Method in a polling system for transmitting queue elements from multiple input queues to a single output with improved queue service performance5623668Abstract Method for the polling of queues. A ratio table is derived from, for example, the numbers of elements in the queues at any one moment. The sum S is calculated of the ratio values R(x) to the various queues. Next, a correction factor C(x) is assigned to each queue, which correction factor is equal to S, decreased by the ratio value of that queue. Further an urgency factor U(x) is assigned to each queue. The queue to be polled next is, in each case, the queue with the highest value of U(x) or, for example where values of U(x) are equal, the first occurring queue with that value. Thereafter the urgency factor U(x) of the selected queue is reduced by the value of its correction factor C(x), whilst the urgency factor U(x) of the remaining queues is increased by their ratio value. The queue to be polled next is then again the queue with the highest value of U(x), and so forth. The method is pre-eminently applicable for ATM, eg. for the polling of subscriber terminals in a passive optical network. Claims I claim: Description A. BACKGROUND TO THE INVENTION
______________________________________
Queue names a b c
Queue values
Q(a) = 3 Q(b) = 6 Q(c) = 3
Ratio factor
R = 1
Ratio values
R(a) = 3 R(b) = 6 R(c) = 3
Sum ratio values
S = 12
Correction values
C(a) = 9 C(b) = 6 C(c) = 9
P U(a) U(b) U(c) Selected queue (i)
1 0* 0 0 a
2 -9 6* 3 b
3 -6 0 6* c
4 -3 6* -3 b
5 0* 0 0 a
6 -9 6* 3 b
7 -6 0 6* c
8 -3 6* -3 b
9 0* 0 0 a
10 -9 6* 3 b
11 -6 0 6* c
12 -3 6* -3 b
13 0* 0 0 a
14 -9 6* 3 b
15 -6 0 6* c
______________________________________
With a ratio distribution of 3 - 6 - 3 the calculated polling sequence is therefore a - b - c - b - a - b - c - b - a - etc. (the highest value of U(x) is indicated by an *). It may be obvious that at the moment the queue values measured on the queues change, the polling sequence may also change; in that case the values of R(a), R(b) and R(c) therefore change as well, as do the values of C(a), C(b) and C(c). Such a change can, directly prior to each calculation cycle P, wherein it is calculated which queue is to be used as the next queue, be included in that calculation so that the polling sequence is always optimally adjusted to the incoming flow of queue elements, such as ATM data cells. Changes may also be included in the calculation once every n calculation cycles. This method thus seems very well suited for application in ATM systems, as has already been confirmed by simulations. The FIGURE depicts flow diagram 100 for carrying out the process of routing queue elements from multiple input queues to a single output of a polling system. With reference to the FIGURE, processing block 101 depicts that each queue has a queue value (Q(i)) determined from characteristics or parameters of the queue elements forming the queue. First, processing block 110 is invoked so as to assign urgency integers U(i) to the queues based upon a pre-selected criterion or pre-selected criteria. As the example demonstrated, the U(i)'s may all be set to the same value if no prior knowledge is available about the relative importance of the queues. Next, processing block 120 is invoked to calculate the ratio integers R(i) in proportion to the queue values Q(i). Then, the processing by block 130 is carried out to calculate the sum S of all the ratio integers. In turn, correction integers C(i) for each queue are computed from S and R(i), as depicted by processing block 140. Once these calculations are complete, it is now possible to select a queue, designated queue j, which has the highest urgency integer U(j) associated with queue j, as indicated by processing block 150. Processing block 160 is now invoked to poll the selected queue to transmit the queue elements from queue j to the single output of the polling system; in this way, the queue having the highest priority is serviced to thereby enhance the performance of the polling system. Once polling has taken place, updating activities occur. Updated first, as evidence by processing block 170, is the U(j) integer which is reduced by the correction integer C(j). Next, all other urgency integers, as per processing block 180, are increased in value by adding the ratio integer R(i) to U(i), i.noteq.j. Finally, processing block 190 is invoked to return to an intermediate block (block 150 or block 120, as exemplified by examples) in flow diagram 100 depending upon the dynamic changes in the queue values. Some further examples will be given below in which use is made of the following program.
