Method and system for varying data packet size for controlling bandwidth

The invention relates to a method and system for controlling the size of data packets transmitted from a sender device to one or more receiving devices, for achieving bandwidth control.

In performing data transmission using various media such as cable, telephone line, satellite communication, etc., it is conventional for the data to be sent from one device to one or more receiving devices. The data is first fragmented, or segmented, into packets and then the packets transmitted, or broadcast. The term "packet" is used herein and conventionally to indicate a sequence of bytes of digital data and represents the smallest unit of transmission and reception. Hereafter, a "device" is defined as a hardware element, usually having a software component, capable of receiving and/or transmitting data packets. Examples of such devices include computers, GPRS mobile phones and network equipment.

The term "content", as used herein, indicates data that is segmented into a sequence of packets. The data contained within the content could be a file, or part of a file, part of a data stream, or any collection of data. The content can comprise pure data, or audio and video data streams, or of any combination. The content is sent in sequentially transmitted packets from a sender device to one or more receiving devices.

The sending and receiving devices typically run software programs in conjunction with a computer, whose purpose at the sender device is to fragment the content to be sent into the packets, and at the receiver device to reassemble the packets received into the original content.

Bandwidth is conventionally defined and used herein to mean the amount of data that can be transmitted over the media in a given time frame, for example, 10 Mbps (10 million bits per second). Essentially, this is the speed of the transmission, that is, the speed at which data of the content is sent from the sending device to the receiving devices. Each transmitting-receiving system has a certain bandwidth capability that is defined by various factors, such as the type of media used for the transmission and the hardware and software equipment used for transmission and reception of the data at the sending and receiving devices. For example, a broadband cable medium has a larger bandwidth capability than a telephone line.

Various types of protocols are used to transmit the data packets forming the content. Some protocols may be considered to be "unreliable", herein meaning any transmission protocol that provides best-effort delivery of packets, and in particular, does not perform automatic re-transmission of lost or corrupted packets. Examples of "unreliable" protocols currently in common use include unicast and multicast User Datagram Protocol (UDP), ATM Adaption Layer (AAL) Types 3/4 and 5 in non-assured transmission mode, AppleTalk DDP datagrams, and unicast and broadcast MPEG-2 transport streams.

Often, data from different files, that is, different content, is to be transmitted at the same time over the same media. For example, if a particular system has a 10 Mbps bandwidth transmission capability, the packets of two different contents can be transmitted at the same time in separate streams, for example, each stream of 5 Mbps bandwidth. In various commercial applications, portions of the available bandwidth of a given media are sold, or leased to several different customers for use at the same time.

Some applications require that packets are to be received in a regular and timely fashion, for example, for audio and video transmissions in which it is important that the jitter (i.e., the amount of variation in the end-to-end packet transit time) is kept as low as possible. Increasing reliability at the receiving device implies use of a high precision control of the bandwidth at the sending device, especially for broadband applications.

In general, the bandwidth usage of a broadcast application is selected according to network load and speed, leased bandwidth capacity, and processing speed of the receiving device. The bandwidth allocated for a given transmission is hereafter referred to as the "target bandwidth" (B_T). During the transmission, the actual bandwidth used can vary relative to the selected target bandwidth.

As the broadcast takes place, the bandwidth varies and can be measured at any time. The "instantaneous bandwidth" is the bandwidth measured in the shortest measurable time. For example, if the bandwidth usage is regularly checked once per second, the instantaneous bandwidth is computed by dividing the quantity of data transferred in that time frame (in this case 1 second) by that time interval.

The broadcast of the content has an "average (or mean) bandwidth" B_M. This is the total amount of data transmitted during a transmission divided by the transmission duration. The terms "mean bandwidth" and "average bandwidth" are used interchangeably herein.

Since the content is transmitted in packets, there is a pause, or wait time (t_w), between the start of transmission of successive packets. The wait time between packets is related to the bandwidth used during transmission of a content. For example, increasing the wait time between packets while maintaining a fixed packet size results in less data transmitted during a given time, and vice versa.

If data is sent without a precise control relative to the target bandwidth, various problems can arise. Basically, there are three different scenarios that can cause a problem: <ol id="ol0001" compact="compact"><li>(1) the average bandwidth B_M is too high relative to the target bandwidth B_T value;</li><li>(2) the average bandwidth B_M is too low relative to the target bandwidth B_T value;</li><li>(3) the average bandwidth B_M is equal, or very close, to the target bandwidth B_T , but the instantaneous bandwidth values measured during the transmission are different from the target bandwidth. This type of transmission is hereafter referred to as heterogeneous, i.e., contains peaks.</li></ol>

The problems caused by the different scenarios described above can adversely affect different components and structures of the system that receive the data, such as routing and switching devices and the destination receiver devices. This is described below. <ol id="ol0002" compact="compact"><li>(1) <u style="single">Sending data packets too fast</u>. Broadcasting of the data at a speed above the target bandwidth B_T usually results in substantial packet losses. Loss of packets reduces the integrity of the data transmission. Also, too high a transmission speed can cause congestion if routing devices are not able to buffer the incoming data stream. The same problem can affect encapsulators if data is sent to a satellite uplink. Here, entire series of packets can be lost. At the receiving device side, packet reassembly can be difficult if the hardware or software application program processing the received data is not designed for an incoming data stream speed higher than the bandwidth mean (average) value. Moreover, a local area network (LAN) to which the receiving devices are connected can experience congestion if no traffic buffering mechanism is provided between the data transmitter and the receiving devices.</li><li>(2) <u style="single">Sending data packets too slow</u>. If the packets are transmitted at a speed lower than the specified target bandwidth, no losses of data packets should be experienced. However, a resource waste is caused, especially if the connection between the data transmitter and receiving devices is purchased or leased on a bandwidth usage basis instead of on a transmitted data amount basis. Also, audio and video streaming quality can be affected if the receiving data rate is too low. That is, the received sound and video will be intermittent, garbled, or otherwise distorted</li><li>(3) <u style="single">Peaks in bandwidth usage</u>. Even when the average output transmission bandwidth is close to the target bandwidth does not always guarantee that the data stream will be free of problems. For example, there can be bandwidth usage peaks at any instant during transmission. Also, if the packets are not homogeneously distributed during the transmission, i.e., a fairly constant time between the successive packets, packet clustering occurs that causes an increase or decrease of the bandwidth used at a given instant, this being the instantaneous bandwidth. As a result, the same problems described above for transmissions with mean bandwidth higher or lower than the target bandwidth can arise, although the adverse affects are experienced to a lesser degree.</li></ol>

