Predictive bursty real-time traffic control for...

Details Predictive bursty real-time traffic control for... Predictive bursty real-time traffic control for...

1998-09-30

2001-06-26

Kuntz, Curtis (Department: 2643)

Telephonic communications

With usage measurement

C379S114030, C379S137000, C379S133000

Reexamination Certificate

active

06252950

ABSTRACT:

FIELD OF THE INVENTION
This application relates to telecommunications switching systems, and more particularly to systems for controlling a telecommunications switching system to optimize performance when confronted with peak loads.
BACKGROUND OF THE INVENTION
The practice of designing telecommunications systems and networks to meet expected traffic demand is well advanced. Nonetheless, sudden bursts of traffic can cause significant congestion in modern telecommunications systems, including telecommunications switching systems.
Most modern telecommunications switching systems employ “common control”, in which a common pool of control components and associated resources are generally available to process arriving calls. The control components and resources are assigned to any particular call for very limited periods during call set-up, call tear-down, and occasionally during a conversation when services are requested by a user. Accordingly, when call arrival rates are high, a limiting factor in a switching system's ability to satisfy traffic demand is the limited availability of the common control components to attend to the calls.
Moreover, most modern telephone switching systems implement common control using “stored-program control” (SPC), in typical applications of which one or more computers, each including a processor and appropriate software, cooperate with peripheral devices to achieve the required functionality of the switching system. Thus, in SPC systems, the aforementioned limiting factor of limited availability of the control components directly translates to limited availability of the processor to perform real-time control tasks.
Many aspects of switching system engineering treat performance in the aggregate, and assume that the distribution of call arrivals approximates a Poisson distribution. In particular, SPC systems are engineered to carry a maximum traffic load based on the assumption that traffic arrival rates are Poissonian. SPC systems use buffer sizing, queuing, token based schemes, and processor occupancy levels to detect when the processor is overloaded. The maximum acceptable level of processor occupancy is predicted on the basis of statistical sampling and on assumptions derived from Traffic Theory. When processor occupancy exceeds the statistically predicted maximum acceptable level, the switching system is determined to be in real-time overload.
A real-time overload can degrade the speed and grade of service provided by the switching system. If due to overload a switching system does not respond within expected periods to communications from another switch, the other switch may abandon the call entirely, or may abandon the current attempt and retry. As a result, effects of the overload propagate through the network. In addition, the overloaded switch must expend additional real-time resources to “clean up” the abandoned call or to process a subsequent retried call, further contributing to the overload.
Accordingly, switching systems have provided certain overload controls for protecting the operation of the system when overload is detected. To mitigate the effects of real-time overload, a sequence of actions ranging from eliminating or deferring non-essential tasks, to deferring or refusing incoming work, depending on the apparent severity of the overload, may be employed. Deferring or refusing incoming work may take a number of forms. In severe overload conditions, a switching system may refuse to accept new calls for processing and may even instruct other switches not to send calls to the overloaded system. In less severe situations, certain ordinarily scheduled tasks which are less time-critical than call processing are readily deferred or reduced in frequency. Although many of these actions are effective in protecting the switching system so that improper operation is avoided with respect to existing calls and those new calls which are accepted for processing, these actions may also increase overhead, thereby actually reducing the real-time available to process calls, and artificially reducing switch capacity.
For example, consider an approach taken in the Number
4
Electronic Switching System (4 ESS), a product of Lucent Technologies, Inc., the assignee of this application. (The 4 ESS switching system is described in
Bell System Technical Journal
, September 1977 (entire issue) and
Bell System Technical Journal
, July-August 1981 (entire issue).) In the 4 ESS switching system, overload conditions of progressively increasing severity are declared when the mean execution time of a basic unit of work in the real-time operating system, measured over the last 8 work units, initially exceeds 130 ms, and at several predefined thresholds. When an overload condition is declared, the switching system may take several protective actions to control the overload or at least minimize its effects on the switch.
In one aspect of the overload control response, the number of available call registers, a data structure used to hold call-related information during the course of processing a call, may be progressively reduced. Since a call register is required in order to accept an incoming call for processing, reducing the number of call registers effectively reduces the size of the batch of calls the switch will accept for processing in a “base-level cycle” (the basic unit of work alluded to above). However, this requires that calls not accepted for processing in the current base-level cycle be inserted in a queue, and later removed from the queue and processed, undesirably increasing overhead and further reducing real-time available for processing calls.
As noted above, SPC-based switching systems are engineered based on the assumption that traffic arrival rates are Poissonian. However, we and others have observed that calls often arrive in bursts. Although the average call arrival rate over some period including the bursts may be well under the time-averaged capacity of the switching system, the instantaneous arrival rate during the burst itself may far exceed the average capacity of the switching system.
This is shown schematically in
FIG. 1
, in which traffic load (corresponding to the vertical axis
84
) is plotted against time (corresponding to the horizontal axis
82
). Line
86
represents the maximum engineered load
86
. The smooth curve
88
represents the expected distribution of call arrivals, which approximates a Poisson distribution, over an engineered busy hour. The engineered busy hour is a one-hour period which reflects the maximum planned traffic to be offered to the switching system. The load increases gradually to the maximum engineered load
86
, and then gradually recedes. Curve
90
represents a typical scenario during a busy period, in which a burst of traffic
92
occurs. The traffic load during burst
92
substantially exceeds the maximum engineered load
86
for a short interval. However, the average load over a period including and surrounding the burst
86
remains well within the maximum engineered load.
The design of existing switching systems and their controls recognizes that non-Poisson traffic arrival rates do occur. However, existing switching systems respond to non-Poisson traffic only when it affects system performance, and, in general only when congestion occurs. Moreover, existing switching systems of which we are aware treat overloads alike, by responding to congestion as though resulting from continuous traffic demand exceeding the maximum engineered load, regardless of whether the congestion is actually the result of excessive continuous demand, or simply a product of a transient burst of high traffic. The systems tend to operate in one of two statesnormal state or overload state. Because the overload controls are set based on the assumption that traffic is Poissonian, the system's response to congestion becomes fixed and results in an artificial limit on the capacity of the switching system.
Accordingly, there exists a need for a switching system traffic control system which provides improved response t

Affiliated with

Also associated with

Rating

Predictive bursty real-time traffic control for... does not yet have a rating. At this time, there are no reviews or comments for this patent.

If you have personal experience with Predictive bursty real-time traffic control for..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Predictive bursty real-time traffic control for... will most certainly appreciate the feedback.

Rate now

Comments { 0 }

Profile ID: LFUS-PAI-O-2526399