D. Bertsekas and R. Gallager, Data Networks, Prentice Hall International, 2nd Edition, 1992, p.211.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....a situation has been prevalent in the Internet until now, and is likely to remain so for another few years. The preferred service model in that situation, known as best effort service, consists in allocating a fair proportion of bandwidth to contending users; see, e.g. Bertsekas and Gallager [1]. There are actually several possible notions of fairness available for this bandwidth allocation problem (see, e.g. Mo and Walrand [2] for a parametric family of fairness criteria covering all other notions proposed so far) although the classical notion proposed in [1] is the so called max min ....

.... Bertsekas and Gallager [1] There are actually several possible notions of fairness available for this bandwidth allocation problem (see, e.g. Mo and Walrand [2] for a parametric family of fairness criteria covering all other notions proposed so far) although the classical notion proposed in [1] is the so called max min fairness. Recent work has led to a relatively good understanding of how bandwidth is shared between network users when a given congestion control algorithm is used; see, e.g. Massouli e and Roberts [3] and references therein. The question of what type of fairness is ....

[Article contains additional citation context not shown here]

D. Bertsekas and R. Gallager, Data Networks, Prentice-Hall International, 2nd edition, 1992.

....any frame synchronization scenario using the modulated Poisson process model. When a large number of video streams are multiplexed, the assumption of using random smoothing at the source can be relaxed since a large number of similar and independent sources can be considered as a Poisson process [43]. Thus, as long as the cells of a frame do not arrive in burst as a result of some form of smoothing whether it is random or deterministic at the source, we can model the Note that this is a cell link delay bound. To calculate the end to end frame delay bound, we should consider ....

....the case in an M=M=1 system analysis. In the M=M=1 system, the packet inter arrival and service times are correlated. To handle this difficulty, Kleinrock proposed to use Independence Approximation in analyzing a communication network using a general queueing network like the Jackson network [43, 47]. It asserts that, in an M=M=1 system, merging several cell streams on a transmission link has an effect akin to restoring the independence of interarrival times and service times. In particular, he emphasized the independence of service times of a packet at different nodes, which is not true in ....

D. Bertsekas and R. Gallager, Data Networks, Prentice-Hall International, Englewood Cliffs, New Jersey, second edition, 1992.

....LFFS and round robin algorithm. V. SIMULATION RESULTS A. Overview The packet error ratio (PER) resulted from the fading and interference in the wireless network is most severe to the system. The investigation in this paper considers that ARQ algorithm is based on selective repeat (SR) algorithm [14], i.e. the reception of a packet is acknowledged by the receiver by sending either an acknowledgment (ACK) or negative acknowledgment (NCK) to the transmitter. Only the erroneous packets are retransmitted. Due to the reason that only the performance of scheduling algorithm is the emp hasized ....

D. Bertsekas and R. Gallarger, Data Networks, second edition, Prentice-Hall International, Inc, 1992

....may be delayed. This is a reasonable assumption to make, since nodes can be expected to monitor the state of their adjacent links and register changes in the link parameters immediately. Propogation of this information either directly or indirectly (by executing distributed Bellman Ford Algorithm [2]) to other nodes will generally be delayed. 3.2 Problem Formulation We model a channel establishment request (also referred to as a call) in the network described above, as a 5 tuple : R = id; s; d; B; Delta) where id is the call request identification number, s ffl V is the source node ....

....ffl LDNHOP (x; d) the first link on the least delay path from x to d. ffl LCNHOP (x; d) the first link on the least cost path from x to d. These values are assumed to be available at each node as a result of executing a distributed distance vector algorithm like the Bellman Ford algorithm [2]. 1. Residual Delay Maximising (RDM) Heuristic : This heuristic is a call specific heuristic. Let a call setup packet P belonging to the callrequest R = id; s; d; B; Delta) arrive at node v. For each link l = v; x) at v , let RDM(l;R) denote the value of the heuristic for link l corresponding ....

