DE10341071B3 - Overload protection, especially for call processing platforms, involves determining processor load, input value queue length, queue length change, overload level, rejecting percentage of incoming information depending on overload level - Google Patents

Abstract

The method involves determining a processor load (y), determine the length of the input value queue (F), determining a change in the queue length, determining a current overload level (OPL), detecting a high overload level if the queue length exceeds a threshold value, manipulating the overload level depending on the queue length and processor loading and rejecting a percentage of the incoming information depending on the overload level.. An independent claim is also included for the following: (a) a communications system with at least one processing platform.

Description

The The present invention relates to a method by which Help overload situations of Processing platforms can be detected and fended off. This Method can be particularly advantageous for so-called call processing Platforms are used in switching systems.

Processing platforms, who generally process messages or other information are subject in principle to the problem that the amount of incoming information can fluctuate strongly, which in turn leads to heavily fluctuating load situations the processing platforms leads. Two parameters that influence the amount of incoming information are, for example, the number of messages received per unit of time and the length the news.

Nice from economic establish processing platforms are often designed for maximum load, the (sometimes significantly) below the maximum information input occurring load is because the maximum information input in the normal operation is very rare or never occurring value. The called maximum load, for a processing platform is designed, in practice often based on the expected values for the incoming information stream in normal operation as well as based of considerations in terms of availability the services provided by the processing platform and others operational considerations established.

By However, this interpretation exists for ongoing operation certain probability that an overload situation a processing platform occurs. Complex processing platforms, For example, processing platforms that have multiple processing units or processors may have doing a certain amount of overload for certain Spanning time spans, for example by means of suitably controlled queues or load distribution mechanisms.

For such Processing platforms are then defined overload levels that the degree of overload specify. An example for such a gradation of overload is given by a system in which 7 overload levels are set, denoted by 0..6, and for which is then determined that for overload level 0 no action are required, i. no in-depth information will be averted whereas for overload level 6 all incoming information will be repelled. For the overload levels 1..5 will be repelled according to parts of the incoming information. For example, with linear distribution of the overload levels are for stage 1 16% of incoming traffic, for level 2 33%, for level 3 50%, for Level 4 66% and for level 5 83% of incoming traffic fended off.

• – Anzahl der im letzten Kontrollintervall eingegangenen Informationseinheiten,
• – Füllstand bestimmter Warteschlangen und
• – Auslastung der Prozessoren.

In known systems, for example described in EP 0 913 772 A2 . EP 216 359 A2 . DE 3328571 . DE 3328574 and DE 3328575 , Overload levels are determined by load indicators using the following load indicators:

• - number of information units received in the last control interval,
• - Level of certain queues and
• - Utilization of the processors.

One Over load defense based on these load indicators however, it tends to "swing", i. the optimal one Operating state, which is characterized by a suitable overload level and in which the utilization of the processing platform is best possible, is never reached, but the processing platform shuttles between overload and bad utilization back and forth. This problem occurs especially then on when in a system with one or more processing platforms both a central element and the individual processing platforms each separately a overload defense make.

Furthermore, it is off US 6,141,323 a device is known in which the change in the size of a queue is measured at fixed time intervals, which device can be used in a system for overload control.

It It is an object of the present invention to provide a method of overload protection avoiding a swinging of the system while ensuring that the usable information throughput possible is high. It is a further object of the invention to provide a corresponding Specify communication system.

These The object is solved by the features of the independent claims. Preferred embodiments are the dependent claims refer to.

• – Bestimmen einer Prozessorauslastung,
• – Bestimmen eines Eingabewarteschlangen-Füllstandes,
• – Bestimmen einer Veränderung des Eingabewarteschlangen-Füllstandes gegenüber des vorangehend bestimmten Eingabewarteschlangen-Füllstandes,
• – Ermitteln einer aktuellen Überlaststufe aus einem Bereich von Überlaststufen anhand von Prozessorauslastung, Eingabewarteschlangen-Füllstand und Veränderung des Eingabewarteschlangen-Füllstandes, und
• – Abweisen eines Prozentsatzes eingehender Informationen, wobei der Prozentsatz abzuweisender Informationen be stimmt wird durch die ermittelte aktuelle Überlaststufe und zwischen 0% und 100% liegt.

According to the invention, for a system having at least one processing platform, wherein the processing platform comprises at least one processor and at least one input queue, a method for the prevention of overload is provided. This comprises the following method steps:

• Determining a processor load,
• Determine an input queue fill article
• Determining a change in the input queue fill level from the previously determined input queue fill level,
• Determine a current overload level from a range of overload levels based on processor utilization, input queue level, and input queue level change, and
• - rejecting a percentage of incoming information, the percentage of information to be rejected being determined by the current overload level determined and between 0% and 100%.