__________________________________________________________________________
100
CLS
110
OPEN"pn.out" FOR APPEND AS#1
120
OPEN"pn.in" FOR INPUT AS#2
130
PRINT "Queue names a b c"
140
PRINT#1, "Queue names a b c"
150
INPUT#2,Q(1),Q(2),Q(3),R
160
IF (Q(1) = Q(1)) AND (Q(2) = Q(2)) AND (Q(3) = Q(3)) GOTO 420
170
Q(1) = Q(1):Q(2) = Q(2):Q(3) = Q(3)
180
PRINT" ":PRINT#1," "
190
PRINT USING "Queue values Q(a) = ### Q(b) = ### Q(c)
= ###";Q(1);Q(2);Q(3)
200
PRINT#1,USING "Queue values Q(a) = ### Q(b) = ###
Q(c) = ###";Q(1);Q(2);Q(3)
210
PRINT USING "Ratio factor R = ###";R
220
PRINT#1,USING "Ratio factor R = ###";R
230
S = 0
240
FOR X = 1 TO 3
250
R(X) = INT((Q(X)/R) + .5)
260
S = S + R(X)
270
NEXT X
280
FOR X = 1 TO 3
290
C(X) = S-R(X)
300
NEXT X
310
PRINT USING "Ratio values R(a) = ### R(b) = ### R(c)
= ###";R(1),R(2),R(3)
320
R(c) = ###";R(1),R(2),R(3)lues R(a) = ### R(b) = ###
330
PRINT USING "Sum ratio values S = ###";S
340
PRINT#1,USING "Sum ratio values S = ###";S
350
PRINT USING "Correction values C(a) = ### C(b) = ### C(c)
= ###";C(1),C(2),C(3)
360
PRINT#1,USING "Correction values C(a) = ### C(b) = ### C(c) = ###";
C(1),C(2),C(3)
370
PRINT" "
380
PRINT#1," "
390
PRINT " P U(a) U(b) U(c) Selected- queue (i)
400
PRINT#1, " P U(a) U(b) U(c) Selected-queue (i)
410
PRINT#1," "
420
FOR P = 1 TO 3
430
IF (U(1)> =U(2)) AND (U(1)> =U(3)) THEN Q$ = "a":M$(1) = "*":
M$(2) = " ":M$(3) = " ":GOTO 460
440
IF (U(2)> =U(3)) AND (U(2)> =U(1)) THEN Q$ = "b":M$(1) = "
":M$(2) = "*":M$(3) = " ":GOTO 460
450
IF (U(3)> = U(1)) AND (U(3)> =U(2)) THEN Q$ = "c":M$(1) = " ":M$(2) =
"
":M$(3) = "*":GOTO 460
460
N = N + 1
470
PRINT USING "### ###& ###& ###& &";N,U(1),M$(1),U(2),
M$(2),U(3),M$(3),Q$
480
PRINT#1,USING "### ###& ###& ###& &";N,U(1),M$(1),
U(2),M$(2),U(3),M$(3),Q$
490
IF Q$ = "a" THEN U(1) = U(1)-C(1):U(2) = U(2) + R(2):U(3) = U(3) +
R(3)
500
IF Q$ = "b" THEN U(2) = U(2)-C(2):U(1) = U(1) + R(1):U(3) = U(3) +
R(3)
510
IF Q$ = "c" THEN U(3) = U(3)-C(3):U(1) = U(1) + R(1):U(2) = U(2) +
R(2)
520
NEXT P
530
IF EOF(2) GOTO 550
540
GOTO 150
550
CLOSE
560
SYSTEM
__________________________________________________________________________
In the program above, a check is carried out, after every third calculation cycle in which a queue is guided, whether any change has occurred in the queues for example a change in the number of elements located in the queue, (or for example a change in the net increase/decrease). In the example above no change occurred; the queue values therefore remained constant in the period considered, (the new values are presented only when changes occur). Below, firstly, a situation in which the queue values are greater than in the above example by a factor of 100.
