JP3105289B2 - Concentration method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention relates to an collecting line system, a fixed length Paquet such cells coming in particular transmitted from a plurality of input links
FIFO (Fifi) to send the packets to a single output link
The rst In FirstOut) buffer is provided for each input link
On by that line concentrator system to the access control of the buffer in the collection-ray device.

In recent years, research on packet networks such as ATM networks has been actively promoted, and their practical use is expected. In a packet network , a single link is shared by a plurality of calls (subscribers and fixed-length packets input from the link) and multiplexed for use. All data stored in the buffer of each input link is used. Fixed length of
It is required to transmit a packet to an output link, and it is desired to realize a line consolidation method as efficient as possible while preventing the occurrence of fixed-length packet discard (loss).

[0003]

2. Description of the Related Art FIG. 11 is an explanatory view of a conventional example. The configuration shown in FIG. 11 is an example called a round robin.
ATM cells (fixed-length packets) arriving from each of the input links 1 to n are stored in buffers 1 to n respectively constituted by FIFOs. Each of these buffers is provided with a reading section 1 to a reading section n.
A token ring for sequentially connecting these read units is provided. When a token arrives, each read unit reads all ATM cells stored in the corresponding buffer.

[0004] The read ATM cells are sent to a single output link via a selector. When the reading operation is completed in one reading unit, a token is sent to the reading unit connected to the next stage of the token ring, and the same reading operation is performed. Thus, the reading unit 1 and the reading unit 2
.. And the buffer read control is performed by the round robin method.

[0005]

According to the above-mentioned conventional method, when the amount of cells stored in a buffer is large, the reading time in that buffer becomes long, and the amount of cells stored in another buffer becomes excessive. In such a case, there is a problem that a cell is discarded.
Further, even when there is an empty buffer, when the turn comes by the token, the reading unit is driven, and there is a problem that a useless processing time is generated.

According to the present invention, there is provided a line concentrating method capable of reading out data at a frequency corresponding to an input fixed length packet amount when concentrating a plurality of fixed length packets on a single output link via a buffer and outputting the same. The purpose is to do.

[0007]

FIG. 1 is a basic configuration diagram of the present invention. In FIG. 1, reference numeral 1 denotes a required bandwidth calculating unit, 2 denotes a pattern determining unit, 3 denotes a holding output unit, 4-1 to 4-n correspond to a plurality of input links, and 5-1 to 5n correspond to each input link. A buffer constituted by a provided FIFO, 6 represents a selector, and 7 represents an output link.

According to the present invention, a bandwidth is calculated based on a required bandwidth of a call corresponding to each input link, an access pattern is determined so as to be most frequent according to the bandwidth of each input link, and the access pattern is determined. To read the buffer.

[0009]

The requested bandwidth calculation unit 1 performs the processing for each of the input links 4-1 to 4-4.
-Calculate the bandwidth requested (or declared) by the n users (subscribers). Next, the pattern determination unit 2 determines that the ratio of the frequency of accessing each of the buffers 5-1 to 5n is equal to the calculated ratio of the bandwidth of each of the input links 4-1 to 4-n.
An access pattern consisting of an access frequency and an order is determined so that the read access interval of each buffer is as constant as possible. The determined access pattern is input to the holding output unit 3. The holding output unit 3 holds the access patterns and also stores each of the buffers 5-1 to 5 according to the patterns.
An output is generated to any one of −n and accessed, and a read operation in the corresponding buffer is executed. The fixed-length packets read from each buffer are output to output link 7 via selector 6.

The determination of an access pattern in the above-mentioned pattern determination section 2 is constituted by the following first to third determination means. The first deciding means 20 is a means for deciding the order of the access pattern. First, the two buffers having the largest required bandwidth are selected and passed to the second deciding means 21 to determine the access pattern. Thereafter, the order of access timing determination is determined such that a buffer of the next largest request is selected based on the pattern, and the timing of accessing the buffer is embedded in the already determined access pattern.

[0011] The second determining means 21 determines an access pattern between the two buffers. That is, assuming that two buffers are i and j, their required bandwidths are integers A and B.
When (A> B) is approximated and the quotient M of A / B and the remainder N are obtained, the access pattern in that case is determined as follows.

After accessing the buffer i M + 1 times, the buffer j is accessed once, and after accessing the buffer i M times, the buffer j is accessed once. Of the B accesses of the buffer j, N times are followed. (B-
N) times follow.