• – Festlegen einer hohen aktuellen Überlaststufe, falls der Eingabewarteschlangen-Füllstand einen Eingabewarteschlangen-Überlaufschwellwert übersteigt, anderenfalls
• – Ermitteln der aktuellen Überlaststufe durch Erhöhen der bislang gültigen Überlaststufe, falls der Eingabewarteschlangen-Füllstand einen Eingabewarteschlangen-Füllstandsschwellwert übersteigt, anderenfalls
• – Ermitteln der aktuellen Überlaststufe durch Verringern der bislang gültigen Überlaststufe, falls die Veränderung des Eingabewarteschlangen-Füllstandes niedriger als ein Eingabewarteschlangen-Füllstandsänderungsschwellwert ist, anderenfalls
• – Ermitteln der aktuellen Überlaststufe durch Erhöhen der bislang gültigen Überlaststufe, falls die Prozessorauslastung größer als ein erster Prozessorauslastungs-Grenzwert ist, anderenfalls
• – Ermitteln der aktuellen Überlaststufe durch Verringern der bislang gültigen Überlaststufe, falls die Prozessorauslastung kleiner als ein zweiter Prozessorauslastungs-Grenzwert ist, anderenfalls
• – Beibehalten der bislang gültigen Überlaststufe.

According to the present invention, the current overload level is determined by the following method steps:

• - Set a high current overload level if the input queue level exceeds an input queue overflow threshold, otherwise
• - Determine the current overload level by increasing the hitherto valid overload level if the input queue level exceeds an input queue level threshold, otherwise
• Determine the current overload level by decreasing the hitherto valid overload level if the change in the input queue level is lower than an input queue level change threshold, otherwise
• - Determine the current overload level by increasing the previous overload level if the processor utilization is greater than a first processor utilization limit, otherwise
• - Determine the current overload level by reducing the previous overload level if processor utilization is less than a second processor utilization limit, otherwise
• - Maintain the current overload level.

One important advantage is the evaluation of the change the input queue level, which prevents the system from oscillating and also in case of overload one possible high throughput is achieved. The evaluation of change the input queue level allows it to determine if the already made overload protection measures Show effects. For example, has the input queue level across from the previous measurement, it can be assumed that that the measures taken at least sufficient, if not already too much information Reject.

By the method steps for determining the overload level according to dependent claim 2 is ensured accordingly that after evaluation of the absolute Input queue fill level its change is considered to the overload level adapt.

The The invention further relates to a communication system, which means to perform of the method.

One Another advantage of the invention is the fact that the processor load for determining the overload level rather than a quantity of incoming information representing Measure how about the number of incoming messages. Thus, the middle one Processing time per message irrelevant for determining the overload level, which is advantageous a more accurate detection of the overload situation for the case allows when the overload produced by a few but especially long messages. One Another advantage is that by evaluating the processor utilization the overload protection according to the invention also for one System works in which in addition to one or more processing platforms both a central element and the individual processing platforms each separately make an overload defense.

in the The following will be an embodiment of Invention explained in more detail with reference to two figures.

1 shows a block diagram of a method for the determination and treatment of overload according to an embodiment of the invention.

2 shows with further details the determination of the overload level based on the determined parameters according to an embodiment of the invention.

1 shows a block diagram of the method for the determination and treatment of overload. The sequence of different steps is merely an example, for example the order of determination of the load parameters or overload indicators. Further, not all aspects of the present invention are all in 1 required steps required. Finally, for the sake of simplicity, the illustration has been omitted from details obvious to those skilled in the art, such as managing timers to determine the end of an overload situation.

The entry into the overload treatment can in a first step 100 by, for example, checking whether an input waiting queue level F is greater than another input queue level threshold S _F or whether an integrating start indicator exceeds a start indicator threshold S _S. If one of these conditions is met, the overload treatment is entered, as explained in detail below. Not shown was the step of determining the input queue level and / or the integrating start indicator.

The integrating start indicator S can be determined, for example, according to the following recursive relationship:
S = max {0, S + y - f * y _planned }, where y is the last determined actual processor utilization, f is a scaling factor and y _{plotted is} the target processor _utilization . The factor f is chosen in the preferred embodiment as f = 1.05. The factor f ensures that overload is not detected too early unnecessarily. The start indicator provides a measure of the extent to which processor utilization y has been above the targeted value y _planned in the past past.