______________________________________
Queue names a b c
Queue values
Q(a) = 286
Q(b) = 598
Q(c) = 326
Ratio factor
R = 100
Ratio values
R(a) = 3 R(b) = 6 R(c) = 3
Sum ratio values
S = 12
Correction values
C(a) = 9 C(b) = 6 C(c) = 9
P U(a) U(b) U(c) Selected queue (i)
1 0* 0 0 a
2 -9 6* 3 b
3 -6 0 6* c
4 -3 6* -3 b
5 0* 0 0 a
6 -9 6* 3 b
7 -6 0 6* c
8 -3 6* -3 b
9 0* 0 0 a
10 -9 6* 3 b
11 -6 0 6* c
12 -3 6* -3 b
13 0* 0 0 a
14 -9 6* 3 b
15 -6 0 6* c
______________________________________
By dividing the queue values by a ratio factor of 100 and using the nearest integers as valves for the ratio table, the same situation is created as in the first example. Owing to the fairly large ratio factor a slight inaccuracy is introduced in the queue selection as may be seen from the example below, in which a smaller ratio factor is used thereby making the selection more accurate.
______________________________________
Queue names a b c
Queue values
Q(a) = 286
Q(b) = 598
Q(c) = 326
Ratio factor
R = 10
Ratio values
R(a) = 29 R(b) = 60 R(c) = 33
Sum ratio values
S = 122
Correction values
C(a) = 93 C(b) = 62 C(c) = 89
P U(a) U(b) U(c) Selected queue (i)
1 0* 0 0 a
2 -93 60* 33 b
3 -64 -2 66* c
4 -35 58* -23 b
5 -6 -4 10* c
6 23 56* -79 b
7 52* -6 -46 a
8 -41 54* -13 b
9 -12 -8 20* c
10 17 52* -69 b
11 46* -10 -36 a
12 -47 50* -3 b
13 -18 -12 30* c
14 11 48* -59 b
15 40* -14 -26 a
______________________________________
An example is given below of a situation in which the queue values change rapidly. The changed situation (which is indicated in each case), is taken into account on every third calculation cycle.
______________________________________
Queue names a b c
Queue values
Q(a) = 286
Q(b) = 598
Q(c) = 326
Ratio factor
R = 10
Ratio values
R(a) = 29 R(b) = 60 R(c) = 33
Sum ratio values
S = 122
Correction values
C(a) = 93 C(b) = 62 C(c) = 89
P U(a) U(b) U(c) Selected queue (i)
1 0* 0 0 a
2 -93 60* 33 b
3 -64 -2 66* c
Queue values
Q(a) = 286
Q(b) = 635
Q(c) = 326
Ratio factor
R = 10
Ratio values
R(a) = 29 R(b) = 64 R(c) = 33
Sum ratio values
S = 126
Correction values
C(a) = 97 C(b) = 62 C(c) = 93
P U(a) U(b) U(c) Selected queue (i)
4 -35 58* -23 b
5 -6 -4 10* c
6 23 60* -83 b
Queue values
Q(a) = 234
Q(b) = 635
Q(c) = 326
Ratio factor
R = 10
Ratio values
R(a) = 23 R(b) = 64 R(c) = 33
Sum ratio values
S = 120
Correction values
C(a) = 97 C(b) = 56 C(c) = 87
P U(a) U(b) U(c) Selected queue (i)
7 52* -2 -50 a
8 -45 62* -17 b
9 -22 6 16* c
Queue values
Q(a) = 234
Q(b) = 635
Q(c) = 376
Ratio factor
R = 10
Ratio values
R(a) = 23 R(b) = 64 R(c) = 38
Sum ratio values
S = 125
Correction values
C(a) = 102
C(b) = 61 C(c) = 87
P U(a) U(b) U(c) Selected queue (i)
10 1 70* -71 b
11 24* 9 -33 a
12 -78 73* 5 b
Queue values
Q(a) = 198
Q(b) = 635
Q(c) = 376
Ratio factor
R = 10
Ratio values
R(a) = 20 R(b) = 64 R(c) = 38
Sum ratio values
S = 122
Correction values
C(a) = 102
C(b) = 58 C(c) = 84
P U(a) U(b) U(c) Selected queue (i)
13 -55 12 43* c
14 -35 76* -41 b
15 -15 18* -3 b
______________________________________
It will be obvious that it is equally possible to check whether a change has occurred in the queue values within the queues on every calculation cycle, rather than on every third calculation cycle. This is assumed in the example below, in which prior to each queue selection, any change in the queue values with respect to the previous selection is detected; only in that case are the changed queue values presented and the values for R(x), S and C(x) recalculated. The progress of queue values in time is as follows:
