According to an aspect of the invention, the link capacity requirement of CBR / VBR traffic is captured by its present filtered rate so that the ABR
transmission rate will be periodically adapted to the remaining link capacity. However, because of multiloop
delay and the CBR / VBR traffic variation, the link capacity will not be fully utilized. Another object of the present invention is therefore to minimize the unused link capacity. An
advantage of the present invention is it eliminates unnecessary highly frequent adaptations of ABR traffic rate as found in most existing control schemes, which are caused either by the
high frequency variation of CBR / VBR traffic or by the rapid change of
queue congestion status. Because of the filtering, the ABR traffic varies very smoothly along with the
slow time variation of the
low frequency CBR / VBR traffic.
is it eliminates unnecessary highly frequent adaptations of ABR traffic rate as found in most existing control schemes, which are caused either by the
high frequency variation of CBR / VBR traffic or by the rapid change of
queue congestion status. Because of the filtering, the ABR traffic varies very smoothly along with the
slow time variation of the low frequency CBR / VBR traffic.
Another object of the invention is to provide an explicit rate ABR traffic control scheme by using a
process control design method, referred to herein as Generated Prediction Control (GPC), to provide a
closed loop stable controller with
linear quadratic optimal performance. An
advantage of the present invention is that multiloop delays of ABR connections are built into the design of the control model. The low frequency high magnitude oscillations of ABR traffic flow, as usually found in most proposed ABR control schemes, are therefore eliminated. As a result, not only is the buffer capacity requirement for ABR traffic substantially reduced, but the control stability condition is also significantly improved.
In a preferred embodiment, a new explicit
rate control scheme based on the results of
frequency domain analysis of
multimedia traffic is provided. Examples of
frequency domain analysis of
multimedia traffic can be found in S. Q. LI and C. HWANG, “
Queue Response to Input Correlation Functions: Continuous
Spectral Analysis,”IEEE / ACM Trans. Networking, Vol. 1, No. 6, December 1993, pp. 678-692 (LI (I) et al. hereinafter), and S. Q. LI, S. CHONG, and C. HWANG, “Link Capacity Allocation and
Network Control by Filtered Input Rate in High Speed Networks,”IEEE / ACM Trans. Networking, Vol. 3, No. 1, February 1995, pp. 10-15 (LI (II) et al. hereinafter), the disclosures of which are expressly incorporated herein by reference in their entireties. The inventors have found that the link capacity required by input traffic at each node is essentially captured by the traffic's low frequency characteristics. In other words, no congestion occurs at the node if the control design guarantees that the aggregate CBR / VBR / ABR traffic rate, filtered in a
low frequency band, never exceeds the link capacity. All the
high frequency traffic is absorbed and smoothed out via limited buffering. Further, the filtered traffic rate is significantly less than its original non-filtered rate.
Another object of the present invention is to provide significant improvement of ABR traffic performance by replacing the
queue threshold detection with the bandwidth threshold detection, using an Explicit Forward Congestion Indication (EFCI) scheme. According to an aspect of the invention, a modified EFCI scheme is provided, which is called EFCI-ECD scheme, where ECD refers to early
congestion detection. A
simulation study has shown about a 60% savings in buffer capacity for the EFCI-ECD scheme to achieve the same
throughput as the original EFCI scheme, providing the same set of source
control parameters. The amplitude of the ABR traffic oscillation is reduced by about 50%. Note that the traffic filtering operation can easily by implemented by
digital signal processing (DSP) chips. Using today's technology, a common DSP
chip only costs a few U.S. dollars whereas the high speed SRAM chips for buffer capacity are relatively much more expensive. Moreover, a single DSP
chip can be shared by many links for multiple traffic measurement purposes.
Thus, there are at least three major advantages for the EFCI-ECD scheme of the present invention: (1) none of the existing EFCI protocols at source and destination needs to be changed; (2) the buffer capacity required at each switching node is substantially reduced; and (3) the ABR traffic oscillation is much reduced.