Other Indicators for an impending overload can Naturally also be used, either separately or in any combination with the mentioned indicators.

Is in the overload treatment in response to the fulfillment of the condition in the block 100 occurred in one step 102 set an initial overload level OPL. In the present example, as explained above, the mean value is given by OPL = 3, which causes 50% of the incoming messages to be rejected. Of course, depending on the implementation, other levels and percentages may be chosen here.

With the steps 104 to 108 begins a repetitive process to adjust the overload level, starting from the in step 102 initial or otherwise determined initial overload level. In step 104 the current processor utilization y is determined. In step 106 the current input queue level F is determined, and in step 108 the change .DELTA.F of the input queue level F is determined with respect to the preceding loop pass. For example, ΔF = 0 can be set in the first pass. The three in the steps 104 to 108 Values determined are also referred to as overload indicators.

In step 110 From the determined values, a new overload stage OPL is calculated, as in connection with 2 explained below by way of example. In order to implement the overload protection measures, it may be necessary to transfer the overload level OPL or other parameters to various elements of the system, for example to peripheral elements to effect the rejection of the percentage of messages determined by the overload level OPL, to increase the overload to prevent and to achieve a decay of the overload.

Finally, in one step 112 Checks whether the overload level has reached the lowest value, in this case the value 0, indicating that the rejection of messages is no longer necessary, reached and held for a certain time. In the example shown, it is checked whether OPL = 0 has been valid for at least 6 seconds. If yes, then the overload treatment is exited and the start conditions for overload treatment are checked in step 100 otherwise, the overload treatment is continued with the steps of determining the parameters y, F, ΔF.

It is possible to wait between two sequences of the two illustrated loops until a control interval has elapsed, steps 114 and 116 , The control interval indicates the frequency with which the overload check and the recalculation of the overload level are checked.

Not The steps for monitoring the time span are shown. in which the overload level has the lowest value. Suitable methods for this are the person skilled in the art and therefore will not be closer here explained.

2 shows the step 110 out 1 for determining the overload level from the parameters processor utilization y, input queue level F and change ΔF of the input queue level with further details.

In a first query 202 It is determined whether the input _queue level F exceeds an input _queue overflow threshold IL _of1 , which serves to protect against overflow of the input _queue and thus protection against messages / information loss. Is the condition 202 is fulfilled, in one step 204 set a high value for the overload level. In the preferred embodiment, the highest value OPL = 6 is selected, so that first by rejecting all incoming messages, the input _queue level may fall below the critical value IL _of1 . After step 204 the determination of the overload level is ended.

Is the condition, on the other hand 202 not satisfied, ie F≤IL _ofl then it is in a query 206 Checks whether the input queue level F exceeds a (less critical) input queue threshold F *. If this is the case, will in one step 208 the overload level is increased, in the present example by the value 1, and the determination of the overload level is ended.

Is the condition, on the other hand 206 not satisfied, ie F≤F *, then in a query 210 whether the change ΔF (which is calculated as a new fill level minus old fill level, this value can also be negative) is smaller than an input queue fill level change threshold IL _var . If this is the case, it will be in one step 212 the overload level is reduced by the value 1 in the present example and the determination of the overload level is ended.

Is the condition, on the other hand 210 not satisfied, that is, ΔF≥IL _var , then in a query 214 checked if the processor utilization y is greater than a first processor utilization limit. The first processor utilization limit can be chosen to y _planned + δy. Therein, y _planned denotes the desired processor utilization and δy a hysteresis in the evaluation of the processor load, which prevents the conditions 210 and 214 relatively small changes in the processor utilization y to a comparatively large change in the percentage of messages to be rejected, in the present example by ± 16% corresponding to an overload level caused. Is condition 214 is fulfilled, in one step 216 the overload level is increased, in the present example by the value 1, and the determination of the overload level is ended.

Is the condition, on the other hand 214 not satisfied, ie y≤y _planned + δy, then it is in a query 218 checks to see if processor utilization y is less than a second processor utilization limit. The second processor utilization limit can be chosen to y _planned -δy. Is condition 218 is fulfilled, in one step 220 the overload level is reduced by the value 1 in the present example and the determination of the overload level is ended.

If none of the above conditions is met, the example remains 2 the overload level unchanged.