Another parameter that can be considered is the transmission burstiness, which is defined as: <maths id="math0001"><math display="block"><mi mathvariant="italic">Burstiness</mi><mo>=</mo><mfrac><msub><mi mathvariant="italic">B</mi><mi mathvariant="italic">peak</mi></msub><msub><mi mathvariant="italic">B</mi><mi mathvariant="italic">M</mi></msub></mfrac></math><img file="EP1584162B1_D0001.tif" /></maths> where B_M is the mean bandwidth of a transmission content, and B_peak is the highest of the absolute instantaneous bandwidth values measured during a transmission. That is: <maths id="math0002"><math display="block"><msub><mi mathvariant="normal">B</mi><mi>peak</mi></msub><mo mathvariant="normal">=</mo><mi>max</mi><mo>⁢</mo><mfenced><msub><mi mathvariant="normal">B</mi><mfenced open="[" close="]"><mi mathvariant="normal">t</mi><mo mathvariant="normal">,</mo><mi mathvariant="normal">t</mi><mo mathvariant="normal">+</mo><mi mathvariant="normal">Δt</mi></mfenced></msub></mfenced></math><img file="EP1584162B1_D0002.tif" /></maths>

B_[t,t+Δt] : instantaneous bandwidth B measured in the interval included between t and t + Δt.

Ideally, the burstiness should tend to one. That is, the bandwidth being used should be centered at B_T. Ideally, B_M = B_T. Note also that under the same conditions of burstiness, the transmission homogeneity could be different, since no information is provided in this value about the recurrence of B_peak. In fact, the burstiness value says how large the peak is relative to the mean transmission bandwidth B_M, but not how many times the peak is repeated during a transmission. That is, a transmission can experience 1 or 1000 peaks, but the value B_peak is the same in most or all cases. Also, no information is provided about the homogeneity of the transmission.

Accordingly, it would be desirable to be able to better control the bandwidth usage for the transmission of packets containing any type and different amounts of content. Prior attempts to solve the various problems discussed above typically enforce a maximum bandwidth. That is, the transmission protocol is such so as to intervene when a maximum value bandwidth is exceeded, but does not intervene when the bandwidth usage is lower than the maximum. Other existing solutions correct existing traffic streams. That is, a protocol is used to solve congestion problems caused by senders without inbuilt traffic control. Still other existing solutions discard packets when buffer space is exceeded. Some of the solutions of this type discard packets using an algorithm, such as the "leaky bucket" algorithm (Tanenbaum, Computer Networks, 3^rd Edition, p, 380, Prentice Hall). Other solutions permit a certain amount of burstiness in the output stream using an algorithm such as the "token bucket" algorithm (Tanenbaum, p. 381).

Prior application Serial No. <patcit id="pcit0001" dnum="US10062830B"><text>10/062,830, filed January 31, 2002</text></patcit>, which is assigned to the same Assignee, provides a method and apparatus by which the bandwidth can be held more closely to a desired target bandwidth B_T value by properly selecting the pause (wait time t_w) between the successive packets and the packets are transmitted using the selected wait time during the transmission of the content. The method and system of that application determines the wait time t_w between packets in accordance with a novel algorithm developed and implemented that relates the desired size P of the packets of a content to be transmitted to the desired target bandwidth B_T. The parameters of desired target bandwidth B_T and size of the packets P are provided or selected, and from these the wait time t_w is computed. By properly selecting and controlling the wait, or pause, time pause t_w, between the transmission of the successive packets the transmission of the content at the desired target bandwidth is more closely controlled and maintained at the selected value.

The invention of that application can be implemented in software and/or hardware. In a preferred embodiment of that invention, the precision of the bandwidth control is maximized by the use of the highest resolution clock or other timing device available at the sending devices. This provides the best compensation for wait time t_w rounding errors, i.e., rounding the wait time value up or down to the nearest whole number value achievable by the system timing devices.

According to a first aspect of the invention, there is provided a method for controlling the bandwidth of transmission of data content by an application program that transmits contiguous collections of data having a minimum block size (d_min) in cycles of successive data packets characterized by the steps of: <ul id="ul0001" list-style="none" compact="compact"><li>selecting a target bandwidth (B_T) for the content transmission and a fixed wait time (t_w) between packets; calculating a minimum and a maximum packet size based on the minimum data block size (d_min) and a theoretical packet size, the theoretical packet size being equal to the target bandwidth (B_T) times the wait time (t_w);</li><li>determining the number of packets of different sizes to be transmitted during a cycle with the fixed wait time (t_w) therebetween; and</li><li>transmitting said packets of different size with said wait time (t_w) between successive packets.</li></ul>