The third deciding means 22 is means for repeatedly using the second deciding means 21. The third deciding means 22 calculates the sum of the required bandwidth of the buffer for which the access pattern has already been decided, and the first deciding means. The required bandwidth of the buffer for synthesizing the next pattern in the order according to 20 is B, and the second determining means 21 is repeatedly used to embed access patterns one after another in descending order of the required bandwidth.

When the access pattern is determined in this manner, the ratio of the access frequency of each buffer can be made equal to the ratio of the required bandwidth, and the access interval of each buffer can be made as constant as possible. The operation of the access decision in the access pattern decision unit 2 will be described with a specific example. It is assumed that a total of four input links share one output link, and that there are three input links requesting a bandwidth among them, respectively connected to buffers i, j, and k. It is assumed that the required bandwidth of the input link connected to each of the buffers i, j, and k approximates an integer ratio of 7, 3, and 15.

The access patterns of the buffers k (band ratio 15) and i (band ratio 7) are determined by the first determining means 20. Next, by the second determining means 21, A = 1
5, B = 7, so A / B = 2, too 1, so M =
2, N = 1, and the access patterns of these two buffers k, i are as follows.

(Kikkikkikkikikkikkiki ...)
·) Next, the third determination means 22 determines A as the sum of the bands of the buffers k and i already determined 15 + 7 = 22, and B as the bandwidth 5 of the next buffer j. Using the new A and B, the access pattern is determined by the second determination means 21.
When j is embedded, the access patterns of the three buffers are as follows, and this is repeated thereafter.

(Kikkikjkikkkikkjkikkkiki)
j)

[0018]

FIG. 2 is a hardware configuration diagram of the embodiment.
In FIG. 2, reference numerals 4 to 7 are the same as those in FIG.
Denotes a plurality of input links, 5 denotes a buffer provided corresponding to each input link, 6 denotes a selector, and 7 denotes a single output link. 10 is a traffic parameter input from a subscriber (terminal) of each input link, and 11 is a central processing unit (C
PU), 12 is a memory for storing a program for determining an access pattern and data, 13 is a RAM for storing the access pattern, and 14 is a decoder for decoding a buffer address extracted from the access pattern.

The CPU 11 receives, from a subscriber (terminal) of each link, a parameter 10 such as a required bandwidth required for communication used by the subscriber or an average or dispersion of cell arrival intervals. At this time, there are cases where ATM cells (fixed length packets) of not only one subscriber but also a plurality of subscribers are multiplexed on one input link, and the CPU 11 receives these traffic parameters from the traffic parameters and stores them in the memory 12. Store. The CPU 11 is notified not only when an ATM call is requested but also when a call that has been performing communication until now releases (stops) communication.

The CPU 11 calculates the speed (required band) of reading out the ATM cells of the buffer 5 connected to each input link in response to a call request and release. A buffer read access pattern is determined using the calculated required bandwidth.
The process of determining the access pattern will be described below.

Here, when there are L buffers for the required bandwidth of each link in the CPU 11, when there are L buffers whose required bandwidth is larger than 0, numbers are assigned from 1 to L in order of the required bandwidth. That is, if the required bandwidth of buffer #i is C _i ′, C ₁ ′ ≧ C ₂ ′ ≧ C ₃ ′.
_L ′> 0.

These C ₁ ′, C ₂ ′, C ₃ ′,...
C _L ′ is converted to an integer ratio, and each integer ratio is expressed as C
_1, C _2, C _3, and ··· C _L. For example, C ₁ =
[C ₁ '/ _CL '], C ₂ = [C ₂ '/ _CL '], ...
C _i = [C _i '/ C _L '], C _L-1 = [C _L-1 / C
_L ′], C _L = 1. At this time, C ₁ ≧ C ₂ ≧
... ≧ C _L = 1> 0: C _i (i = 1, 2,..., L is an integer).

In the following description, it is assumed that the read pattern of the buffer is represented as a row vector having the buffer number as an element as in the following example. P = (1, 1, 1, 2, 1, 3,..., 2) Each element ≦ L, the number of elements = {C _i (Σ changes i from 1 to L) In this example, P # 1, # 1, # 1, # 2, #
Accessing in the order of 1, # 3... # 2 means repeating this.

