Citation Context

...terature Survey and Related Work The earliest of recent transport layer flow control research has concentrated on modifications of the byte-oriented 16-bit window-based algorithms used in TP4 and TCP =-=[26]-=-[27][29][38]. (A 16-bit window only allows for 64KB of outstanding data). XTP [12] uses a byte-oriented 32-bit window in conjunction 0-7803-4386-7/98/$10.00 (c) 1998 IEEE with a rate/burst-specified o...

Citation Context

...vey and Related Work The earliest of recent transport layer flow control research has concentrated on modifications of the byte-oriented 16-bit window-based algorithms used in TP4 and TCP [26][27][29]=-=[38]-=-. (A 16-bit window only allows for 64KB of outstanding data). XTP [12] uses a byte-oriented 32-bit window in conjunction 0-7803-4386-7/98/$10.00 (c) 1998 IEEE with a rate/burst-specified open-loop rat...

Citation Context

...s primarily due to inefficient use of bandwidth, and that loss performance can be grossly inadequate without substantial buffering requirements, which in turn results in poor packet delay performance =-=[7]-=-[40][42]. Clearly, new mechanisms for efficient transport layer flow control are needed. 1.1 Literature Survey and Related Work The earliest of recent transport layer flow control research has concent...

Citation Context

...d rate behavior, meaning that congested nodes cycle through periods of over- and under-utilization.sFlow control windows have also been extensively studied at the network layer. Ramakrishnan and Jain =-=[39] depend on-=- "congestion indication bits" to throttle/revive a flow control window. Jain [29] performs a similar study using round-trip packet delays as congestion indicators. Mitra and Seery [34] and H...

Citation Context

...ontrol research at the transport layer has focused on rate-based control. Danthine [15] shows that ratebased flow control responds better to QoS requirements than window-based flow control does. VMTP =-=[11]-=- (streamlined for request-response transactions) and HeiTP [16] (designed to work with the ST-II multimedia network protocol) use an interpacket gap specification. Jacquet and Muhlethaler [28] suggest...

Citation Context

...d to work with the ST-II multimedia network protocol) use an interpacket gap specification. Jacquet and Muhlethaler [28] suggest a multilayer flow control scheme based on bandwidth reservation. Tenet =-=[3]-=- contains a lightweight transport layer protocol, which defers all flow control to the network layer, trusting that end-user resources will be protected by the admission control procedure. A great dea...

Citation Context

... deployed transport layer protocols were designed for networks with small bandwidth-delay products. We are now faced with "Long Fat Networks" (LFNs)--networks with very large bandwidth-delay=-= products [27]-=-. Under these new conditions, the old transport protocols will no longer perform with the same utilization efficiency. Multi-gigabit data rates are already available at the link and network levels, bu...

Citation Context

... Survey and Related Work The earliest of recent transport layer flow control research has concentrated on modifications of the byte-oriented 16-bit window-based algorithms used in TP4 and TCP [26][27]=-=[29]-=-[38]. (A 16-bit window only allows for 64KB of outstanding data). XTP [12] uses a byte-oriented 32-bit window in conjunction 0-7803-4386-7/98/$10.00 (c) 1998 IEEE with a rate/burst-specified open-loop...

Citation Context

...and Lin [41] analyze a credit manager algorithm with feedback at the usernetwork access point, using a zero-delay assumption Another popular network layer flow control algorithm is the "choke pac=-=ket" [5]-=-[36] in which a single bit is used to return congestion information to the sender, indicating whether the queue-size at a bottleneck node is above or below a given threshold. Congestion causes a propo...

Citation Context

...research explores the effects of variable thresholds, proper selection of gain and damping parameters, the effect of multiple connections on a single link, and long propagation delays [24][25][36][23]=-=[6]-=-. Yin and Hluchyj [44] analyze a 2-rate server whose serving rate depends on bottleneck node queue-size. Threshold-based algorithms naturally result in periodic queue-size and rate behavior, meaning t...