According to a second aspect of the invention, there is provided a system for controlling the bandwidth of transmission of data content in cycles of successive data packets comprising: <ul id="ul0002" list-style="none" compact="compact"><li>computer means including an application program, characterized by a minimum data block size (d_min) representing the smallest contiguous collection of data that the application program can transmit, for receiving parameters of a target bandwidth (B_t) for the content transmission and a fixed wait time (t_w) between packets, said application program being operable to: <ul id="ul0003" list-style="none" compact="compact"><li>calculate a minimum and a maximum packet size based on the minimum data block size and a theoretical packet size, said theoretical packet size being equal to the target bandwidth times the wait time; and</li><li>determine the number of packets of different sizes to be transmitted during each cycle with the fixed wait time (t_w) therebetween;</li><li>said computer means operating in response to the application program to determine and control transmission of a number of packets of different sizes during a cycle with a fixed wait time (t_w) therebetween to achieve transmission at the target bandwidth (B_T).</li></ul></li></ul>

The present invention is directed to another method and system for controlling bandwidth to a target value B_T during transmission of the packets of a content. In the present invention, the size of the packets transmitted during a "cycle" is varied. As used herein, the term "cycle" means a given number of packets having different sizes. A number of cycles of packets is used during the transmission of a content.

In implementing the system and method of the invention the content to be sent is divided into packets. The packets are regularly sent with a fixed wait time interval t_w between successive packets. The packets are of two different sizes, a minimum P_min and a maximum P_max, during the transmission and possibly one of a third size between P_min and P_max for the last packet of the content. This results in the average bandwidth usage value B_M being as near as possible to the target bandwidth B_T for the given transmission. Also, the instantaneous bandwidth B_I should be as near as possible to the target bandwidth B_T (in order to keep the burstiness as low as possible).

In considering the present invention as compared to the prior application, in the prior application the size of all packets was fixed to a value P (with the possible exception of the last packet). In this invention, the packets can have two different sizes: P_min or P_max) (again, with an exception for the last packet).

The method and system of the present invention finds particular use in applications using "unreliable" protocols, such as UDP, which do not use a comeback or returning traffic, as distinguished from so-called "reliable" protocols, such as TCP, which utilize such return traffic.

Other objects and advantages of the present invention will become more apparent upon reference to the following specification and annexed drawings, in which: <ul id="ul0004" list-style="none" compact="compact"><li><figref idref="f0001 f0002 f0003 f0004">Figs. 1A-1D</figref> are flow charts showing the computation of the packet sizes;</li><li><figref idref="f0005">Fig. 2</figref> is a diagram illustrating transmission during one cycle; and</li><li><figref idref="f0006">Fig. 3</figref> is a schematic diagram showing a typical transmission and receiving system.</li></ul>

In the method and system of the invention, the sending device is computer controlled and transmission of the content at the sending device is controlled by a computer application program. The program is structured to make the necessary calculations and control the size of the packets and distribution of packets of different sizes and the transmission of such packets. This is described below. The computer program is written using any suitable conventional programming language.

In using the preferred embodiment of method and system, the input parameters for the application program are: <ul id="ul0005" list-style="dash" compact="compact"><li>target bandwidth B_T</li><li>fixed waiting interval between packets t_w</li><li>minimal data block size used in forming a packet that can be handled d_min</li></ul>

The target bandwidth B_T represents the bandwidth at which the content (including the protocol overhead) is to be sent. This parameter is typically specified on a per-transmission basis. As explained above, several transmissions with the same or different target bandwidth B_T can be taking place at the same time over the same medium. A properly constructed system can use one computer and one application program to accomplish this, or there can be a separate computer and application program for each transmission.

The wait time interval t_w is the time interval that occurs between successive packets of the transmission. The use of a high-resolution timer is not required in the preferred embodiment of the invention, but it is important that the hardware and/or software implementation of the sending device is capable of reliably waiting for the desired interval with a relatively high degree of precision. In general, the wait time interval is chosen according to the hardware and software (operating system, programming language) specifications. Usually, the limitations are on the software side. When programming, there typically exist functions that can be implemented to wait for a desired time. Specifications exist that describe the precision limitations of these functions, also according to the operating system used. It should also be considered that the intercorrelations between functions implemented in software development kits and operating systems are not always well predictable. The following is a procedure used in the preferred embodiment of the invention in choosing the wait time interval: <ul id="ul0006" list-style="dash" compact="compact"><li>choose a function according to the software application kit and/or programming language used for development. For example, there can be a given function called "wait" where the input parameter is the waiting time in a given time unit, e.g., wait (10) means wait 10 milliseconds.</li><li>carefully select the function limitation in relation to the operating system to be used. For example, if the time limitation for the wait() function on a given operating system is 1 millisecond, the input parameter for the function cannot be smaller than 1 millisecond.</li><li>consider that usually the limit provided in the specifications is not real. For example, if the program calls for wait(1), it can happen that, in reality, the wait time will be more than 1 millisecond. For this reason, it is preferred that the effective precision of the time function be proven. This can be efficiently done by repeating many times (e.g., 1 million) the function with its limit and computing an average inaccuracy. For example, the function waits 1.3 milliseconds instead of the specified 1.0 millisecond value.</li><li>once this real limit is found, it must be considered that if concurring software applications or processes run on the same machine, that delays can result. It is very important to be sure that a precise wait time can be ensured during the entire transmission. For this reason, a "security factor" must be chosen in the development of the sending application. It also must be considered that by increasing the wait time, the wait time precision increases, but also the transmission burstiness, defined above, will increase (if the other parameters are fixed). Here, the decision must be made considering both factors and also other factors, e.g., the transmission protocol used, the maximum transmission bandwidth to be reached, etc. In a typical application, the security factor of 10 can be chosen. That is, the wait time interval will be 10 milliseconds (instead of 1 millisecond) as described in the programming specifications. This is valid for the example mentioned above, but it is not a general rule. The choice of the values 1 millisecond / 10 milliseconds are strictly related to the example.</li></ul>

Since in the present invention the wait time interval between successive packets is fixed, this method and system are suitable for use where a high-resolution timer is not available.

