Multiplex communications – Data flow congestion prevention or control – Control of data admission to the network
Reexamination Certificate
2000-03-13
2004-02-03
Kizou, Hassan (Department: 2697)
Multiplex communications
Data flow congestion prevention or control
Control of data admission to the network
C370S236000, C370S349000, C370S401000
Reexamination Certificate
active
06687227
ABSTRACT:
FIELD OF THE INVENTION
The invention relates to channel utilization enhancements for long and variable delay channels, such as channel utilization enhancements for TCP (Transport Control Protocol) over high-speed wireless channels.
BACKGROUND OF THE INVENTION
The utilization of a channel, for example during a bulk data transfer, is determined by the balance between the capacity of the channel and the amount of data available to feed into that channel. This balance needs to be managed. If the balance is in favor of the capacity, under-utilization occurs. If the balance is in favor of the amount of data put in transit, optimal utilization is achieved. However, for channels with a variable capacity, there will be excessive packets that may have to be either buffered or dropped, as they will not be accommodated by the transmission channel when the capacity shrinks periodically.
This is particularly true for high-speed third generation (3G) wireless communications systems where one of the major issues in effectively transporting packet data using Internet protocols such as TCP/IP (Transport Control Protocol/Internet Protocol) is such an under-utilization of allocated high-speed radio channels. This is a significant issue since such radio channels are scarce resources due to the fundamental limit on the available radio spectrum. In high-speed 3G wireless environments, a high-speed channel results in a larger bandwidth delay product which is enlarged further by the length and variable delay characteristics of such channels.
The channel capacity of a TCP connection is the product of channel bandwidth and the round trip delay and is a measure of the volume of the “pipe” between the end-hosts. A longer round trip delay increases the capacity and therefore will demand more data onto the channel such that the pipe can be filled. A higher channel speed also allows the channel to accommodate more packets, and will also require the server host to send data more quickly as the pipe empties faster. When a link has long and variable delay the channel capacity will increase, and will also be variable.
Over a wireless link, high error rates may exist (compared with those on fixed wire-line links) and despite various coding techniques these error rates may still translate into a high radio frame erasure rate (typically, 10%), which in turn translates into more than a 10% packet loss for data transfer rates of 64-144-384 kbps. For this reason, link layer retransmissions are employed to compensate for this high frame erasure rate. Typically an Automatic Repeat Request (ARQ) based method is used such as RLP-III (Radio Link Protocol-III). These retransmissions of course contribute to the size and variability of the delay. There is an accumulative delay per retransmission as the frame is inserted into the current transmission stream. Furthermore, a lost frame cannot be released to upper layers until it is retransmitted correctly. If the retransmission protocol provides in-order data delivery, frames will be put off on the receiving side by the lost frame in front of them. These frames will have to wait for the correct retransmission before being released. The residual frame error may be reduced, e.g., to below 0.2% with 2-4 retransmissions for a resulting packet loss of around 1%. These retransmissions extend the one-way delay by 160-320 ms. Such delays only occur when the initial frame is lost, they are variable to TCP/IP, and are proportional to the frame error rate of the link.
These delays will interact with TCP to produce a number of effects with the end result that under-utilization of the channel occurs. Firstly, retransmissions will delay the forwarding of effected packets to the TCP/IP layers. Retransmissions will also push back all the packets closely following each effected packet during the retransmission. Secondly, the corresponding TCP client-to-server direction acknowledgements for all the effected packets will experience delay. The server will wait for the duration of the delay before sending data because TCP uses these acknowledgements to regulate the data flow from the server to the client. Thirdly, if the acknowledgements do not arrive at the TCP server host, it will not transmit new data onto the channel, creating periods of idling of the channel. This will result in under-utilization and also stretch the overall TCP session time. Fourthly, delayed acknowledgement implementation of TCP and possible loss and retransmission of acknowledgements in the client-to-server direction will also contribute to under-utilization.
FIG. 1
illustrates an example delay characteristic experienced by packets to be sent between a wire-line end-host
10
and a mobile end-host
12
. The plot shows a relatively stable delay region
14
for a wire-line portion of the link, and a variable delay region
16
shown by the fluctuating plot lines for a wireless portion of the link.
The impact of stable delay on utilization will be described with reference to
FIG. 2
so that it may be contrasted with the impact of variable delay which is described below with reference to
FIG. 3. A
TCP end-to-end packet flow pattern is plotted in
FIG. 2
where the round trip delay is stable for example for communication between a first wire-line host
10
and a second wire-line host
18
. After TCP completes slow-start, the congestion window and the received advertised window size limits the server to send data in packet bunches corresponding to the size of the client advertised window size. Given a large enough delay of the channel or a high channel speed, TCP utilization of the channel will degrade.
The operations of TCP proceed as follows. Assuming TCP flow control has completed slow-start, the first wire-line host
10
fills in the channel with the maximum number of packets allowed by the minimum of congestion window, the received advertised window size, and the sender's retransmission buffer size. It will wait for the first client-to-server direction ACK packet to come back from the second wire-line host
18
before sending more data. The waiting period becomes one of the root causes of channel under-utilization as soon as synchronization between the number of packets sent and the round trip delay is disrupted. In
FIG. 2
, t is the size of the packet in time; T is the round trip time of the channel from the time the first of the packets is sent until the corresponding ACK is received; n is the number of packets that will be sent as dictated by the smallest of the congestion window size, the received advertised window size, and the sender's buffer size. For the channel to be fully utilized to the total of the bandwidth and delay product, the following equation must hold: T≧n·t.
When T>n·t, there is an idle time; “I” in the radio channel. I equates to the time between when the current full window of packets have been transmitted and the next window of packets triggered by the reception of ACKs for those packets. As this pattern of packet transmission is repeated the overall channel utilization is reduced by I=T−n·t, or as a percentage of reduced utilization U=(T−n·t)/T. The relationship of T and t can be illustrated by the relative difference between channel speed and the overall delay. With a T>>t, which can be either interpreted as long delay of the network, or high data rate channel, even a large n will still cause an idle time I. n=T/t can be achieved by enlarging the joint effect of the TCP congestion window, the received advertised window size, and the sender's buffer size. With a higher channel speed (and/or longer overall network delay), a larger n is required to reduce I, thus to achieve a lower U. For such a channel, an n
1
can be calculated such that if applied, it can allow TCP flow control to fill in the idle time I of the channel and increase utilization to (near) the optimum.
The impact of variable delay will be described with reference to FIG.
3
. As stated previously, variable delay can be caused by link layer retransmissions. For high speed 3
Brost Leslie K.
Cao Carl F.
Li Yalun
Kizou Hassan
Nortel Networks Limited
Swickhamer Christopher M
LandOfFree
Systems and methods for requesting packets for transmission... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Systems and methods for requesting packets for transmission..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Systems and methods for requesting packets for transmission... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3318671