Villamizar, C., and Song, C., High Performance TCP in ANSNET. Computer Communications Review, V. 24 N. 5, October 1994, pp. 45-60. URL http://ftp.ans.net/pub/papers/tcp-performance.ps.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....queuing policy. Queuing policy in the Internet is governed by two antagonistic requirements: First, buffer space should be maximized to accommodate sometimes huge transient bursts; often, sufficient memory is provided to buffer an entire end to end round trip time s (RTT) worth of link bandwidth [2]. Second, buffer occupation should be minimum so as not to introduce unnecessary end to end delays: if each of the often as many as 30 routers between source and destination hosts were to buffer 200 ms of traffic in both directions, each message would require 6 s round trip, which clearly is not ....

....PURPLE. Therefore, we reverted to the use of the network simulator NS 2 [11] for all simulations and compare the results with RED and A RED. For our simulations, we first configured queues to achieve high link utilization by setting their capacities in the order of the bandwidth delay product [2], a widely accepted recommendation. Unfortunately, this leads to high queuing delays, and is often not practiced owing to the large amount of buffer space required, which can amount to several hundred megabytes of high speed memory for fast links. Nevertheless, this simulation helps to understand ....

[Article contains additional citation context not shown here]

Curtis Villamizar and Cheng Song. High performance TCP in ANSNET. Computer Communications Review, 24(5):45--60, October 1995.

....determining the severity of congestion. As a result, RED requires a wide range of parameters to operate correctly under different congestion scenarios. While RED can achieve an ideal operating point, it can only do so when it has a sufficient amount of buffer space and is correctly parameterized [5, 29]. In light of the above observation, this paper proposes BLUE, a fundamentally different active queue management algorithm which uses packet loss and link utilization history to manage congestion. BLUE maintains a single probability, which it uses to mark (or drop) packets when they are queued. ....

....of the bottleneck link. One way to solve this problem is to use a large amount of buffer space at the RED gateways. For example, it has been suggested that in order for RED to work well, an intermediate router requires buffer space that amounts to twice the bandwidth delay product [29]. This approach, in fact, has been taken by an increasingly large number of router vendors. Unfortunately, in networks with large bandwidth delay products, the use of a large amounts of buffer adds considerable end to end delay and delay jitter. This severely impacts the ability to run interactive ....

[Article contains additional citation context not shown here]

C. Villamizar and C. Song. High Performance TCP in ANSNET. Computer Communication Review, 24(5):45--60, October 1994. 27

....determining the severity of congestion. As a result, RED requires a wide range of parameters to operate correctly under different congestion scenarios. While RED can achieve an ideal operating point, it can only do so when it has a sufficient amount of buffer space and is correctly parameterized [13]. In light of the above observation, this paper proposes BLUE, a fundamentally different active queue management algorithm which uses packet loss and link utilization history to manage congestion. BLUE maintains a single probability, which it uses to mark (or drop) packets when they are queued. ....

....of the bottleneck link. One way to solve this problem is to use a large amount of buffer space at the RED gateways. For example, it has been suggested that in order for RED to work well, an intermediate router requires buffer space that amounts to twice the bandwidth delay product [13]. This approach, in fact, has been taken by an increasingly large number of router vendors. Unfortunately, in networks with large bandwidthdelay products, the use of a large amounts of buffer adds considerable end to end delay and delay jitter. This severely impacts the ability to run interactive ....

[Article contains additional citation context not shown here]

C. Villamizar and C. Song. High Performance TCP in ANSNET. Computer Communication Review, 24(5):45-- 60, October 1994.

....determining the severity of congestion. As a result, RED requires a wide range of parameters to operate correctly under different congestion scenarios. While RED can achieve an ideal operating point, it can only do so when it has a sufficient amount of buffer space and is correctly parameterized [6, 34]. In light of the above observation, we propose a fundamentally different active queue management algorithm, called BLUE, which uses packet loss and link utilization history to manage congestion. BLUE maintains a single probability, which it uses to mark (or drop) packets when they are queued. If ....