In the invention, the minimal data block size d_min represents the smallest contiguous collection of data that the application program can transmit. As explained below, this value must be at least equal to the protocol overhead of a packet plus 1 byte. By increasing the minimal data block size d_min, the percentage of "useful" data (content data) contained in a given cycle packet is increased, i.e., the proportion "content data vs. header information" increases. Nevertheless, increasing the minimal data block size will increase the difference between P_max and P_min and, consequently, the transmission burstiness (if the other parameters are fixed).

In the system and method, the content of a transmission to be sent is divided into packets. The packets are regularly sent with a fixed interval t_w between successive packets. The packets are of two different sizes, a minimum P_min and a maximum P_max, during the transmission and possibly one of a third size between P_min and P_max for the last packet of the content. This results in the average bandwidth usage value B_M being as near as possible to the target bandwidth B_T for the given transmission.

Moreover, the wait time t_w should be chosen according to the minimum data block size d_min. If the difference between P_min and P_max is large, the wait time should be kept as small as possible (but according to the precision limit described above). Again, the difference between P_min and P_max is equal to d_min.

The maximum and minimum of the instantaneous bandwidth, defined above, used during the wait time t_w between successive packets can be calculated using the following formulae: <maths id="math0003"><math display="block"><msub><mi>B</mi><mi>max</mi></msub><mo>=</mo><mfrac><msub><mi>P</mi><mi>max</mi></msub><msub><mi>t</mi><mi>w</mi></msub></mfrac></math><img file="EP1584162B1_D0003.tif" /></maths><maths id="math0004"><math display="block"><msub><mi>B</mi><mi>min</mi></msub><mo>=</mo><mfrac><msub><mi>P</mi><mi>min</mi></msub><msub><mi>t</mi><mi>w</mi></msub></mfrac></math><img file="EP1584162B1_D0004.tif" /></maths> where B_max = maximum instantaneous bandwidth B_min = minimum instantaneous bandwidth P_max = maximum size packet P_min = minimum size packet

<figref idref="f0001 f0002 f0003">Figs. 1A-1C</figref> show the implementation of the invention. In particular, the drawings 1A-1D show the following processes:

<figref idref="f0001">Fig. 1A</figref> (steps S101-S130) depicts the computation of packet composition by computing the different packet sizes and the number of packets for each packet size during a cycle. During a cycle, the average bandwidth B_M must be equal to the target bandwidth B_T (exception can be the last cycle).

Referring to <figref idref="f0001">Fig. 1A</figref>, in S101 and S103 the target bandwidth B_T and wait time t_w, are input to the computer application program. The process steps are sequentially numbered. <ul id="ul0007" list-style="none" compact="compact"><li>1. Using the values input from S101 and S103, in S105 a theoretical packet size P_theoretical is calculated where <maths id="math0005" num="(Eq.1)"><math display="block"><msub><mi mathvariant="normal">P</mi><mi>theoretical</mi></msub><mo mathvariant="normal">=</mo><msub><mi mathvariant="normal">B</mi><mi mathvariant="normal">T</mi></msub><mo mathvariant="normal">*</mo><msub><mi mathvariant="normal">t</mi><mi mathvariant="normal">w</mi></msub></math><img file="EP1584162B1_D0005.tif" /></maths></li></ul>

The theoretical packet size P_theoretical is the theoretical size that every packet should have in order to ensure that the target bandwidth B_T for a given content transmission is achieved when sending packets with a fixed wait time interval t_w between successive packets. The packet size granularity (rounding packet size to a whole number) is not infinitesimally small, but is determined by the minimum data block size d_min that makes up the packet. For example, if the minimum data block size that the application program can handle is 50 Bytes, and the P_theoretical packet size value (resulting from (Eq. 1) is 976.45 Bytes, the possible packet sizes rounded down to the theoretical value is 950 Bytes and 1000 Bytes. Here, 950 Bytes = P_min and 1000 Bytes = P_max.

It is also possible to consider P_theoretical as the average packet size during the transmission in order to achieve the target bandwidth B_T by sending the packets of the theoretical size with a fixed t_w interval therebetween.

As explained above, a relationship exists between the various input parameters. In a typical example, the maximum target bandwidth B_T for the application is selected. For example, the software program will reasonably send the data at a maximum of 50 Mbps. Then the theoretical packet size P_theoretical is computed using the previously selected t_w. Considering now the transmission protocol packet size and header information, a reasonable value for d_min is computed for carrying the content data, but also trying not to increase the burstiness too much.

