High performance TCP in ANSNET

Abstract

> Advanti sAbstrac t This report concentrates on specific requirements an d goals of the research networks supported by ANSNET , but applies to any TCP dominated high speed WA N and in particular those striving to support high spee d end-to-end flows . Measurements have been made unde r conditions intended to better understand performanc e barriers imposed by network equipment queueing capacities and queue drop strategies .The IBM RS/6000 based routers currently supportin g ANSNET performed very well in these tests . Measurements have been made with the current software an d performance enhanced software . Single TCP flows are able to achieve 40 Mb/s and competing multiple TC P flows achieve over 41 Mb/s link utilization on 44 .7 Mb/ s DS3 links with delays comparable to US cross continen t ANSNET delays . Congestion collapse is demonstrate d with intentionally reduced queueing capacity and usin g window sizes much larger than optimal .A variation of Floyd and Jacobson's Random Earl y Detection (RED) algorithm [1] is tested . Performanc e improved with the use of RED for tests involving multiple flows . With RED and queueing capacity at o r above the delay bandwidth product, congestion collaps e is avoided, allowing the maximum window size to safel y be set arbitrarily high .Queueing capacity greater than or equal to the dela y bandwidth product and RED are recommended . RED provides performance improvement in all but the singl e flow case, but cannot substitute for adequate queuein g capacity, particularly if high speed flows are to be supported .

Understanding the performance of many TCP flows

“…There has been a vast volume of research on studying RED dropping policy, ranging from analytical modeling [5,31] to simulations and experimental evaluation [45,10,8,48], from recommendations for parameters and architecture configuration [18,19,13,22] to proposals of alternatives, which include BLUE [12], SRED (Stabilized RED) [34], Adaptive RED [11], FRED (Fair Random Early Drop) [28], and BRED (Balanced RED) [4]. Our simulation study of RED performance complements the existing body of research by comparing the aggregate throughput, goodput, loss rate, and roundtrip bias of TCP connections under RED gateways with those under drop-tail gateways.…”

The utility of characterizing the buffer of network devices in order to improve real-time interactive services

“…This rule was proposed in 1994 [21] and it is given by B = C × RT T , where B is the buffer size, RT T is the average round-trip time and C the capacity of the router's network interface. In [22], proposed a reduced buffer size by dividing BDP by the square root of the number of used TCP flows B = C × RT T / √ N .…”

Buffer Sizing in 802.11 Wireless Mesh Networks

“…A widely used rule-of-thumb is to have buffers larger than the bandwidth-delay product (BDP) of the network [5], i.e., ≥ × , where is the bottleneck link capacity along the path, and is the effective round-trip propagation delay through the bottleneck link. This buffer size reflects the number of packets queued at a bottleneck link to keep it fully utilized while a TCP source recovers from a loss-induced window reduction.…”