______________________________________
1 286 - 598 - 326
326 286 - 635
635 - 326
4 234 - 635 - 376
635 - 376
6 198 - 635 - 376
7 198 - 635 - 376
8 198 - 635 - 376
467 255
10 255 - 698 - 467
501 287
12 287 - 751 - 501
578 302
14 302 - 923 - 578
15 302 - 923 - 578
______________________________________
The polling sequence table calculation then looks as follows:
______________________________________
Queue names a b c
Queue values
Q(a) = 286
Q(b) = 598
Q(c) = 326
Ratio factor
R = 10
Ratio values
R(a) = 29 R(b) = 60 R(c) = 33
Sum ratio values
S = 122
Correction values
C(a) = 93 C(b) = 62 C(c) = 89
P U(a) U(b) U(c) Selected queue (i)
1 0* 0 0 a
Queue values
Q(a) = 286
Q(b) = 635
Q(c) = 326
Ratio factor
R = 10
Ratio values
R(a) = 29 R(b) = 64 R(c) = 33
Sum ratio values
S = 126
Correction values
C(a) = 97 C(b) = 62 C(c) = 93
P U(a) U(b) U(c) Selected queue (i)
2 -93 60* 33 b
Queue values
Q(a) = 234
Q(b) = 635
Q(c) = 326
Ratio factor
R = 10
Ratio values
R(a) = 23 R(b) = 64 R(c) = 33
Sum ratio values
S = 120
Correction values
C(a) = 97 C(b) = 56 C(c) = 87
P U(a) U(b) U(c) Selected queue (i)
3 -64 -2 66* c
Queue values
Q(a) = 234
Q(b) = 635
Q(c) = 376
Ratio factor
R = 10
Ratio values
R(a) = 23 R(b) = 64 R(c) = 38
Sum ratio values
S = 125
Correction values
C(a) = 102
C(b) = 61 C(c) = 87
P U(a) U(b) U(c) Selected queue (i)
4 -41 62* -21 b
Queue values
Q(a) = 198
Q(b) = 635
Q(c) = 376
Ratio factor
R = 10
Ratio values
R(a) = 20 R(b) = 64 R(c) = 38
Sum ratio values
S = 122
Correction values
C(a) = 102
C(b) = 58 C(c) = 84
P U(a) U(b) U(c) Selected queue (i)
5 -18 1 17* c
6 2 65* -67 b
7 22* 7 -29 a
8 -80 71* 9 b
Queue values
Q(a) = 255
Q(b) = 698
Q(c) = 467
Ratio factor
R = 10
Ratio values
R(a) = 26 R(b) = 70 R(c) = 47
Sum ratio values
S = 143
Correction values
C(a) = 117
C(b) = 73 C(c) = 96
P U(a) U(b) U(c) Selected queue (i)
9 -60 13 47* c
10 -34 83* -49 b
Queue values
Q(a) = 287
Q(b) = 751
Q(c) = 501
Ratio factor
R = 10
Ratio values
R(a) = 29 R(b) = 75 R(c) = 50
Sum ratio values
S = 154
Correction values
C(a) = 125
C(b) = 79 C(c) = 104
P U(a) U(b) U(c) Selected queue (i)
11 -8 10* -2 b
12 21 -69 48* c
Queue values
Q(a) = 302
Q(b) = 923
Q(c) = 578
Ratio factor
R = 10
Ratio values
R(a) = 30 R(b) = 92 R(c) = 58
Sum ratio values
S = 180
Correction values
C(a) = 150
C(b) = 88 C(c) = 122
P U(a) U(b) U(c) Selected queue (i)
13 50* 6 -56 a
14 -100 98* 2 b
15 -70 10 60* c
______________________________________
Presented in shorter form:
______________________________________
Calculation Cycle
Queue values
Selected queue
______________________________________
1 286 - 598 - 326
a
326 286 - 635
b
635 - 326 234
c
4 234 - 635 - 376
b
635 - 376 198
c
6 198 - 635 - 376
b
7 198 - 635 - 376
a
8 198 - 635 - 376
b
467 255
c
10 255 - 698 - 467
b
501 287
b
12 287 - 751 - 501
c
578 302
a
14 302 - 923 - 578
b
15 302 - 923 - 578
c