For example: if the target bandwidth B_T is 50 Mbps and the wait time t_w is chosen as 100 milliseconds (0.1 seconds), then the theoretical packet size is 5 Megabits (=5000000 bits), that is, 625000 Bytes. Consider a transmission protocol with a maximum packet size of 350000 bytes and overhead of 35 bytes is used (theoretical transmission protocol, for example). Also, consider that it is desired to send protocol packets as large as possible. That is, the cycle packet size when rounded will vary between 350000 and 700000 bytes. This can cause a relatively high burstiness, depending on the chosen wait time t_w. Alternatively, the choice can be made to send protocol packets of 35000 bytes (10 times less), with cycle packet sizes of 595000 and 630000 bytes, respectively. Since the overhead size per protocol packet is very small (35 bytes), the content data is efficiently sent and the burstiness is dramatically decreased (considering the same wait time t_w). <ul id="ul0008" list-style="none" compact="compact"><li>2. Because of the impossibility of creating packets with a size granularity smaller than d_min, in S107 the minimal and maximal packet sizes are determined, and computed as follows: <maths id="math0006" num="(Eq.2)"><math display="block"><mi>a</mi><mo>=</mo><mfrac><msub><mi>P</mi><mi mathvariant="italic">theoretical</mi></msub><msub><mi>d</mi><mi>min</mi></msub></mfrac></math><img file="EP1584162B1_D0006.tif" /></maths></li><li>3. In S109, the computed value a is rounded to an integer by defect (round down), to obtain an integer value b used for S110.</li><li>4. In S111, the computed value a is rounded to an integer by excess (round up) to obtain an integer value c used for S112 Note that b ≤ a and c ≥ a, therefore: b ≤ a ≤ c.</li><li>5. In S115 a minimum packet size value P_min is computed as: <maths id="math0007" num="(Eq.3)"><math display="block"><msub><mi>P</mi><mi>min</mi></msub><mo>=</mo><mi>b</mi><mo>⋅</mo><msub><mi>d</mi><mi>min</mi></msub></math><img file="EP1584162B1_D0007.tif" /></maths></li><li>6. In S117 a maximum packet size value P_max is computed as: <maths id="math0008" num="(Eq.4)"><math display="block"><msub><mi>P</mi><mi>max</mi></msub><mo>=</mo><mi>c</mi><mo>⋅</mo><msub><mi>d</mi><mi>min</mi></msub></math><img file="EP1584162B1_D0008.tif" /></maths> Note that P_min ≤ P_theoretical ≤ P_max.</li></ul>

After having found the minimum P_min and maximum P_max packet sizes, it is necessary to find the right proportion of such packets to be used during transmission of the content in order to achieve on average a transmission of the content at the target bandwidth B_T. That is, on average (of the P_min and P_max packets) the packet size must be as near as possible to P_theoretical.

The purpose of this also can be described as trying to distribute the packets in order to maximize homogeneity, i.e., decrease burstiness as much as possible.