From the point of view of the present invention it is incidental, with what frequency, for example indicated by the control interval in the optional step 114 , the loop for determining the current overload level is executed. This frequency depends, for example, on how strong the traffic sources connected to the processing platforms are, whether their transmission rate predominantly corresponds to the average transmission rate, or whether the traffic sources transmit in a bursty manner and how high the processing rate of the platform (s) is. In the considered systems with quite high message throughput, it can be assumed that the overload level is updated several times per second.

moreover It should be noted that it for the present invention is irrelevant, whether the control interval, ie the time span between two consecutive cycles Determination of the overload level, always the same length. Rather, they are scheduled or random fluctuations the control interval length the overload defense do not negatively influence.

The exact determination of the parameters S _F , S _S , F, IL of _I , IL _var , F *. For example, y _planned and δy may be hit depending on processor speed, average message processing time, and other system parameters.

• – Je größer die Schwellwerte S_F und S_S sind, um so seltener wird die Überlastabwehr aktiviert, wenn nur kurze Lastspitzen auftreten, jedoch dauert es um so länger, bis eine Überlastsituation als solche erkannt wird. S_F sollte kleiner als F* gewählt werden, z.B. S_F = 0,5·F*. S_S kann so gewählt werden, daß bei voller Auslastung des Prozessors die Erkennung auf Überlast durch den Startindikator S erst nach mehreren Kontrollintervallen erfolgt.
• – Sinnvolle Werte für Faktor f liegen beispielsweise zwischen 1% und 10%.
• – Bezeichnet IL_lenght die maximale Länge der Eingabewarteschlange, so ist z.B. IL_ofl = 0,8·IL_lenght ein geeigneter Wert.
• – y_planned kann für sehr schnellen Prozessoren mit einem großen Nachrichtendurchsatz nahe bei 100% liegen (z.B. größer oder gleich 90%)
• – Die Wahl von y_planned beeinflußt δy, da y_planned + δy < 100% gelten muß. Für y_planned = 90% könnte δy = 5% gewählt werden.

Simulations can be used to define and optimize these parameters. The following considerations may help in finding the optimal parameter values:

• The greater the threshold values S _F and S _S , the less frequently the overload protection is activated when only short peak loads occur, but it takes the longer until an overload situation is recognized as such. S _F should be smaller than F *, eg S _F = 0.5 · F *. S _S can be chosen so that when full utilization of the processor, the detection of overload by the start indicator S only after several control intervals.
• - Meaningful values for factor f are, for example, between 1% and 10%.
• - If IL _means the maximum length of the input _queue , then IL _ofl = 0.8 · IL _is an appropriate value.
• - y _planned can be close to 100% for very fast processors with a high message throughput (eg greater than or equal to 90%)
• - The choice of y _planned affects δy, since y _planned + δy <100% must apply. For y _planned = 90%, δy = 5% could be chosen.

• – Um zu verhindern, daß die Eingabewarteschlange zu groß wird und damit große Verzögerungen auftreten, sollte IL_var nicht zu groß gewählt werden. Ist dieser Parameter andererseits zu klein, dann wird die Überlaststufe OPL erst sehr spät wieder verringert, was zu einem Leerlaufen der Eingabewarteschlage und anschließend zu einem schlecht ausgenutzten Prozessor führt. Das System würde so ins Schwingen geraten. Ein geeigneter Wert für IL_var ist z.B. gegeben durch IL_var = 0,05·m_cap. Damit würde die Überlaststufe verringert, wenn die Eingabewarteschlage um höchstens 5% der in einem Kontrollintervall bearbeitbaren Nachrichten wächst. Damit werden wieder frühzeitig neue Anfragen/Nachrichten bearbeitet.
• – F* sollte im Bereich von einem kleinen Vielfachen von m_cap liegen. Damit wird die Überlaststufe erhöht, wenn die Nachrichten in der Eingabewarteschlage nicht mehr in einem kleinen Vielfachen der Kontrollintervall-Länge abgearbeitet werden kann. Als Formel ausgedrückt, F* = k·m_cap, worin k ein kleiner ganzzahliger Wert ist, beispielsweise k = 2 oder k = 3.

An additional quantity can be used to select the appropriate values for the remaining parameters: the mean number m _cap of messages that can be processed in a control interval.

• - In order to prevent the input queue from becoming too large and causing long delays, IL _var should not be too large. On the other hand, if this parameter is too small, then the overload stage OPL is reduced again very late, which leads to an idling of the input queue and then to a poorly utilized processor. The system would start to oscillate like that. A suitable value for IL _var is for example given by IL _var = 0.05 * m _cap . This would reduce the overload level if the input queue fails to process at most 5% of the one in a control interval growing news. This will be processed again early new requests / messages.
• F * should be in the range of a small multiple of m _cap . This increases the overload level if the messages in the input queue can no longer be processed in a small multiple of the control interval length. Expressed as a formula, F * = k * _mcap , where k is a small integer value, for example k = 2 or k = 3.