______________________________________
For simplicity three queues are assumed in the above examples. It will be obvious that this number may be increased in a simple manner. One important application for the method according to the invention is to be found in systems in which it must be possible to adjust the service allocation quickly to a changing service requirement. One example of a system in which a rapidly changing queue service requirement occurs is the ATM Passive Optical Network, in which multiple users are connected to one local exchange by means of a glass fibre branched like a tree. Here, the user queues with ATM data cells can be efficiently read out making use of the method presented above. If the polling table is continuously adjusted to the service requirement, the choice of the initial values of U(a), U(b) and U(c) is not critical. Where, however, the polling table is used in a semi-static way, the initial values are more importance. Practical experience has shown that a good solution is found if the initial values used are the sum S of the ratio values minus the whole number portion of the quotient of this sum S and the ratio value R(a), R(b) and R(c) respectively. In the last example the following hold good: Ratio values R(a)=29 R(b)=60 R(c)=33 Sum ratio values S=122 If the above rule is used, the initial values of U(a,b,c) become: U.sub.i (a)=S-(S DIV R(a))=122-(122 DIV 29)=122-4=118 U.sub.i (b)=S-(S DIV R(b))=122-(122 DIV 60)=122-2=120 U.sub.i (c)=S-(S DIV R(c))=122-(122 DIV 33)=122-3=119 The polling table calculation result then becomes (starting from the same queue values above):
______________________________________
Queue names a b c
Queue values
Q(a) = 286
Q(b) = 598
Q(c) = 326
Ratio factor
R = 10
Ratio values
R(a) = 29 R(b) = 60 R(c) = 33
Sum ratio values
S = 122
Correction values
C(a) = 93 C(b) = 62 C(c) = 89
P U(a) U(b) U(c) Selected queue (i)
1 118 120* 119 b
Queue values
Q(a) = 286
Q(b) = 635
Q(c) = 326
Ratio factor
R = 10
Ratio values
R(a) = 29 R(b) = 64 R(c) = 33
Sum ratio values
S = 126
Correction values
C(a) = 97 C(b) = 62 C(c) = 93
P U(a) U(b) U(c) Selected queue (i)
2 147 58 152* c
Queue values
Q(a) = 234
Q(b) = 635
Q(c) = 326
Ratio factor
R = 10
Ratio values
R(a) = 23 R(b) = 64 R(c) = 33
Sum ratio values
S = 120
Correction values
C(a) = 97 C(b) = 56 C(c) = 87
P U(a) U(b) U(c) Selected queue (i)
3 176* 122 59 a
Queue values
Q(a) = 234
Q(b) = 635
Q(c) = 376
Ratio factor
R = 10
Ratio values
R(a) = 23 R(b) = 64 R(c) = 38
Sum ratio values
S = 125
Correction values
C(a) = 102
C(b) = 61 C(c) = 87
P U(a) U(b) U(c) Selected queue (i)
4 79 186* 92 b
Queue values
Q(a) = 198
Q(b) = 635
Q(c) = 376
Ratio factor
R = 10
Ratio values
R(a) = 20 R(b) = 64 R(c) = 38
Sum ratio values
S = 122
Correction values
C(a) = 102
C(b) = 58 C(c) = 84
P U(a) U(b) U(c) Selected queue (i)
5 102 125 130* c
6 122 189* 46 b
7 142* 131 84 a
8 40 195* 122 b
Queue values
Q(a) = 255
Q(b) = 698
Q(c) = 467
Ratio factor
R = 10
Ratio values
R(a) = 26 R(b) = 70 R(c) = 47
Sum ratio values
S = 143
Correction values
C(a) = 117
C(b) = 73 C(c) = 96
P U(a) U(b) U(c) Selected queue (i)
9 60 137 160* c
10 86 207* 64 b
Queue values
Q(a) = 287
Q(b) = 751
Q(c) = 501
Ratio factor
R = 10
Ratio values
R(a) = 29 R(b) = 75 R(c) = 50
Sum ratio values
S = 154
Correction values
C(a) = 125
C(b) = 79 C(c) = 104
P U(a) U(b) U(c) Selected queue (i)
11 112 134* 111 b
12 141 55 161* c
Queue values
Q(a) = 302
Q(b) = 923
Q(c) = 578
Ratio factor
R = 10
Ratio values
R(a) = 30 R(b) = 92 R(c) = 58
Sum ratio values
S = 180
Correction values
C(a) = 150
C(b) = 88 C(c) = 122
P U(a) U(b) U(c) Selected queue (i)
13 170* 130 57 a
14 20 222* 115 b
15 50 134 173* c