Bertsekas D and Gallager R, Data Networks, Prentice-Hall International, 2nd ed., 1992

....may be delayed. This is a reasonable assumption to make, since nodes can be expected to monitor the state of their adjacent links and register changes in the link parameters immediately. Propagation of this information either directly or indirectly (by executing distributed Bellman Ford Algorithm [5]) to other nodes will generally be delayed. In the remainder of the thesis, unless otherwise stated, the static network model will be assumed. A path P = v 0 ; v 1 ; v 2 ; Delta Delta Delta ; v n ) in this network, has two associated characteristics : 5 Cost C(P ) n Gamma1 X i=0 C(v i ....

....ffl LDNHOP (x; d) the first link on the least delay path from x to d. ffl LCNHOP (x; d) the first link on the least cost path from x to d. These values are assumed to be available at each node as a result of executing a distributed distance vector algorithm like the Bellman Ford algorithm [5]. 1. Residual Delay Maximizing (RDM) Heuristic : This heuristic is a call specific heuristic. Let a call setup packet P belonging to the call request R = id; s; d; B; Delta) arrive at node v. For each link l = v; x) at v , let RDM(l;R) denote the value of the heuristic for link l corresponding ....

D. Bertsekas and R. Gallager, Data Networks, Prentice-Hall International, 2nd ed., 1992

....may be delayed. This is a reasonable assumption to make, since nodes can be expected to monitor the state of their adjacent links and register changes in the link parameters immediately. Propogation of this information either directly or indirectly (by executing distributed Bellman Ford Algorithm [2]) to other nodes will generally be delayed. 3.2 Problem Formulation We model a channel establishment request (also referred to as a call) in the network described above, as a 5 tuple : 5 R = id; s; d; B; where id is the call request identi cation number, s V is the source node for the ....

....node d. LDNHOP (x; d) the rst link on the least delay path from x to d. LCNHOP (x; d) the rst link on the least cost path from x to d. These values are assumed to be available at each node as a result of executing a distributed distance vector algorithm like the Bellman Ford algorithm [2]. 1. Residual Delay Maximising (RDM) Heuristic : This heuristic is a call speci c heuristic. Let a call setup packet P belonging to the callrequest R = id; s; d; B; arrive at node v. For each link l = v; x) at v , let RDM(l;R) denote the value of the heuristic for link l corresponding to ....

Bertsekas D and Gallager R, Data Networks, Prentice-Hall International, 2nd ed., 1992

....Such a situation has been prevalent in the Internet until now, and is likely to remain so for another few years. The preferred service model in that situation, known as best effort service, consists in allocating a fair proportion of bandwidth to contending users; see, e.g. Bertsekas and Gallager [1]. There are actually several possible notions of fairness available for this bandwidth allocation problem (see, e.g. Mo and Walrand [2] for a parametric family of fairness criteria covering all other notions proposed so far) although the classical notion proposed in [1] is the so called max min ....

.... Bertsekas and Gallager [1] There are actually several possible notions of fairness available for this bandwidth allocation problem (see, e.g. Mo and Walrand [2] for a parametric family of fairness criteria covering all other notions proposed so far) although the classical notion proposed in [1] is the so called max min fairness. Recent work has led to a relatively good understanding of how bandwidth is shared between network users when a given congestion control algorithm is used; see, e.g. Massoulie and Roberts [3] and references therein. The question of what type of fairness is ....

[Article contains additional citation context not shown here]

D. Bertsekas and R. Gallager, Data Networks, Prentice-Hall International, 2nd edition, 1992.

....is obtained by simply convolving rate histograms of component video streams. When a large number of video streams are multiplexed, the assumption of using random smoothing at the source can be relaxed since a large number of similar and independent sources can be considered as a Poisson process [4]. Thus, as long as the cells belonging to the same frame do not arrive in burst (i.e. some form of smoothing is executed at the source) we can model the cell arrivals of the aggregate video as a modulated Poisson process. 45 0 50 100 150 200 250 300 350 400 450 500 0 2 4 6 8 10 12 x ....