Citation Context

...te control. A well-known open-loop rate control algorithm is the "leaky bucket," in which network access is controlled by a quota of tokens which accumulate at a fixed rate, up to the bucket=-= capacity [2]-=-[32]. Konstantopoulos et. al. [30] prove that the classic leaky bucket controller provides optimal cell loss performance under finite buffer constraints. An enhancement to the leaky bucket is a markin...

Citation Context

...king algorithm, which allows traffic with insufficient credit to still enter the network with a violation tag, at the risk of being dropped first if it later encounters congestion [2][10][41]. Berger =-=[4]-=- examines an implementation in which traffic arriving at a queue with no tokens is dropped. Rubin and Lin [41] analyze a credit manager algorithm with feedback at the usernetwork access point, using a...

Citation Context

... Lin [41] analyze a credit manager algorithm with feedback at the usernetwork access point, using a zero-delay assumption Another popular network layer flow control algorithm is the "choke packet=-=" [5][36]-=- in which a single bit is used to return congestion information to the sender, indicating whether the queue-size at a bottleneck node is above or below a given threshold. Congestion causes a proportio...

Citation Context

...p rate control algorithm is the "leaky bucket," in which network access is controlled by a quota of tokens which accumulate at a fixed rate, up to the bucket capacity [2][32]. Konstantopoulo=-=s et. al. [30]-=- prove that the classic leaky bucket controller provides optimal cell loss performance under finite buffer constraints. An enhancement to the leaky bucket is a marking algorithm, which allows traffic ...

Citation Context

... effects of variable thresholds, proper selection of gain and damping parameters, the effect of multiple connections on a single link, and long propagation delays [24][25][36][23][6]. Yin and Hluchyj =-=[44]-=- analyze a 2-rate server whose serving rate depends on bottleneck node queue-size. Threshold-based algorithms naturally result in periodic queue-size and rate behavior, meaning that congested nodes cy...

Citation Context

...ontrol scheme for coupled error recovery and flow control. NETBLT [13] employs a block-oriented window with a rate/burst-specified open-loop rate control scheme for efficient bulk data transfers. SNR =-=[35]-=- uses a packet/block-oriented window flow control with periodic state information exchanges, rather than standard asynchronous acknowledgements. See Table 1 for a comparison of some of the transport l...

Citation Context

...ee Table 1 for a comparison of some of the transport layer flow control algorithms discussed here. More recent flow control research at the transport layer has focused on rate-based control. Danthine =-=[15]-=- shows that ratebased flow control responds better to QoS requirements than window-based flow control does. VMTP [11] (streamlined for request-response transactions) and HeiTP [16] (designed to work w...

Citation Context

...nal research explores the effects of variable thresholds, proper selection of gain and damping parameters, the effect of multiple connections on a single link, and long propagation delays [24][25][36]=-=[23]-=-[6]. Yin and Hluchyj [44] analyze a 2-rate server whose serving rate depends on bottleneck node queue-size. Threshold-based algorithms naturally result in periodic queue-size and rate behavior, meanin...

Citation Context

...ongestion indication bits" to throttle/revive a flow control window. Jain [29] performs a similar study using round-trip packet delays as congestion indicators. Mitra and Seery [34] and Hahne et.=-= al. [20]-=- also explore methods of dynamically controlling flow control window sizes. Much research has been done at the data link/network layer. However, the results of these network layer studies are not alwa...

Citation Context

...rily due to inefficient use of bandwidth, and that loss performance can be grossly inadequate without substantial buffering requirements, which in turn results in poor packet delay performance [7][40]=-=[42]-=-. Clearly, new mechanisms for efficient transport layer flow control are needed. 1.1 Literature Survey and Related Work The earliest of recent transport layer flow control research has concentrated on...

Citation Context

... Jain [39] depend on "congestion indication bits" to throttle/revive a flow control window. Jain [29] performs a similar study using round-trip packet delays as congestion indicators. Mitra =-=and Seery [34]-=- and Hahne et. al. [20] also explore methods of dynamically controlling flow control window sizes. Much research has been done at the data link/network layer. However, the results of these network lay...