....ECN marks exactly the correct amount of packets to keep sending rate of sources at L Mbs Sinks generate DupAcks or ECN Fig. 2. Ideal scenario been suggested that in order for RED to work well, an intermediate router requires buffer space that amounts to twice the bandwidth delay product [34]. This approach, in fact, has been taken by an increasingly large number of router vendors. Unfortunately, in networks with large bandwidth delay products, the use of large amounts of buffer adds considerable end to end delay and delay jitter. This severely impairs the ability to run interactive ....

[Article contains additional citation context not shown here]

C. Villamizar and C. Song. High Performance TCP in ANSNET. Computer Communication Review, 24(5):45--60, October 1994.

....determining the severity of congestion. As a result, RED requires a wide range of parameters to operate correctly under different congestion scenarios. While RED can achieve an ideal operating point, it can only do so when it has a sufficient amount of buffer space and is correctly parameterized [18]. 2 The idea behind BLUE, on the other hand, is to perform queue management based directly on packet loss and link utilization rather than on the instantaneous or average queue lengths. BLUE maintains a single probability, which it uses to mark (or drop) packets when they are enqueued. If the ....

C. Villamizar and C. Song. High Performance TCP in ANSNET. Computer Communication Review, 24(5):45-- 60, October 1994.

....no clear advantages over TD concerning response times of Web transfers and that the setting of RED parameters influences response time performance only in the high load case. Several publications discussing the setting of RED parameters give either rules of thumb and qualitative recommendations [15, 8, 25] or quantitative models assuming infinite time averages [7] which are not capable of modelling the oscillatory behavior of the RED queue. A control theoretic approach to model the parameter setting of RED is employed in [11] In [26] a quantitative model how to set RED parameters with TCP traffic ....

C. Villamizar and C. Song. High performance TCP in ANSNET. Computer Communication Review, 24(5):45--60, 1994.

....rate, loss rates can be made to increase linearly with the number of connections 2 . Without changing end host congestion control, Equation 2 shows that decreasing the segment size, increasing the bottleneck bandwidth, and increasing the round trip times through the use of additional buffers [26] can slow the rate of congestion notification considerably and thus, decrease the amount of packet loss observed. Note that the dependence of packet loss rates on round trip time explains why RED queues, which attempt to reduce network queueing delays, can potentially increase packet loss rates ....

C. Villamizar and C. Song. High Performance TCP in ANSNET. Computer Communication Review, 24(5):45--60, October 1994. 18

....determining the severity of congestion. As a result, RED requires a wide range of parameters to operate correctly under different congestion scenarios. While RED can achieve an ideal operating point, it can only do so when it has a sufficient amount of buffer space and is correctly parameterized [18]. The idea behind BLUE, on the other hand, is to perform queue management based directly on packet loss and link utilization rather than on the instantaneous or average queue lengths. BLUE maintains a single probability, which it uses to mark (or drop) packets when they are enqueued. If the queue ....

C. Villamizar and C. Song. High Performance TCP in ANSNET. Computer Communication Review, 24(5):45--60, October 1994.

....size while still accommodating bursts of packets without loss. RED s use of randomness breaks up synchronized processes that lead to lock out phenomena. There have been several implementations of RED in routers, and papers have been published reporting on experience with these implementations ([Villamizar94], Gaynor96] Additional reports of implementation experience would be welcome. All available empirical evidence shows that the deployment of active queue management mechanisms in the Internet would have substantial performance benefits. There are seemingly no disadvantages to using the RED ....

Villamizar, C., and Song, C., High Performance TCP in ANSNET. Computer Communications Review, V. 24 N. 5, October 1994, pp. 45-60. URL http://ftp.ans.net/pub/papers/tcpperformance. ps.