______________________________________
As may be seen, a change to the initial values affects only the result of the first three calculation cycles (P=1 . . . 3) A comment such as that made with respect to the choice of the initial values of the urgency factors U(a,b,c) may also be made in relation to the choice of the ratio factor R. A ratio factor R=1 gives the most accurate result; the (measured) queue values are then also the ratio values to be used for the calculation. If, however, the queue values are extremely large and in particular if the nature of the polling table is static, and must therefore (during a period) be stored, for example, in a RAM (with limited buffer capacity), a reduced representation, represented by the ratio values of queue values must be made using a ratio factor >1. An additional problem here is that the maximum transmission speed may not be exceeded (if use is made of queue values which represent the supply of queue elements per time unit). The following is a good solution for the calculation of the ratio values. Assume the following queue values (supply of queue elements per time unit in kbit/sec): 70000, 19000, 18500, 22333, and 8000. The total band width required for this is 137833 kbit/sec. Assume that the maximum transmission band width is 140000 kbit/sec. An initial solution is to carry out an integer division of the remaining band width of 140000-137833=2167 by the number of queues (5): 2167 DIV 5=433, and subsequently to divide the various queue growth values by that quotient: 70000 DIV 433=161; 19000 DIV 433=43; 18500 DIV 433=42; 22333 DIV 433=51 and 8000 DIV 433=18. Preferably, these ratio values are then each incremented by 1, making use of almost the entire transmission capacity of 140000 kbit/sec; the ratio values therefore then become 162, 44, 43, 52 and 19. It should be pointed out that the ratio factor for different queues has a somewhat different value, namely in this case 70000/162=432, 19000/44=431, 18500/43=430, 22333/52=429 and 8000/19=421. Another solution for the calculation of the optimal reduced representation of the queue values operates as follows. Step 1: add 1 to the odd queue values; these values then become 70000, 19000, 18500, 22333+1=22334 and 8000; the remaining band width now becomes 2167-1=2166. Step 2: divide the queue values and the remaining band width by 2 and add 1 to the resulting odd values; the new values then become 35000, 9500, 9250, 11167+1=11168, 4000; the new remainder is 1082 (2166/2=1083, from which 1 is subtracted to increase the value of 11167 by 1). Step 2 is repeated until the remaining band width is minimal, that is to say, if step 2 were repeated once more, then the remaining band width would become negative. In this case the following ratio values are the result of repeating step 2 8 times: 137, 38, 37, 44 and 16; the remaining band width is 1 (kbit/sec). The above methods for calculating optimal ratio values from the queue values, which ratio values in turn then form the input for the calculation of the optimal polling sequence, achieve maximum use of the (total) band width--thus maximising the transmission speed of the queue elements--without exceeding that band width. In this way, the aspects discussed ensure that the queue elements from the various queues are read out (polled) in a regular sequence and sufficiently frequently, (that is to say so frequently that the queue does not overflow), and that transmission speed of these elements, eg. of ATM cells from subscribers in a passive optical subscriber network, is maximised, without however allowing the permissible transmission speed (bit rate) to be exceeded.
|
Same subclass Same class Consider this |
||||||||||