....the case in an M=M=1 system analysis. In the M=M=1 system, the packet inter arrival and service times are correlated. To handle this difficulty, Kleinrock proposed to use Independence Approximation in analyzing a communication network using a general queueing network like the Jackson network [4, 40]. It asserts that, in an M=M=1 system, merging several cell streams on a transmission link has an effect akin to restoring the independence of interarrival times and service times. In particular, he emphasized the independence of service times of a packet at different nodes, which is not true in ....

D. Bertsekas and R. Gallager, Data Networks, Prentice-Hall International, Englewood Cliffs, New Jersey, second edition, 1992.

....Poisson processes. If FCFS queueing policy is used and packet arrival rate is , the average total delay per packet is t d = 1= Gamma) For non preemptive and preemptive resume priority policies, the average total delays per packet are given in Table 1, assuming that there are two priority levels [2]. 1 is the arrival rate of the level 1 (high priority) packets and 2 the arrival rate of the level 2 (low priroty) packets. From the table, it can be seen that when the communication demand is very high (i.e. 1 2 ) preemptive resume policy has a better performance for high priority ....

Dimitri Bertsekas and Robert Gallager, Data Network, Prentice-Hall International, 1992.

....is chosen with the first one excluded, and the same procedure is repeated. Since choosing candidate routes requires the knowledge of the network topology, this approach can be considered as a distributed approach augmented with a centralized feature. The well known Dijkstra shortest path algorithm [2] can be used for finding candidate routes. However, this approach could be potentially very time consuming for the complete search of all possible routes, and its operational overhead which is measured in number of connection establishment request messages is proportional to n. One way to reduce ....

....requests by employing temporary resource reservation. In this scheme, each node in the network maintains certain information of the real time traffic going through it and exchanges this information with its neighbors, so that an algorithm similar to the Bellman Ford shortest path algorithm [2, 17] can be used to find a qualified route. Although the proposed scheme starts with searching all possible routes in parallel , it prunes infeasible and or inferior routes very quickly, thereby reducing the operational overhead, i.e. the number of request messages. Under the realistic assumption ....

[Article contains additional citation context not shown here]

D. Bertsekas and R. Gallager, Data Networks, Prentice-Hall International, Englewood Cliffs, New Jersey, second edition, 1992.

....times) with load chosen uniformly from 0.5 to 1.0. Although this is an unrealistic model, it is used here for two reasons. First, because of the buffer, the data exhibits correlation between nearby samples. Second, the underlying loss rate as a function of load for this is a standard result ([9]) that can be used for comparison, and it possesses the property that the loss rate is exponentially large with large load as is shown to be the case for more realistic situations ( 10] Thus, the M M 1 will give a clear insights into the implementation issues. The values of sample size T in the ....

Bertsekas D. and Gallager, R., Data Networks, Prentice-Hall International, Inc., 1992.

....CDPA is conceived for multipacket sources. These sources, when ready to transmit packets, signal their need to the BS via a signaling channel, e.g. using slots explicitly reserved for this purpose, by means of some random access protocol, e.g. ALOHA or stack algorithms (see, for example, [15]) The BS coordinates the transmissions of reserved MTs within its cell, whereas no attempt is made to coordinate the transmissions of MTs belonging to different cells [10, 11] In other words, perfect coordination is achieved among transmissions within a cell, whereas an ALOHA like mechanism ....

....need some control to operate in the best conditions. In practice, the control procedures are very different. S ALOHA is a completely distributed system, and a control scheme coherent to its philosophy would call for distributed procedures able to estimate the channel traffic or the backlog [15]. In CDPA the traffic control procedure is easily implemented because of its centralized architecture. In fact, the channel traffic is completely under control of the BS,which can easily measure the channel capture performance and manage the traffic accordingly. As an example, a probabilistic ....