....else normal tcp send( 3) opencwnd( if (CWND = 1 scale 1) scale = scale 2 ; if (scale 1) scale = 1; return; normal tcp opencwnd( 4) Every timer interval counter = counter 1; Figure 3. 17: SUBTCP algorithm round trip time, through the use of additional network buffers [65] increases the bandwidth delay product which slows the rate of congestion notification. 3.3 End Host Congestion Control While properly designed active queue management mechanisms like Adaptive RED can help reduce packet losses, such techniques alone cannot guarantee low loss rates especially ....

....determining the severity of congestion. As a result, RED requires a wide range of parameters to operate correctly under different congestion scenarios. While RED can achieve an ideal operating point, it can only do so when it has a sufficient amount of buffer space and is correctly parameterized [15, 65]. In light of the above observation, this chapter proposes a fundamentally different active queue management algorithm, called BLUE, which uses packet loss and link utilization history to manage congestion. BLUE maintains a single probability, which it uses to mark (or drop) packets when they 42 ....

[Article contains additional citation context not shown here]

C. Villamizar and C. Song. High Performance TCP in ANSNET. Computer Communication Review, 24(5):45--60, October 1994.

....determining the severity of congestion. As a result, RED requires a wide range of parameters to operate correctly under different congestion scenarios. While RED can achieve an ideal operating point, it can only do so when it has a sufficient amount of buffer space and is correctly parameterized [5, 29]. In light of the above observation, this paper proposes BLUE, a fundamentally different active queue management algorithm which uses packet loss and link utilization history to manage congestion. BLUE maintains a single probability, which it uses to mark (or drop) packets when they are queued. ....

....of the bottleneck link. One way to solve this problem is to use a large amount of buffer space at the RED gateways. For example, it has been suggested that in order for RED to work well, an intermediate router requires buffer space that amounts to twice the bandwidth delay product [29]. This approach, in fact, has been taken by an increasingly large number of router vendors. Unfortunately, in networks with large bandwidth delay products, the use of a large amounts of buffer adds considerable end to end delay and delay jitter. This severely impacts the ability to run interactive ....

[Article contains additional citation context not shown here]

C. Villamizar and C. Song. High Performance TCP in ANSNET. Computer Communication Review, 24(5):45--60, October 1994. 27

....connections 1 . Without 1 Note that such a scheme may impact the max min fairness of linear changing end host congestion control, Equation 2 shows that decreasing the segment size, increasing the bottleneck bandwidth, and increasing the round trip times through the use of additional buffers [23] can slow the rate of congestion notification considerably and thus, decrease the amount of packet loss observed. Note that the dependence of packet loss rates on round trip time explains why RED queues, which attempt to reduce network queueing delays, can potentially increase packet loss rates ....

C. Villamizar and C. Song. High Performance TCP in ANSNET. Computer Communication Review, 24(5):45--60, October 1994.

....in 2.3 show that larger effective windows allow TCP to fully utilize the capacity of satellite channels. TCP options to provide larger windows have been defined [JB88] BJZ90] JBB92] These extensions have been proven useful in networks with high latency [AHKO97] and in high bandwidth networks [VS95] 23 2.4.2 Selective Acknowledgments XFTP shows that errors can be repaired better if TCP is given a better idea about which segments have been lost. Using N data connections allows TCP to detect up to N lost segments simultaneously. TCP is able to repair up to N lost segments using fast ....

....amount of unacknowledged 36 data a TCP sender can inject into the network, advertising a window larger than permitted by standard TCP should not be harmful to the network. Larger advertised windows have been shown to work well in satellite networks [AHKO97] and over wide area terrestrial networks [VS95] 4.1.2 Selective Acknowledgments The need for SACKs has been shown in section 2.4.2. SACKs allow the receiver to report precisely which segments have arrived. This explicit information allows the sender to more effectively retransmit segments that have not been received. The sender is able to ....