When P is determined, P is created in order from the buffer having the larger required bandwidth.
Access patterns ^Pk up to 1, # 2,..., #K (where k <L) are determined. That is, P ^k = (1, 1, 1, 2, 1,..., 2) Each element ≦ k, the number of each element = ΣC _k ( _k varies from 1 to k) Then, a process of determining an access pattern is executed, and the process will be described below in the order of processes with reference to the drawings.

FIG. 3 is a processing flow for numbering buffers. When the buffers are numbered from 1 to L-1 in descending order of the required bandwidth, a new call setup request (CALL SETUP) is newly generated in a certain buffer, and the call originates from the subscriber. (Or data registered in advance) determines the required bandwidth C ', the buffer is numbered.

At the start of FIG. 3, the required bandwidth of the buffer that has already been determined is defined as C ₁ '≧ C ₂ ' ≧ C ₃ '.
C _L-1 '. First, L-1 is set as i (FIG. 3)
1), it is determined whether the required bandwidth C 'of the new call is larger than C _i (2), and if it is larger, the buffer #i is renumbered to # i + 1 (3) and i is decremented by -1. It is determined whether it is 0 (4, 5). If it is not 0, the process returns to step 2 and the next C _i ′ is determined.

If it is determined in step 2 that C ′ is not larger than C _i , the buffer is set to # i + 1
And the required bandwidth C 'of this new call is C _{i + 1} '
And the process is terminated. Further, in step 5, in the case of i = 0, the buffer # 1 and numbering the new call is entered, the required bandwidth C 'of the new call and C ₁ (the 7).

When the flow shown in FIG. 3 is executed, a buffer having issued a new bandwidth request is given an equal number in the order of the size of the bandwidth, and the number of each buffer having a required bandwidth smaller than that bandwidth is assigned to each buffer. Go up one. Next, a processing flow for determining an access pattern will be described with reference to FIG.

In this case, it is assumed that the required bandwidth of each buffer is sequentially stored in the memory 12 (see FIG. 2) by the processing of FIG. First, "1" is set in the register holding the variable i (one of the registers set in the memory) (1 in FIG. 4). Next, an integer ratio C _i = [C _i '/ C _L '] of the required band of the number i is calculated (the same as in the second example), and it is determined whether or not i reaches L (the same as the third example). to 4 repetitions reaches the P '= (11 · · 1) and the set C ₁ piece (read pattern only buffer # 1 with integer ratio of the maximum required bandwidth) (the 5).

Subsequently, it is determined whether L = 1 (No. 6), and if yes, since there is only one buffer having the required bandwidth, P = P 'is determined (No. 7). If no, the register holding the variable k (the other one of the registers set in the memory)
Is set to “2” (8), and ΔC _i / C
By executing _k (i is 1 to k-1), the quotient is obtained as M and the remainder is obtained as N (9). Then sets 1 to register for holding the other variables n in memory (at 10), P ^k-1 of the n (m +
1) Insert k after the element (11). Thereafter, it is determined whether n = N (12), and if they do not match, n is incremented by +1.
Then, the process returns to step 11 to repeat the processing.

When n = N, "1" is set in the register holding the other variable 1 (lowercase L) (1).
4). Next, k is inserted after the (M + 1) N + Ml-th element (1 is a lowercase letter) of P ^k-1 (15). Next, it is determined whether l = C _k −N (No. 16), and if they do not match, l (lowercase L) is set to l + 1 (No. 17),
Returning to step 15, the process is repeated. When 1 = C _k −N holds, a new vector obtained by adding C _k elements to ΣC _i elements of P ^k−1 becomes P ^k , and the read access pattern to be obtained is obtained (18).

[0032] As in the ATM network by such constant as possible read access interval of each buffer
In a simple packet network , the output link usage rate (traffic volume) can be improved with a low discard rate. FIG. 5 shows the effect of equalizing the cell read interval of the buffer. That is, the vertical axis in FIG. 5 is the buffer length (unit is the cell length) required when the cell loss rate is defined as the connection quality standard (QOS).
, And the horizontal axis represents the usage rate of the output link (the amount of traffic carried: the unit is Erlang). The symbols in the figure are Kendall symbols used in queuing theory, and X / Y / Z (U)
X represents arrival distribution, Y represents service time distribution, Z represents the number of servers, and U represents the number of queues (queue length) allowed. In the example of the figure, “M” used as X or Y represents an exponential distribution (the interval of services etc. fluctuates), and “D” used for Y represents a constant distribution (the intervals of services etc. are fixed).