To find the proportional number of packets having sizes P_min and P_max, the following steps are followed: <ul id="ul0009" list-style="none" compact="compact"><li>7. In S119 a value m is computed as: <maths id="math0009" num="(Eq.5)"><math display="block"><mi mathvariant="normal">m</mi><mo mathvariant="normal">=</mo><msub><mi mathvariant="normal">P</mi><mi>theoretical</mi></msub><mo mathvariant="normal">-</mo><msub><mi mathvariant="normal">P</mi><mi>min</mi></msub></math><img file="EP1584162B1_D0009.tif" /></maths></li><li>8. In S121, a value n is computed as: <maths id="math0010" num="(Eq.6)"><math display="block"><mi mathvariant="normal">n</mi><mo mathvariant="normal">=</mo><msub><mi mathvariant="normal">P</mi><mi>max</mi></msub><mo>-</mo><msub><mi mathvariant="normal">P</mi><mi>theoretical</mi></msub></math><img file="EP1584162B1_D0010.tif" /></maths> where each of m and n is a whole number</li><li>9. The values m and n determine the population of the different size packets. That is: <ul id="ul0010" list-style="none" compact="compact"><li>if m &lt; n there will be a majority of packets with packet size P_min;</li><li>if n &lt; m there will be a majority of packets with packet size P_max; and</li><li>If m = n there will be the same number of packets with packet size P_min as packets with packet size P_max.</li></ul> Note that if m and n are equal to zero, the size of all packets will be equal to P_theoretical.</li><li>10. In S123 the value <maths id="math0011"><math display="inline"><mfrac><mi>n</mi><mi>m</mi></mfrac></math><img file="EP1584162B1_D0011.tif" /></maths> is reduced to the lowest common denominator, in order to obtain values p in S124 and q in S126 to restate the value <maths id="math0012"><math display="inline"><mfrac><mi>n</mi><mi>m</mi></mfrac></math><img file="EP1584162B1_D0012.tif" /></maths> in the form <maths id="math0013"><math display="inline"><mfrac><mi>q</mi><mi>p</mi></mfrac><mn>.</mn></math><img file="EP1584162B1_D0013.tif" /></maths> That is (mathematically), there exists an integer x ≥ 1 for which: <maths id="math0014"><math display="block"><mfrac><mi>n</mi><mi>m</mi></mfrac><mo>=</mo><mfrac><mrow><mi>q</mi><mo>⋅</mo><mi>x</mi></mrow><mrow><mi>p</mi><mo>⋅</mo><mi>x</mi></mrow></mfrac></math><img file="EP1584162B1_D0014.tif" /></maths></li><li>11. In S127, a value <i>cycle_packets</i> (the number of packets contained in a cycle) is computed as: <maths id="math0015" num="(Eq.7)"><math display="block"><mi mathvariant="italic">cycle_packets</mi><mo>=</mo><mi mathvariant="normal">p</mi><mo mathvariant="normal">+</mo><mi mathvariant="normal">q</mi></math><img file="EP1584162B1_D0015.tif" /></maths> That is, a cycle is composed in Step S130 of: <ul id="ul0011" list-style="none" compact="compact"><li><i>cycle_packets</i> packets, with: <ul id="ul0012" list-style="none" compact="compact"><li>q packets with packet size P_min, and</li><li>p packets with packet size P_max.</li></ul> Note that, if p and q are equal to zero (that is, the packet size is P_theoretical for all packets) no cycle will be performed, but the content will be divided into packets of constant size and sent with a time interval t_w between successive packets.</li></ul> The transmission of content involves transmission of many cycles composed of a number of cycle_packet packets until the last packet of the content is sent. Note that the size of the last packet can be smaller than P_max and also smaller than P_min, depending on how many bytes must be sent in order to complete the content transmission. It is preferred to achieve transmission homogeneity. That is, the packets transmitted in a cycle are to be distributed in a smoother sequence than simply sending q packets with packet size P_min and p packets with packet size P_max. To do this, reference is made to <figref idref="f0002">Fig. 1B</figref><figref idref="f0002">Fig. 1B</figref> (steps S131-S183) depicts computation of the distribution of packets during a cycle. The purpose of this proportional distribution is that the instantaneous bandwidth must be as near as possible to the target bandwidth, i.e., to decrease as much as possible the burstiness. Referring to <figref idref="f0002">Fig. 1B</figref>, the following steps are performed: </li><li>12. Using the values p from S124 and q from S126, a value z is computed in one of S 131, S 133, S 135 by testing q against p to produce z where: <maths id="math0016" num="(Eq.8)"><math display="block"><mi>In S</mi><mo>⁢</mo><mn>132</mn><mspace width="1em" /><mi>z</mi><mo>=</mo><mfrac><mi>q</mi><mi>p</mi></mfrac><mo>,</mo><mi>if q</mi><mo>&gt;</mo><mi mathvariant="normal">p</mi></math><img file="EP1584162B1_D0016.tif" /></maths><maths id="math0017"><math display="block"><mi>In S</mi><mo>⁢</mo><mn>134</mn><mspace width="1em" /><mi>z</mi><mo>=</mo><mfrac><mi>p</mi><mi>q</mi></mfrac><mo>,</mo><mi>if p</mi><mo>&gt;</mo><mi mathvariant="normal">q</mi></math><img file="EP1584162B1_D0017.tif" /></maths> In <i>S136 z</i> = 1, if p = q (also in the case p = q = 0) </li><li>13. In S139 the quantity z is rounded by defect (down) to the nearest integer value to produce the value prop in S140.</li><li>14. Using the results of the tests of p against q of S131, S133 and S 135 in S160 a value r is computed which in a cycle composed of <i>cycle_packets</i> packets, where r is the number of "isolated" packets as follows: <maths id="math0018" num="&lt;segemimath dpn=&quot;9&quot; segid=&quot;creer4&quot; xbd=&quot;1710&quot; xhg=&quot;796&quot; (Eq.9)"><math display="block"><mi mathvariant="normal">r</mi><mo mathvariant="normal">=</mo><mi mathvariant="normal">q</mi><mo>-</mo><mfenced><mi mathvariant="italic">prop</mi><mo>*</mo><mi mathvariant="normal">p</mi></mfenced><mo>,</mo><mi>if q</mi><mo>&gt;</mo><mi mathvariant="normal">p</mi></math><img file="EP1584162B1_D0018.tif" /></maths> r = p - (<i>prop</i> * q), if p &gt; q r = 0, if p = q</li><li>15. In a cycle, the packets will be sent in the following sequence, using the results of the p against q test of S131, S133 and S135, as set forth in a), b) and c) below: <ol id="ol0003" compact="compact"><li>a) If q &gt; p: <ul id="ul0013" list-style="none" compact="compact"><li>then loop p times as controlled by S 160, send prop P_min packets in S161, send 1 P_max packet in S163, and send r P_max packets in S 165</li></ul></li><li>b) If p &gt; q: <ul id="ul0014" list-style="none" compact="compact"><li>then loop q times as controlled by S170, send prop P_max packets in S 171, send 1 P_min packet in S 173, and send r P_min packets S 175</li></ul></li><li>c) If p = q: <ul id="ul0015" list-style="none" compact="compact"><li>then loop the number of cycle_packets times as controlled by S180, send 1 P_max packet in S181, and send 1 P_min packet S183. In this way the process is complete: the transmission is composed of many cycles of <i>cycle_packets</i> number of packets.</li></ul></li></ol></li></ul>

In a cycle, the packet transmissions are distributed as described in point 16 of Example 1 below.

It is typical for an application program sending data on a network to use network protocols in order to address the packets to the receiver devices. Network protocols typically have a fixed maximum protocol packet size. For example, for the ATM protocol this size (size of a cell) is 53 Bytes. This means that the herebefore called packets are blocks of data which must be fragmented into protocol packets depending on the maximal protocol packet size. For example, if a packet of size 200 Bytes must be sent on an ATM protocol network, it will be split into 3 protocol packets of 53 Bytes (max protocol packet size for ATM) and one protocol packet of 41 Bytes. This does not impact the process described above relative to the determination of the number of packets of each of P_max and P_min to be transmitted during a cycle, since this fragmentation of a transmitted packet into its constituent smaller size protocol packet does not impact the transmission of the protocol packets in a way that these can be considered as a single larger packet.

On average during a content transmission, the packets size would be P_theoretical and therefore the bandwidth used equal to B_T.

An increase of the minimal data block size d_min handled by the application program and/or of the time interval between the successive packets can cause an increase of the instantaneous bandwidth (depending on the transmission sample rate). Also, an increase of the minimal data block size handled can cause an increase of the difference of the value P_max - P_min, leading to an increase of the burstiness.

A refinement of the invention used in accommodating data protocol packet size is described referring to <figref idref="f0003">Figs. 1C</figref> and <figref idref="f0004">1D</figref>.