Curtis Villamizar and Cheng Song. High Performance TCP in ANSNET. Computer Communications Review, 24(5):45--60, October 1995.

....window varies up and down by a factor of two over time. Two router design considerations spring immediately from these algorithms. First, a router must provide about one delay bandwidth of buffering if it wishes to make sure that a single TCP can sustain a send rate equal to the link bandwidth [51]. Otherwise TCP will send at less than the link rate after it cuts its window in half. Second, TCP will tend to keep router buffers full no matter how large they are. This means that building routers with huge buffer memories is an invitation to excessive queuing delay. 2.1.4 Fast Retransmit TCP ....

.... for this is that it allows for the natural delay at which the end systems react to signals from the network [31, 29] A TCP specific justification is that this is the minimum amount of buffering that will ensure full utilization when a number of TCP connections share a drop tail router [51]. With any less than a full delay bandwidth product of router memory, the TCPs windows would sum to less than a delay bandwidth product after they all halve their window sizes. At least one major router vendor [49, 48] uses a delay bandwidth product of memory. Another answer is that buffering is ....

[Article contains additional citation context not shown here]

Curtis Villamizar and Cheng Song. High performance tcp in ansnet. Computer Communications Review, 24(5), October 1994.

....to prevent congestion collapse. In lieu of explicit congestion notification, Equation (2) provides some insight on how to reduce loss rates. In particular, decreasing the segment size, increasing the bottleneck bandwidth, and increasing the round trip times through the use of additional buffers [24] can slow the rate of congestion notification considerably and thus, decrease the amount of packet loss observed. Note that the dependence of packet loss rates on round trip times explains why red queues, which attempt to reduce network queueing delays, can potentially increase packet loss rates ....

....red and subtcp modifications are used together, they can provide optimal performance regardless of the amount of buffering in the network. This allows the network to maintain high degrees of efficiency without having to incur large queueing delays through the use of additional network buffers [19, 24]. The previous experiments fixed both the round trip times and the percent bandwidth increase used (1 ) Since the performance of the proposed congestion control mechanisms have an inherent dependence on both, we ran several experiments which varied them. Figure 20(b) plots the loss rates using ....

C. Villamizar and C. Song. High Performance TCP in ANSNET. Computer Communication Review, 24(5):45--60, October 1994.

....lieu of explicit congestion notification, Equation 2 provides some insight on how to reduce loss rates. In particular, decreasing the segment size used in congestion avoidance, increasing the bandwidth of the congested link, and increasing the round trip times through the use of additional buffers [24] can slow the rate of congestion notification considerably and thus, decrease the amount of packet loss observed. Note that the dependence of packet loss rates on round trip times explains why red queues, which attempt to reduce network queueing delays, can potentially increase packet loss rates ....

C. Villamizar and C. Song. High Performance TCP in ANSNET. Computer Communication Review, 24(5):45--60, October 1994.

....make Vegas about 3 8 faster than Reno. Alternatively, changing the granularity of the Reno coarse timer from 500ms to 200ms (which doesn t require cooperating receivers) would reduce the savings that Vegas obtains by decreasing the frequency of timeouts. Random Early Drop (RED) gateways [12] are another approach to TCP congestion avoidance. While implementing selective acknowledgments requires changing TCP receivers, RED gateways require changing network switches. The question of how RED gateways treat competing Reno and Vegas traffic deserves attention. Reno, since it tends to ....

Curtis Villamizar and Cheng Song. High performance TCP in ANSNET. Computer Communication Review, 24(5):45--61, October 1995.

No context found.

Villamizar, C., and Song, C., High Performance TCP in ANSNET. Computer Communications Review, V. 24 N. 5, October 1994, pp. 45-60. URL http://ftp.ans.net/pub/papers/tcp-performance.ps.

No context found.

Curtis Villamizar and Cheng Song. High performance TCP in ANSNET. Computer Communications Review, 24(5):45--60, October 1994.

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