According to FIG. 5, according to the present invention, when the access interval is constant, the characteristics indicated by M / D / 1 (S) are provided, and the cell discard rate achieves approximately 10 ^-10 and approximately 10 ^-6 . In
When the access interval is random as in the related art (M in FIG. 5)
/ M / 1 (S)) according to the present invention, the required buffer length is about 50%.

Next, the required cell read speed C 'of the buffer
Is shown in FIG. This flow is executed in order to obtain the requested cell read speed for each buffer, and is processed in real time when a call connection request or a change request is issued, or is processed in advance offline.

First, a buffer length S of the buffer is set, and a required connection quality condition (this is called QOS, for example, a cell loss rate) to be satisfied by the buffer is set (FIG. 6).
1). Next, traffic parameters for a new connection request call or disconnection request call are input (2). In response to this, the traffic of all the calls that share the buffer, excluding a new connection request call or excluding a disconnection request call, is represented by a parameter T (τ, σ ² ) (3). In this case, τ
Is the average of the cell arrival time intervals, and σ ² is its variance. Note that τ and σ ² in this case are obtained by using a process flow of the cell arrival process approximation of FIG. 8 described later.

Next, the initial value of C 'is selected in the register in which the requested cell read speed C' is set (4). Here, it is appropriately selected from a plurality of values prepared in advance. Next, each flag is set to "0" in a register storing the flags PY and PN for controlling the processing (5). Thereafter, the connection quality Q in the buffer is calculated using T, C ', and S (Same as 6). It is determined whether or not the calculated connection quality Q satisfies the required connection quality condition QOS (step 1) (7).

At this time, if satisfied, it is determined whether or not the flag PN is "1" (8), and if it is 1, C 'at this time is set as the requested cell reading speed for obtaining C', and the processing is terminated. (Ibid. 9). If the flag PN is not “1”, C ′
−ΔC is set in the register, the flag PY is set to “1” (10, 11), the process returns to step 6, and the process is repeated. Here, ΔC is the minimum management unit of the reading speed.

In step 7, if the connection quality Q does not satisfy the required connection quality condition QOS, it is determined whether PY = 1 (No. 12), and if yes, C '+ ΔC is set in the register for the required cell read speed. And ends (13). In the case of no, C ′ + ΔC is set in the register similarly, and the flag PN is set to “1” (14, 15).
The process returns to step 6 through the path described above, and the process is repeated again.

Next, a description will be given of a case where the average and variance of the arrival intervals of all the ATM cells added to the buffer are used based on the traffic parameters applied by the user when characterizing the traffic added to each buffer. FIG. 7 is a characteristic diagram showing the influence of the average and variance of cell arrival. The vertical axis in FIG.
Cell loss rate of buffer that can store 0 cells,
The horizontal axis represents the average traffic volume (the unit is Erlang), the symbols in the figure are the Kendall symbols used in the queuing theory as in FIG. 5, X in X / Y / Z is the arrival distribution, and Y is Is the distribution of service time, and Z is the number of servers (including output links).

[0040] When G1 / M / 1 expressed (arrival distribution generally distributed in accordance with the regeneration process, if the link is 1 service time exponential distribution) the variance of the average arrival time at at C _v, each type as arrival distribution Represents a variance value (square of variance C _v ² ) corresponding to the distribution of. FIG.
Is an example of C _v ² = 2 when the arrival distribution is a two-dimensional super-exponential distribution (H ₂ / M / 1), and the arrival distribution is an exponential distribution (M / M /
C _v ² = 1 in the example of the case of 1), examples of C _v ² = 0.5 in the case of arrival distribution is two-dimensional Erlang distribution _{(E 2 / M / 1)} ,
Also, an example is shown where C _v ² = 0 when the arrival distribution is a constant distribution.

Thus, it can be seen that the cell discard rate and the amount of traffic to be transmitted are greatly affected by the dispersion rate of the arriving cell. Therefore, a method of obtaining a variance other than the average of the cell arrival rate as the traffic parameter applied by the user will be described.

A method for obtaining the average and variance of the arrival times of all the ATM cells added to the buffer based on the traffic parameters will be described below. It is assumed that N buffers 1 to N are added to a certain buffer. The average of tau _i, dispersing sigma _i ² of the arrival interval of the cells of a call #i, the average number of cells that arrive over a period of time with a call #i m _i, and the dispersion v _i.