<figref idref="f0003">Fig. 1C</figref> depicts the composition of packets according to transmission protocol packets. That is, the method and system of the invention can be applied to any transmission protocol. Here, the purpose of the process is to compute how many protocol packets (e.g., UDP packets) must be put together to create a cycle packet. As defined herein, each protocol packet is of a size ppsize. This computation is done for both P_max and P_min. This is important because, when implementing a software application, the transmission control is implemented at transmission protocol level, i.e., the programmer can control how many packets of a given protocol are sent. In this way, the programmer can say, for example, a cycle packet of size P_max is composed by a number of UDP packets. The number of the protocol packets is computed as described below. It should be noted that for every properly working software application, d_min must be equal or smaller than ppsize. Otherwise, the application is not able to handle data blocks of size ppsize. That is, the application is not able to transmit using the chosen transmission protocol (with protocol packet size ppsize).

In S200, the protocol packet size <i>ppsize</i> [Bytes] is used as an input parameter. The protocol packet size ppsize represents the maximum protocol packet size for those protocols, which can have a varying protocol packet size (e.g., UDP). Other transmission protocols have fixed protocol packet size (e.g., ATM). Knowing the values of P_min and P_max from S 115 and S117, the packet composition at network protocol level can be computed as follows: <ol id="ol0004" compact="compact"><li>1. Compute in S203 and S205 the values <i>v1</i> and <i>v2</i> as: <maths id="math0019" num="(Eq.10)"><math display="block"><mi>v</mi><mo>⁢</mo><mn>1</mn><mo>=</mo><mfrac><msub><mi>P</mi><mi>max</mi></msub><mi mathvariant="italic">ppsize</mi></mfrac></math><img file="EP1584162B1_D0019.tif" /></maths><maths id="math0020" num="(Eq.11)"><math display="block"><mi>v</mi><mo>⁢</mo><mn>2</mn><mo>=</mo><mfrac><msub><mi>P</mi><mi>min</mi></msub><mi mathvariant="italic">ppsize</mi></mfrac></math><img file="EP1584162B1_D0020.tif" /></maths></li><li>2. In S204, round <i>v1</i> by defect [down] to the nearest integer value, obtaining <i>v3</i> in S205.</li><li>3. In S206, round <i>v2</i> by defect to the nearest integer value, obtaining <i>v4 in S207</i>.</li><li>4. Compute in S211 <i>maxrest</i> as: <maths id="math0021" num="(Eq.12)"><math display="block"><mi mathvariant="italic">maxrest</mi><mo>=</mo><mfenced><msub><mi mathvariant="normal">P</mi><mi>max</mi></msub><mo>-</mo><mi mathvariant="italic">v</mi><mo>⁢</mo><mn mathvariant="italic">3</mn><mo mathvariant="italic">*</mo><mi mathvariant="italic">ppsize</mi></mfenced></math><img file="EP1584162B1_D0021.tif" /></maths></li><li>5. Compute in S21 3 <i>minrest</i> as: <maths id="math0022" num="(Eq.13)"><math display="block"><mi mathvariant="italic">minrest</mi><mo>=</mo><mfenced><msub><mi mathvariant="normal">P</mi><mi>min</mi></msub><mo>-</mo><mi mathvariant="italic">v</mi><mo>⁢</mo><mn mathvariant="italic">4</mn><mo mathvariant="italic">*</mo><mi mathvariant="italic">ppsize</mi></mfenced></math><img file="EP1584162B1_D0022.tif" /></maths></li><li>6. As shown in S215, a packet of size P_max will be split into: v3 protocol packets of size <i>ppsize,</i> and 1 protocol packet of size <i>maxrest.</i></li><li>7. As shown in S217, a packet of size P_min will be split into: v4 protocol packets of size <i>ppsize,</i> and 1 protocol packet of size <i>minrest</i>.</li></ol>

Typically, the address of the receiver devices to which the content is to be transmitted is contained in the same packet in which the content data is stored. The part of the packet which contains the address of the receiving devices and other information (not belonging to the content to be sent) is hereafter called overhead.

A protocol packet sent on the network will therefore contain a given overhead of a value "overhead size", input in S220, and a given amount of content data. The amount of content data sent in a packet of size P_max and P_min, respectively, depends on the number of protocol packets resulting from the splitting process needed by the network protocol, described above, and also on the overhead of every protocol packet.

As is described referring to <figref idref="f0004">Fig. 1D</figref>, this process computes the data content of a given packet. In fact, common transmission protocols use a header, containing protocol data (e.g., the receiver address). In this header, no "useful" data of the content to be sent is contained. The issue is still that when implementing a software application, the programmer must be able to say how much content should be put in a packet. This allows knowing how much space of a packet is reserved for "useful" data and how much space is spent for header information.

Referring to <figref idref="f0004">Fig. 1D</figref>, in S220, given as an input parameter the size of the overhead <i>overhead_size</i> [Bytes], the content data contained in the protocol packets are: <ul id="ul0016" list-style="none" compact="compact"><li>8. for computing the information in the packet of size P_max: <i>v3</i> protocol packets containing (<i>ppsize - overhead_size)</i> bytes of content data, from S204, and 1 protocol packet containing (<i>maxrest - overhead_size</i>) bytes of content data. That is, in a packet of size P_max, the actual bytes of content data <i>cdpmax</i> are given in S230 by: <maths id="math0023" num="(Eq.14)"><math display="block"><mrow><mi mathvariant="italic">cdpmax</mi><mo mathvariant="italic">=</mo><mi mathvariant="italic">v</mi><mo>⁢</mo><mn mathvariant="italic">3</mn><mo mathvariant="italic">*</mo><mfenced><mi mathvariant="italic">ppsize</mi><mo mathvariant="italic">-</mo><mi mathvariant="italic">overhead_size</mi></mfenced><mo mathvariant="italic">+</mo><mo mathvariant="italic">(</mo><mi mathvariant="italic">maxrest</mi><mo mathvariant="italic">-</mo><mi mathvariant="italic">overhead_size</mi><mo mathvariant="italic">)</mo></mrow></math><img file="EP1584162B1_D0023.tif" /></maths></li><li>9. for computing the information in the packet of size P_min: <ul id="ul0017" list-style="none" compact="compact"><li><i>V4</i>, from S206, protocol packets containing (<i>ppsize - overhead_size)</i> bytes of content data, and</li><li>In S240, it is determined that 1 protocol packet containing (<i>minrest - overhead_size</i>) bytes of content data.</li></ul></li></ul>