Citation Context

...d control. Danthine [15] shows that ratebased flow control responds better to QoS requirements than window-based flow control does. VMTP [11] (streamlined for request-response transactions) and HeiTP =-=[16]-=- (designed to work with the ST-II multimedia network protocol) use an interpacket gap specification. Jacquet and Muhlethaler [28] suggest a multilayer flow control scheme based on bandwidth reservatio...

Citation Context

...control. A well-known open-loop rate control algorithm is the "leaky bucket," in which network access is controlled by a quota of tokens which accumulate at a fixed rate, up to the bucket ca=-=pacity [2][32]-=-. Konstantopoulos et. al. [30] prove that the classic leaky bucket controller provides optimal cell loss performance under finite buffer constraints. An enhancement to the leaky bucket is a marking al...

Citation Context

...timates, an algorithm for optimal selection of h f and α in order to minimize the maximum packet delay, and a discussion on the effects of random network delays. These results may be found in [8] and=-= [9]-=-. 4) System loading. If there is insufficient traffic loading to make use of reallocated bandwidth, there is no need to worry about secondary rate limits and re-sharing underutilized bandwidth. For ca...

Citation Context

...y bucket is a marking algorithm, which allows traffic with insufficient credit to still enter the network with a violation tag, at the risk of being dropped first if it later encounters congestion [2]=-=[10]-=-[41]. Berger [4] examines an implementation in which traffic arriving at a queue with no tokens is dropped. Rubin and Lin [41] analyze a credit manager algorithm with feedback at the usernetwork acces...

Citation Context

...] uses a byte-oriented 32-bit window in conjunction 0-7803-4386-7/98/$10.00 (c) 1998 IEEE with a rate/burst-specified open-loop rate control scheme for coupled error recovery and flow control. NETBLT =-=[13]-=- employs a block-oriented window with a rate/burst-specified open-loop rate control scheme for efficient bulk data transfers. SNR [35] uses a packet/block-oriented window flow control with periodic st...

Citation Context

...fic. Additional research explores the effects of variable thresholds, proper selection of gain and damping parameters, the effect of multiple connections on a single link, and long propagation delays =-=[24]-=-[25][36][23][6]. Yin and Hluchyj [44] analyze a 2-rate server whose serving rate depends on bottleneck node queue-size. Threshold-based algorithms naturally result in periodic queue-size and rate beha...

Citation Context

... Additional research explores the effects of variable thresholds, proper selection of gain and damping parameters, the effect of multiple connections on a single link, and long propagation delays [24]=-=[25]-=-[36][23][6]. Yin and Hluchyj [44] analyze a 2-rate server whose serving rate depends on bottleneck node queue-size. Threshold-based algorithms naturally result in periodic queue-size and rate behavior...

Citation Context

...s. VMTP [11] (streamlined for request-response transactions) and HeiTP [16] (designed to work with the ST-II multimedia network protocol) use an interpacket gap specification. Jacquet and Muhlethaler =-=[28]-=- suggest a multilayer flow control scheme based on bandwidth reservation. Tenet [3] contains a lightweight transport layer protocol, which defers all flow control to the network layer, trusting that e...

Citation Context

...rimarily due to inefficient use of bandwidth, and that loss performance can be grossly inadequate without substantial buffering requirements, which in turn results in poor packet delay performance [7]=-=[40]-=-[42]. Clearly, new mechanisms for efficient transport layer flow control are needed. 1.1 Literature Survey and Related Work The earliest of recent transport layer flow control research has concentrate...

Citation Context

...cket is a marking algorithm, which allows traffic with insufficient credit to still enter the network with a violation tag, at the risk of being dropped first if it later encounters congestion [2][10]=-=[41]-=-. Berger [4] examines an implementation in which traffic arriving at a queue with no tokens is dropped. Rubin and Lin [41] analyze a credit manager algorithm with feedback at the usernetwork access po...