Further, the average arrival time τ, the variance σ ² , and the square variance coefficient C _v ² = (σ ² / τ ² ) of the arrival intervals of the cells in the N cells added to the buffer, and the number of cells arriving during a certain period Let m be the average and v be the variance. Step 3 in FIG. 6 is characterized by the arrival process of the cells of all N calls sharing a certain buffer. Here, as an example, τ _i , σ _{i of} each call are obtained from the traffic parameters of each call.
² (i = 1, 2,... N) is obtained, and on the premise of this, the average τ and the variance σ ² of the cell arrival intervals over the N calls are calculated from this value as T (τ, σ ² ).

That is, FIG. 8 is an approximate processing flow of the cell arrival process of a plurality of calls based on the traffic parameters of each call, and is an example that can be adopted in step 3 of FIG. First, τ _i and σ _i ² are input (1 in FIG. 8). Next, m _i = 1 / τ _i and v _i = σ _i ² / τ _i ³ are obtained (the same as above). Further, m = Σm _i , v = Σv _i (where i = 1, 2,... N) is obtained (3), and the following values are obtained using these values (4).

Τ = 1 / m, σ ² = v / m ³ , _Cv ² = v / m In the method of FIG. 8, the average τ _i and the variance σ _i ² of the arrival interval of the cell of the call i are respectively it was assumed that obtained from the traffic parameters of the call, wherein an example of using P _1, P _2, P ₃ of the following traffic parameters, by using the tau _i, the sigma _i ² The method for obtaining the value will be described.

P ₁ : Transmission speed of the user terminal (maximum bandwidth of the user terminal) P ₂ : Average packet length P ₃ : Average packet arrival interval The following method can be applied even when parameters equivalent to these are used. For example, instead of the above-mentioned P _3, P ₃ ': When using the traffic parameters as the mean used bandwidth P _1, P ₂ of the user terminal, P _3' = P ₁ - relationship (P ₂ / P ₃₎ Is used to convert P ₃ ′ to obtain P ₃ .

The above three parameters P ₁ , P ₂ , P ₃
Are the following parameters β ₁ , β ₂ , β in the IPP approximation
Can be converted to ₃ . β ₁ : average cell arrival rate when random switch is on (β ₁ ^-1 is average cell arrival interval) β ₂ : rate at which random switch is turned on (β ₂ ^-1 is average time when random switch is turned on) β ₃ : rate at which the random switch is turned off (β ₃ ^-1 is the average time during which the random switch is turned off) where β ₁ ^-1 = cell length (53 bytes) / P ₁ β ₂ ^-1 = P ₂ / P ₁ β ₃ ⁻¹ = P ₃ − (P ₂ / P ₁ ) The arrival process of the cell approximated by IPP is a two-dimensional hyperexponential distribution (H ₂ ), and the average τ of the cell arrival interval and the variance σ
² can be obtained as follows.

[0048] tau = The _{(β 2 + β 3) /} β 1 β 2 σ 2 = τ 2 + (2β 2 / β 1 β 3 2), square dispersion coefficient C _v ² is obtained as follows. C _v ² = σ ² / τ ² Next, in steps 4 to 15 of the processing flow of FIG. 6, the buffer T is determined based on the traffic T added to a buffer, the length S of the buffer, and the required connection quality condition QOS of the buffer. Is performed to obtain the cell read speed C ′. On the other hand, FIGS. 9 and 10 show how the cell read speed C 'is obtained when T (the average and variance of a plurality of calls) is obtained as shown in FIG.

In the processing (steps 4 to 15) of FIG.
The minimum cell read speed C ′ that satisfies the OS is obtained while changing the read speed C ′ set by assumption and checking the connection quality Q in the buffer each time.
On the other hand, in the case of FIGS. 9 and 10, Q is the square dispersion coefficient C.
_v If ² is considered to be a good as becomes the smaller, than the C _v ^'2 that can be acceptable for QOS, from T C
_v or ² is reduced by checking every time changing the C ', which is exactly equivalent to the operation and process of FIG 6.

FIGS. 9 and 10 are flow charts (1) and (2) for determining the cell reading speed using the average and variance of the cell arrival rate. Note that this processing flow shows only a case where a call connection request is generated, and does not include a case where a call is disconnected. To outline this processing flow, first, PY is set to "0" (step 1 in FIG. 9) as in step 5 in FIG. 6, and the available bandwidth = output link speed-read speed of other buffers excluding the buffer. C _i ′ and the vacant band is set as C ′ (2).