• – die Überlaststufe OPL reduziert wird, wenn der Eingabewarteschlangen-Füllstand F einen bestimmten Wert unterschreitet bzw.
• – die Überlaststufe OPL erhöht wird, wenn die Änderung ΔF des Eingabewarteschlangen-Füllstands F einen bestimmten Wert überschreitet.

In one embodiment, a further step (not shown) can be provided to determine the overload level, accordingly additionally

• - The overload level OPL is reduced when the input queue level F falls below a certain value or
• - The overload level OPL is increased when the change .DELTA.F of the input queue level F exceeds a certain value.

F

Input queue level

S _F

Input queue level threshold

S

integral start indicator

S _S

Start indicator threshold

y

processor utilization

f

scaling factor

y _planned

sought Processor utilization (in percent)

OPL

Overload level (Overload Priority Level)

.DELTA.F

change of the input queue level F

IL _ofl

Input queue Überlaufschwellwert

F *

Input queues threshold

IL _var

Input queue level change

threshold

y _planned + δy

first Processor Auslastungs limit

Dy

hysteresis in the evaluation of the processor load y

y _planned -δy

second Processor Auslastungs limit

IL _lenght

maximum Length of input queue

m _cap

the mean number of in a control interval

editable news

k

integer factor

Claims (8)

Procedure for the defense against overload for a system which has at least one processing platform, wherein the Processing platform at least one processor and at least one Input queue, with the following steps: - Determine a processor load, Determining an input queue level, - Determine a change the input queue level across from the previously determined input queue level, - Determine a current overload level from a range of overload levels based on processor utilization, input queue level and change the input queue level by: - Establish a high current overload level, if the input queue level exceeds an input queue overflow threshold, otherwise - Determine the current overload level by elevating the currently valid overload level, if the input queue level exceeds an input queue level threshold, otherwise - Determine the current overload level by reducing the current overload level, if the change lower the input queue level as an input queue level change threshold is, otherwise - Determine the current overload level by elevating the currently valid overload level, if the processor load is greater than a first processor utilization limit, otherwise - Determine the current overload level by reducing the current overload level, if the processor utilization is less than a second processor utilization limit is, otherwise - Maintain the currently valid overload level, and - reject a percentage of incoming information, the percentage to be rejected is determined by the determined current overload level and between 0% and 100. The method of claim 1, thus the first processor utilization limit bigger than that second processor utilization limit are chosen so that both Values by the same amount of actual processor utilization differ. Method according to one of claims 1 to 2, characterized that in an initial step, an initial overload level and set in the overload defense is entered when the input queue level exceeds another input queue level threshold or an integrating start indicator that indicates the extent to which the Processor utilization in the past above the actual target Processor load has exceeded a start indicator threshold. Method according to claim 3, characterized that for the initial overload level a middle is selected from the range of overload levels. Method according to one of claims 1 to 4, characterized that in a last step, the overload defense will leave, if over a predetermined period constant the lowest overload level was determined. Communication system, which at least one processing platform wherein the processing platform comprises at least one processor and at least one input queue, with the following Funds for the defense against overload: - Medium for determining a processor load, - means for determining a Input queue filling level, - Medium to determine a change the input queue level across from the previously determined input queue level, - Medium to determine a current overload level a range of overload levels based on processor utilization, input queue level and change the input queue level With: - funds to set a high current overload level if the input queue level exceeds an input queue overflow threshold, - funds to determine the current overload level Increase the hitherto valid overload level, if the input queue level exceeds an input queue level threshold, - funds to determine the current overload level Reduction of the current overload level, if the change the input queue level is lower than an input queue level change threshold, - funds to determine the current overload level Increase the hitherto valid overload level, if the processor utilization is greater than is a first processor utilization limit, and - funds to determine the current overload level Reduction of the current overload level, if the processor utilization is less than and - Medium for rejecting a percentage of incoming information, wherein the percentage of information to be rejected is determined by the determined current overload level and between 0% and 100%. Communication system according to claim 6, characterized by means for defining an initial overload level and causing it the entry into the overload defense, when the input queue level exceeds one or more input queue level thresholds an integrating startup indicator that indicates the extent to which the processor utilization is in the past about the actual targeted processor utilization exceeds a start indicator threshold. Communication system according to one of claims 6 or 7, characterized by means for causing the abandonment of the overload defense when over one predetermined period constant the lowest overload level was determined.