D. Bertsekas, R. Gallager Data Networks, Prentice-Hall International, 1992.

....in parallel but lowers the throughput of the connection per time interval. 5.3 A quantitative measure of the TCP fairness To complete the discussion of the tiny network setup this section takes a short glance at the quantitative fairness of the RM protocol with TCP. Besides min max fairness [30], the fairness index F = P n i=1 x i ) 2 = n P n i=1 x 2 i ) with x i being the resource usage of i ( 31] 32] is frequently used [23] It allows a quantitative analysis of fairness and therefore meets our needs. Figure 17: Fairness index F as a function of the packet drop probability. ....

Bertsekas D., Gallager R., Data Networks. Prentice-Hall International, Inc., 1992.

....the network is utilised inefficiently: rejecting more traffic than would be necessary if the load were more evenly balanced. One means of performing such load balancing is to centrally compute optimal allocations of traffic across the network s paths using predictions of expected traffic [Bertsekas Gallager87] When such calculations have been completed, the network management function can configure the network s routing plan to make the best use of the available resources given the predicted traffic. However, as networks grow larger and involve more complex elements, the amount of operational data ....

D. Bertsekas & M. Gallager, Data Networks, Prentice Hall International , Inc. (1987)

....messages, and retransmission. When the hardware is extremely reliable, as in the case of the Myrinet, hardware failures can be treated as catastrophic events; the MSA may focus solely on preventing buffer overflow problems by using window based or return to sender flow control techniques [2]. These techniques obviate the need for timeouts, but still require ACK NACK messages and occasional retransmissions. 4. Overview of MSAs for Myrinet. Several MSAs have been developed for the Myrinet. In the following subsections we present the main characteristics of the most important of these ....

D. Bertsekas and R. Gallager. Data Networks. Prentice-Hall International, 1992.

....for each real time channel establishment request. Basically, the scheme uses the algorithm proposed in [10, 11] to estimate the minimum worst case response time (MWRT) of a real time channel over each link of a route under consideration. By using MWRTs, we then applied the Bellman Ford algorithm [2, 19] to solve the distributed route selection problem. However, this scheme suffers the problem of over estimating link delays. That is, when there are multiple simultaneous channel establishment requests, the link delays may be over estimated as a result of treating the pending channels as if they ....

....t such that W new (t) t, i.e. the smallest t such that 2 Theta d t 10 e 4 = t. This channel s MWRT on this link is therefore 6, i.e. the delay of this link is 6 ms. 2 3. 2 Building Real Time Delay Tables Based on the above definition of link delay, we can apply the Bellman Ford algorithm [2, 19] and a loop free version of the APARNET s previous routing strategy (APRS) 3, 13, 14, 17] to build real time delay tables on each node. As mentioned earlier, the MWRT used to construct real time delay tables does not include the load of pending channels for two reasons: ffl The maximum ....

D. Bertsekas and R. Gallager, Data Networks, Prentice-Hall International, 1987.

No context found.

D. Bertsekas and R. Gallager, Data Networks, Prentice Hall International, 2nd Edition, 1992, p.211.

No context found.

D. Bertsekas and R. Gallager, Data networks, Prentice-Hall International, Inc., 1992.

No context found.

D. Bertsekas and R. Gallager, Data Networks, Prentice Hall International, 2nd Edition, 1992, p.211.

No context found.

D.P. Bertsekas and R. Gallager, Data Networks (Prentice Hall International, Englewood Cliffs, NJ, 1992).

No context found.

Dimitri Bertsekas. Data Networks. Prentice-Hall International, 1992.

No context found.

D. Bertsekas & M. Gallager, Data Networks, Prentice Hall International , Inc. (1987)

No context found.

D. Bertsekas and R. Gallager, Data Networks, Prentice-Hall International, Inc., 1992.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC