"... In PAGE 8: ........... 38 Table4 -1. Approximate Storage Space and Costs for One Complete Year of TransGuide Loop Detector Data .... In PAGE 8: ...................................... 45 Table4 -2. Performance Measures in ITS DataLink .... In PAGE 51: ... At this time, the cost of computer storage has become relatively inexpensive ($100-200 per GB in 1997 dollars, depending on the computer platform). Table4 -1 contains estimates of approximate data... In PAGE 54: ...2.5 Calculate Performance Measures The research team designed the ITS DataLink system to calculate and summarize the performance measures shown in Table4 -2. The table indicates that performance measure can be calculated for three different spatial levels: point, link, and corridor or system.... In PAGE 54: ... These assumptions are usually reasonable in free-flow traffic conditions but may be questionable for congested or stop-and-go traffic. Table4 -2. Performance Measures in ITS DataLink Analysis Level Performance Measures Point Measures spot speed (kilometers per hour) person volume or throughput (persons per hour) Link-Based Measures travel time (seconds or minutes) average speed (kilometers per hour) person volume or throughput (persons per hour) person-movement speed (person-kilometers per hour) person delay (person-hours) Corridor or System Measures average person speed (person-kilometers per hour) (not yet implemented) total person delay (person-hours) person-km (mi) of travel in congestion (total and percent) person-hours of travel in congestion (total and percent) corridor mobility index roadway congestion index 4.... ..."

"... In PAGE 12: ... The rst veri cation was of the Alternating Bit Protocol, a simple stop-and-go protocol for correcting data corruptions. In Table1 proofs with di erent numbers of message-types... ..."

"... In PAGE 9: ...Table 1: Special Cases of the Delay-Jitter Controlling Regulator The definition of the delay-jitter controlling policy is general. As shown in Table1 , regulators used in previously proposed non-work-conserving such as RCSP [28], Jitter-EDD [26], and Stop-and-Go [8] are its special cases. In RCSP, i;j in (3) is 0.... In PAGE 9: ...pecial cases. In RCSP, i;j in (3) is 0. In Jitter-EDD, ETk i;j is defined to be ATk i;j + Aheadk i?1;j, where ATk i;j is the arrival time of the kth packet at the ith server, and Aheadk i?1;j is the amount of time the kth packet was ahead of schedule at the i ? 1th server. It is easy to show that this definition is equivalent to the definition in Table1 when the link delay is constant, or i = i;j. In Stop-and-Go, the frame time Tm is the local delay bound for the connection at all servers along the path.... ..."

"... In PAGE 11: ...80 622.52 (38:82 ; ) Table1 : Upper bounds on queueing delay for FCFS in M 1 through M 5 . Since M (k)= S (k;k), M (k)= C l + 31 C l k : Step 2: Q M H 16 C l H X j=1 S (H;j 1) + M (j 1) = 32 C l H 1 X j =0 C l + 31 C l j = 32 31 C l + 31 C l H 1 2 79 48 H since C l = 48 : C BOUNDS ON DELAY FOR STOP-AND-GO QUEUEING In [11, 10] Golestani presents a queueing discipline (stop- and-go queueing) which supports an upper and lower bo- und on delay in a connection-oriented network.... ..."

"... In PAGE 8: ... We investigate the common mechanisms in these disciplines that contribute to such desired features as bounded delays, bounded delay jitters, decoupling of delay and bandwidth allocation, evenly distributed buffer space reservation to prevent packet loss, and easy implementation in high-speed switches. In Table2 , we can see that each desired feature is im- plemented by nearly identical mechanisms in each service discipline. The only exception is that Jitter-EDD uses a sorted priority queue to allocate delays, while HRR, Stop- and-Go and RCSP use static multi-level priorities.... ..."

"... In PAGE 10: ... Zhang and Keshav [35] have also investigated hard real-time communication in a survey paper. Table1 attempts to classify the above according to criteria to be discussed in more detail below. Due to space constraints, we omit discussing the extensive work on real-time communication in local area networks [36{40].... In PAGE 15: ... Also, it may attempt another route. An important point to make about rate-based methods is the coupling between the service parameters, as shown in the last line of Table1 . For instance, Golestani shows that for the Stop-and-Go method [28], jitter, bu er space, and end-to-end delay are all linearly proportional to a parameter called the frame size Fg, and the increments of rate allocation are inversely proportional to Fg.... In PAGE 16: ... A method without any input regulation whatsoever would therefore necessarily be work-conserving. Jitter-bounded approaches are always non-work-conserving, while the converse is not true, as illustrated in Table1 . Work conservation seems like a more attractive option, since the message delivery times promise to be faster.... ..."

"... In PAGE 4: ... These seven batching strategies are listed in Table 1. In Table1 , batching strategies from S1 to S4 are denoted as simple mix, while batching strategies from S5 to S7 are denoted as pool mixes. From Table 1, we can see that the transmission of a batch of packets can be triggered by different events, for example, queue length reaching a predefined threshold, a timer having a time out or some combination of these two.... In PAGE 4: ... In Table 1, batching strategies from S1 to S4 are denoted as simple mix, while batching strategies from S5 to S7 are denoted as pool mixes. From Table1 , we can see that the transmission of a batch of packets can be triggered by different events, for example, queue length reaching a predefined threshold, a timer having a time out or some combination of these two. Flow-based mix networks can also use stop-and-go and similar mixing strategies (Danezis, 2004; Kesdogan et al.... ..."

"... In PAGE 8: ... We investigate the common mechanisms in these disciplines that contribute to such desired features as bounded delays, bounded delay jitters, decoupling of delay and bandwidth allocation, evenly distributed buffer space reservation to prevent packet loss, and easy implementation in high-speed switches. In Table2 , we can see that each desired feature is im- plemented by nearly identical mechanisms in each service discipline. The only exception is that Jitter-EDD uses a sorted priority queue to allocate delays, while HRR, Stop- and-Go and RCSP use static multi-level priorities.... ..."

"... In PAGE 9: ...Table 1: Special Cases of the Delay-Jitter Controlling Regulator The definition of the delay-jitter controlling policy is general. As shown in Table1 , regulators used in previously proposed non-work-conserving such as RCSP [28], Jitter-EDD [26], and Stop-and-Go [8] are its special cases. In RCSP, a127 a56 a83 a78 in (3) is 0.... In PAGE 9: ... In Jitter-EDD, a115a121a116 a118 a56 a83 a78 is defined to be a124 a116 a118 a56 a83 a78 a53 a124 a85 a134a12a66 a82 a118 a56 a43 1a83 a78 , where a124 a116 a118 a56 a83 a78 is the arrival time of the a120 a72a46a74 packet at the a71 a72a46a74 server, and a124 a85 a134a4a66 a82 a118 a56 a43 1a83 a78 is the amount of time the a120 a72a46a74 packet was ahead of schedule at the a71a26a112 1a72a46a74 server. It is easy to show that this definition is equivalent to the definition in Table1 when the link delay is constant, or a114 a56 a122 a114 a56 a83 a78 . In Stop-and-Go, the frame time a116 a47 is the local delay bound for the connection at all servers along the path.... ..."

"... In PAGE 16: ... In contrast, in an RCSP server, a connection can be assigned to any priority level regardless of its rate parameters. Table3 summarizes the regulators and schedulers for the four disciplines. Notice that there are two equivalent definitions of eligibility times for each of the DJ,, DJ, and DJ, regulators.... In PAGE 16: ... F. Delay-Jitter-Control and Rate-Jitter-Control Regulators As shown in Table3 , the regulators for RCSP/DJ,, jitter-EDD, and stop-and-go are very similar. For each of the three regulators, the eligibility time of a packet at a switch is defined with respect to the eligibility time of the same packet at the previous switch.... In PAGE 16: ... G. End-to-End Delay Characteristics and Buffer Space Requirements The end-to-end delay characteristics and buffer space requirement for nonwork-conserving disciplines are shown in Table3 . In the table, D(bj,b*) is the worst-case delay ZHANG SERVICE DISCIPLINES FOR GUARANTEED PERFORMANCE IN PACKET-SWITCHING NETWORKS - ... In PAGE 17: ... Compared to FFQ-based disciplines, rate-controlled ser- vice disciplines have the additional advantage of requiring less buffer space inside the network to prevent packet loss. Based on (17)-(19), and Table 4, it can be easily shown that the total amount of buffer space required for connection j in a network of rate-controlled servers is which is less than Alternatively, based on Table3 , the total amount of buffer space required for connection j in a network of WFQ servers is Since aj, which is the maximum burst size, is usually much larger than a packet size, the terms with aj dominate (21) and (22). While the amount of the buffer space required for a connection increases linearly with the number of hops when WFQ is used, the amount of buffer space is almost independent of the number of hops when rate-controlled service disciplines are used.... ..."