Next, τ, C _v ² (square variation coefficient obtained from the traffic parameter reported by the user) obtained by the processing flow of FIG. 8 is input (3). Next, the constant h
Is stored in a specific register to store (53 octets) /
C ′ is set, and h / τ is set in a specific register for storing a constant ρ (4). Next, QOS input is performed (5). In this case, for example, the probability P that the waiting time exceeds T is input as QOS. Then, step 6
Is obtained from the relational expression concerning P shown in (6). This ω is a numerical value that means. Next, C _v ² > 1
Is determined (No. 7).

Thereafter, the flow shifts to FIG.
In the case of S, step 8 in FIG.
The calculation in step 9 is performed. Step 8 and Step
Satisfies the relational expression of the corresponding expression ρ
C_v ^TwoAnd find it C _v’^Two(Satisfy QOS
Square variation coefficient). Next, in step 10,
C_v ^TwoAnd C obtained in step 8 or 9_v’^Twoof
A comparison is made (10 in FIG. 10).

As a result, if _Cv ² is larger, the process goes to step 11 to determine whether PY = 1. If PY
If not = 1, the process of step 14 is executed. That is, the last call to be added is rejected.
If the answer is affirmative in step 11, C '+. DELTA.C is set in the register C' (12), and C 'is obtained as the minimum required read access speed for the buffer (13).

In step 10, C _v ' ² and C _v ²
If they match, C 'at that time is set in the register C' (15), and the process proceeds to step 13 in the same manner as described above. Further, the step of C _v ^'2> C _v is ^2, then set the "1" PY (the 16), the C'-[Delta] C register C' is set to (the 17), above 9 Return to 2 and repeat the process.

According to the present invention, a buffer memory for sharing a single output link with a plurality of input links and holding received ATM cells (fixed length packets) until the output link becomes empty is provided for each input link. It can be implemented in a configuration. Therefore, A
The present invention can be applied not only to a concentrator of a packet network such as a TM network, but also to a switching mechanism of an ATM exchange and a home network. That is, one input link may be multiplexed by a plurality of calls, such as when it is used as a part of a communication path of an ATM exchange, or AT may be transmitted from several terminals.
As in the case of concentrating and multiplexing M cells to share one subscriber line, one input link may be occupied by only one call, and the present invention can be implemented in any case. In addition, a plurality of calls are multiplexed on one input link, which can be dealt with by assuming the average and dispersion of the cell arrival rates of the entire input link.

Further, the present invention can be effectively used if the unit carrying information is not limited to the ATM cell composed of the header field of 5 bytes and the information field of 48 bytes, but has a fixed length.

[0057]

According to the present invention, if the read access interval of each buffer is made as constant as possible, the buffer length required for obtaining a predetermined cell loss rate can be reduced by the characteristics shown in FIG.

By determining the access pattern according to the present invention, the number of input links for which a bandwidth request is
Then, the range of fluctuation of the access interval of each buffer is at most X, and the number of accesses can be made to match the ratio of the required bandwidth. As a result, traffic smoothing can be realized, and connection quality such as a fixed-length packet discard rate of a packet network such as an ATM network can be improved.

Further, according to the present invention, the following effects can be obtained. Fixed-length packets such as cells that arrive periodically
Output can be performed while maintaining the cycle of the bracket . Even a fixed length packet such as a cell arriving in a burst can be output periodically. This means that the bandwidth of the output link is equivalently divided according to the required bandwidth of each input link.

By independently calculating the required bandwidth for each input link, the connection quality unique to each buffer can be guaranteed.
That is, a call requesting high quality and a call permitting low quality are accommodated in separate buffers, and the frequency of access to the buffer requesting high quality is increased, thereby enabling flexible buffer read control.

By considering the dispersion as well as the average of the cell arrival intervals (or cell arrival rates) in the calculation of the required bandwidth, the required bandwidth can be accurately estimated, and a readout cell can be made taking this into account. That is, even if the average is the same, the required bandwidth is different if the variance is different, and the same quality cannot be satisfied unless the frequency of reading out the buffer is increased as the variance is larger.

[Brief description of the drawings]

FIG. 1 is a basic configuration diagram of the present invention.

FIG. 2 is a hardware configuration diagram of an embodiment.

FIG. 3 is a processing flow of buffer numbering.

