communication. 4 Conclusion The good fit between estimated and actual latencies in Table1 suggests that the scaling model of [1] is accu- rate also with respect to the effect of network speeds. References [1] J.... ..."

"... In PAGE 2: ... The choices of the bounds and the number of the service classes depend on the types of applications. Table1 shows possible end-to-end delay classes for vir- tual collaboration applications. Note that our application- level service classes are different from network-level ser- vice classes in Differentiated Services [5] and the reduced service-set architecture [6].... In PAGE 3: ... 3.1 New Core-based Routing Under Service Class Framework We revisit Table1 showing four possible end-to-end delay classes for virtual collaboration applications. Each class in the table indicates the upper bound of the end-to-end... In PAGE 13: ... We map the link distance to a link delay such that the longest path in the network has the total delay of approximately 70 ms, a coast-to-coast delay observed in the United States. Four service classes in Table1 were used in the experiments. We conducted experiments under various group sizes ranging from 5 to 30 members.... ..."

"... In PAGE 6: ...2. Results Table2 shows a portion of the end-to-end monitor- ing data collected by the MPs for a certain work unit at a sample run of the experiment. It highlights data col- lected from MPs instrumenting service image list and image retrieve B and omits the service pointers.... In PAGE 6: ... The table shows the work unit was properly classified and monitoring data pertaining to it were correctly corre- lated together in an end-to-end manner. Table2 ac- counts for what platforms and services were involved in processing the work unit and the amount of time spent on each of them. The elapsed time data are calculated using Equation 3.... ..."

"... In PAGE 24: ...ig. 10 Exp 2: resource utilization. a Processor utilization; b Bandwidth utilization These results demonstrate that when the system was operated with indepen- dent feedback loops, wireless network bandwidth was severely under-utilized, which therefore leads to a high target tracking error, or a low QoS. Table5 shows the end-to-end delay when the system was operated with indepen- dent feedback loops. From Tables 3 and 5 it can be seen that when the system tracked 2 objects of interest, the same end-to-end delay was achieved when the system was operated with independent feedback loops, with HiDRA, and without HiDRA as the system resource utilization was maintained below the specified utilization set- point.... ..."