Multiplex communications – Data flow congestion prevention or control
Reexamination Certificate
1999-06-30
2002-10-29
Ton, Dang (Department: 2732)
Multiplex communications
Data flow congestion prevention or control
C370S395100
Reexamination Certificate
active
06473398
ABSTRACT:
This invention relates to congestion management in managed packet-switched networks, such as frame relay networks, and encompasses both the real-time aspects of congestion control and avoidance, and Operations, Administrations and Maintenance procedures to detect the onset of congestion.
Congestion management for frame relay switched networks, which are designed to handle bursty traffic, is an important issue due to the possibility that frame relay resources will become depleted, causing congestion. In this case congestion management includes real-time mechanisms to prevent or recover from congestion, as well as procedures to detect the onset of congestion.
Congestion management procedures have several important requirements. The number of frame discards and level of degradation of service to user devices during congestion must be minimized. They must simple to implement and place little overhead upon the network, be fair in allocating resources among competing PVCs, limit the spread of congestion to other network elements, and optimize the use of network resources. The configuration complexity should be minimized, yet be flexible enough to allow tuning of the network congestion procedures. Discard frames with the Discard Eligibility bit (DE) set should be discarded in preference to other frames during a congestion situation. Facilities that notify severely congested circuits should be provided and they should provide sufficient information so that network operators can isolate the source of congestion.
Existing practice as implemented, for example, on the Newbridge Networks Corporation MainStreet product line uses several mechanisms shown in FIG.
1
. These include negotiating a minimum bandwidth (CIR) among other class of service parameters for each PVC; and monitoring the actual bandwidth utilization at ingress, which involves policing the incoming information rate and marking traffic which exceeds CIR in support of two levels of priority and fairness; monitoring the depletion of network resources; initiating standards-based implicit congestion mechanisms, also referred to as congestion avoidance procedures, during times of low levels of congestion; and turning to standards-based explicit congestion mechanisms to efficiently and effectively recover from congestion during times of high levels of congestion.
The above mechanisms generate congestion alarms to reflect the severe, mild and non-congestion state that can exist within a frame, stream and/or the whole card. Alarms are generated and logged to the Newbridge Networks Corporation 46020 network management system if the circuit/card enters the RED congested state. These indicators are debounced by a low passing filter to manage the transition between the states as shown in FIG.
2
.
Because of the low pass filter implementation, spikes shown in
FIG. 2
are not accounted for when generating the alarm. This causes loss of frames without an actual indicator of this happening. For a congestion alarm to be raised, the frame stream buffer must be over the Severe Congestion Threshold (SCT) for five continuous seconds. If the buffer utilization dips below SCT the counter is reset to zero regardless of the time spent in the SCT range. When in Severe Congestion it is referred to as being in the RED state. The following statistics are also subject to the five second filter: Number of RED alarms, Percentage of time spent in RED state, Duration of longest RED state.
The RED state is significant because in this state Frames with the DE bit set are discarded. The discarding of flames is not subject to the five second delay. There are therefore two definitions of the RED state, one for statistics and alarms (five second delay) and one for discards (instantaneous).
For a traffic pattern shown in
FIG. 3
, the system is in the discarding RED State for 98% of the time (4.9s/5.0s). However the statistical and alarming RED state was never reached. Therefore no alarms are raised and there is no indication that a RED state has been entered, while significant frames have been discarded without any operator alert.
At first glance, the traffic profile in
FIG. 3
may seem very improbable, and unrealistic. In actual practice a similar profile with identical characteristics and the resulting discards and alarms is very common. It is actually the intentional result of the frame relay congestion avoidance scheme.
The traffic profile in
FIG. 4
is a profile that results from the existing frame relay congestion avoidance scheme. When the switch buffers reach the SCT level, all the frames with DE set are discarded. This results in the buffer utilization dropping instantly below the threshold. The buffer then fills up again with the same result. While less time is spent in the discarding RED state than in
FIG. 3
, there are still a significant number of of frames (all frames with DE set at the time the threshold is exceeded) discarded without any alarms or statistics to reflect the reason for their being discarded.
The actual amount of frames discarded will be a function of the number of frames that have DE set. The more frames with DE set, the more discards that will occur. This is shown in FIG.
5
.
The number of frames with DE set (if all setting is done by Newbridge Networks Corporation FREs/FRSs [Frame Relay Engines and Frame Relay Switches) and not the end devices] will be determined by how “tightly” the network is engineered. In a “loosely” engineered network, where the Class of Service (COS) parameters CIR, Bc and Be are significantly greater than the actual traffic, there will not be many DE bits set since CIR and Bc are rarely exceeded. Since relatively few frames are discarded, the buffer may not fall below the SCT threshold, the five second window will expire, and the alarms will be raised and the statistics collected.
In a tightly engineered network where the COS parameters accurately reflect the traffic profiles, there may be many DE bits set. Since many frames may be discarded, the buffer may always fall below the SCT level resetting the five second counter. Therefore the tightly engineered network may discard more frames (more frames with DE set) and never get any alarms.
According to the present invention there is provided a method of managing a packet switched network comprising a plurality of switch nodes and a network manager and wherein packets are discarded in the event of congestion in the network. A congestion event indicator is notified to the network manager immediately whenever the congestion level of a resource in a switch node in the network exceeds a congestion threshold, and the network manager maintains a congestion status for the switch node based on the receipt of said congestion event indicators.
The invention addresses the network requirements with regards to the enhanced frame relay network congestion indicators. These indicators can be monitored by a network management entity in the generation of trouble tickets.
The frame relay engine supports a mechanism for notifying congestion and the occurrence of frame discards on the switch. When the switch buffer usage or a frame Networks Corporation 46020 will be notified. CPU congestion will not cause a ‘Congestion Indication’ to be sent to the network manager.
Preferably, a single congestion indicator event will be sent to the network manager when a resource on the switch exceeds a Congestion Indicator Threshold (CIT). The Congestion Indicator Threshold can be set equal to ACT, SCT or MCT for each resource on the switch. Whenever a frame is to be placed in a resource (e.g. transmit queue for a frame stream) that has reached or exceeded the Congestion Indicator Threshold then the resource is declared as congested. The Congestion Indicator status is updated to indicate that the switch has a congested resource, and an event is sent to the network manager. The status of the Congestion Indicator is changed to ‘No Congestion’ only if the level of congestion of all resources on the switch remain below the ‘Congestion Indicator Threshold’ for a period of a user config
Giuhat Micaela
Katz Felix
Mayer Johanne
Wall Wayne
(Marks & Clerk)
Alcatel Canada Inc.
Ton Dang
LandOfFree
Congestion management in managed packet-switched networks does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Congestion management in managed packet-switched networks, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Congestion management in managed packet-switched networks will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2919900