FIG. 4 is a processing flow of access pattern determination.

FIG. 5 is a diagram illustrating an effect of equalizing a cell read interval of a buffer.

FIG. 6 is a processing flow for obtaining a required cell reading speed of a buffer.

FIG. 7 is a characteristic diagram showing the influence of average and variance of cell arrival.

FIG. 8 is a process flow of approximating a cell arrival process of a plurality of calls by using traffic parameters of each call.

FIG. 9 is a processing flow (1) of determining a cell reading speed using an average and a variance of a cell arrival rate.

FIG. 10 is a process flow (2) for determining a cell reading speed using the average and variance of a cell arrival rate.

FIG. 11 is an explanatory diagram of a conventional example.

[Explanation of symbols]

Reference Signs List 1 Required bandwidth calculation unit 2 Pattern determination unit 3 Holding output unit 4-1 to 4-n Plural input links 5-1 to 5-n Buffers composed of FIFOs provided corresponding to each input link 6 Selector 7 Output link

──────────────────────────────────────────────────続 き Continuing on the front page (72) Inventor: Kazuto Jomasu, Fujitsu Limited, 1015, Kamidadanaka, Nakahara-ku, Kawasaki City, Kanagawa Prefecture (72) Inventor: Yuji Kato 1015, Kamiodanaka, Nakahara-ku, Kawasaki City, Kanagawa Prefecture, Fujitsu Limited (56) References JP-A-62-102637 (JP, A) JP-A-2-260845 (JP, A) JP-A-2-164158 (JP, A) JP-A-3-297245 (JP, A) Kaihei 4-137941 (JP, A) (58) Fields studied (Int. Cl. ⁷ , DB name) H04L 12/28 H04L 12/56 H04J 3/22

Claims (5)

(57) [Claims] 1. A fixed length packet of a plurality of input links is
In the concentrator method of sending data to a single output link, for each buffer provided corresponding to the input link,
A calculating unit for calculating a required bandwidth of the fixed-length packet to be input ;
Equal to the integer ratio of the proportion occupied by each required band,
A determining unit for determining an access pattern including an access frequency and an order so that a read access interval of each buffer is substantially constant; and reading from each buffer to an output link is performed according to the determined access pattern. Concentration method . 2. The method according to claim 1, wherein the fixed length packets of the plurality of input links are
In the line concentrator method of transmitting data to a single output link, for each buffer provided corresponding to the input link,
The required bandwidth (read speed) of the fixed-length packet to be input is
A calculating unit for calculating, and an access frequency of each of a plurality of buffers in all required bandwidths.
Equal to the integer ratio of the proportion occupied by each required band,
Make the read access interval of each buffer almost constant
Determine access patterns, including access frequency and order
A determination unit for determining an access pattern determination order from the buffers in the order of the required bandwidth; and determining the required bandwidth of the two selected buffers as an integer (A , B, A> B), and the division of both (A / B)
Quotient (M) and remainder (N) of the buffer, an integer of the required bandwidth, a second determining means for determining an access pattern between the two buffers using the quotient and the remainder, and a buffer for which the access pattern has already been determined. And the third determining means for setting the required bandwidth of the buffer in the next order to be the required bandwidth of the two buffers in the second determining means, and from each of the buffers according to the determined access pattern. Out
A line concentrating method characterized by reading out to a force link . 3. The access pattern determination unit according to claim 2, wherein the second determination means of the access pattern determination unit calculates an integer approximating a required bandwidth of the two buffers (i, j) by A, B (A> B), and division ( Using the quotient M and remainder N of (A / B), (1) access buffer i M + 1 times and access buffer j once, (2) access buffer i M times and then buffer j once
Accessing, (3) collecting the N times of the buffer j to B times that by (1), B-N times is characterized by determining the access pattern to be accessed by (2)
Line method . 4. The method according to claim 1, wherein the required bandwidth of the buffer of each of the input links is based on a traffic parameter requested by a user for each buffer, and a total traffic added to the buffer is used as a statistical parameter. A line concentrating method characterized in that the required bandwidth of each buffer is calculated based on the parameters acquired and the quality guaranteed by each buffer. 5. The method according to claim 4, wherein when obtaining the total traffic added to each buffer as a statistical parameter, the total fixed length packet added to the buffer is based on the traffic parameter applied by the user.
And obtaining the mean and variance of Tsu city of arrival interval
Concentration method .