That is, in a packet of size P_min, the actual bytes of content data <i>cdpmin</i> are given in S240 by: <maths id="math0024" num="(Eq.15)"><math display="block"><mrow><mi mathvariant="italic">cdpmin</mi><mo mathvariant="italic">=</mo><mi mathvariant="italic">v</mi><mo>⁢</mo><mn mathvariant="italic">4</mn><mo mathvariant="italic">*</mo><mfenced><mi mathvariant="italic">ppsize</mi><mo mathvariant="italic">-</mo><mi mathvariant="italic">overhead_size</mi></mfenced><mo mathvariant="italic">+</mo><mo mathvariant="italic">(</mo><mi mathvariant="italic">minrest</mi><mo mathvariant="italic">-</mo><mi mathvariant="italic">overhead_size</mi><mo mathvariant="italic">)</mo></mrow></math><img file="EP1584162B1_D0024.tif" /></maths>

EXAMPLE 1

Reference is made to the equations above and also to F<figref idref="f0001 f0002 f0003 f0004">igs. 1A-1D</figref>. Reference is also made to <figref idref="f0005">Fig. 2</figref> which illustrates the proportional distribution of the different packets during a cycle, in order to decrease burstiness (i.e., the instantaneous bandwidth must be as near as possible to the target bandwidth). <figref idref="f0005">Fig. 2</figref> further shows a pictorial representation of some of the results of the Example. In the example: <ol id="ol0005" compact="compact"><li>(a) the application program is to send content with a target bandwidth B_T of 15 Mbps.</li><li>(b) the machine on which the program is installed can reach a high precision time granularity of 100 milliseconds (= t_w).</li><li>(c) the application program will send UDP protocol packets on a network to one or many receiver devices (depending on what is specified by the user).</li><li>(d) the minimal data block d_min size that can be handled is fixed as 177 Bytes.</li></ol>

Solution

:

UDP protocol packets header is 28 Bytes (20 bytes for IP and 8 bytes for UDP). The maximum size of UDP protocol packets is 65535 Bytes.

The input parameters are summarized as: <ul id="ul0018" list-style="none" compact="compact"><li>B_T = 15 Mbps = 15000000 bps = 1875000 Bytes per second (1 byte = 8 bits)</li><li>t_w = 100 milliseconds = 0.1 seconds</li><li>d_min = 177 Bytes</li><li><i>ppsize</i> = 65535 Bytes</li><li><i>overhead_size =</i> 28 Bytes</li></ul>

Following the process described above: <ol id="ol0006" compact="compact"><li>1. P_theoretical = 1875000 * 0.1 = 187500 Bytes</li><li>2. a = 187500 / 177 = 1059.322</li><li>3. b = 1059</li><li>4. c = 1060</li><li>5. P_min = 1059 * 177 = 187443 Bytes</li><li>6. P_max = 1060 * 177 = 187620 Bytes</li><li>7. m = 187500 - 187443 = 57</li><li>8. n = 187620 - 187500 = 120</li><li>9. m &lt; n : there will be a majority of packets with packet size P_min</li><li>10. n/m is already reduced to the lowest common integer multiple: p = 57, q = 120</li><li>11. <i>cycle_packets</i> = 57 + 120 = 177</li><li>12. A cycle will be composed of 177 packets, with 120 packets of packet size 187443 Bytes and 57 packets of packet size 187620 Bytes.</li><li>13. q &gt; p, therefore z = q / p = 2.105263</li><li>14. <i>prop = 2</i></li><li>15. q &gt; p, therefore r = 6</li><li>16. In a cycle, the packets will be sent in the following sequence: <ul id="ul0019" list-style="none" compact="compact"><li>Repeat 57 times: <ul id="ul0020" list-style="none" compact="compact"><li>send 2 packets with packet size 187443 Bytes (P_min)</li><li>send 1 packet with packet size 187620 Bytes (P_max)</li></ul></li><li>then send 6 packets with packet size 187443 Bytes. (P_max)</li></ul></li></ol>

Protocol packets composition

:

<ol id="ol0007" compact="compact"><li>1. v1 = 2.8629, v2 = 2.8602</li><li>2. v3 = 2</li><li>3. v4 = 2</li><li>4. maxrest = 56550 Bytes</li><li>5. minrest = 56373 Bytes</li><li>6. A packet of size 187620 Bytes will se split into: <ul id="ul0021" list-style="none" compact="compact"><li>2 UDP packets of size 65535 Bytes, and</li><li>1 UDP packet of size 56550 Bytes</li></ul></li><li>7. A packet of size 187443 Bytes will be split into: <ul id="ul0022" list-style="none" compact="compact"><li>2 UDP packets of size 65535 Bytes, and</li><li>1 UDP packet of size 56373 Bytes</li></ul></li><li>8. For the packet of size 187620 Bytes, the actual content data are 187536 Bytes (≈ 99.95523% of the packet size).</li><li>9. For the packet of size 187443 Bytes, the actual content data are 187359 Bytes (≈ 99.95519% of the packet size).</li></ol>

<figref idref="f0006">Fig. 3</figref> is a diagram that explains the transmission of the data packets. At the sending station there is a computer 10 that has a broadcast application program installed. The broadcast application program operates on a content 12 that is to be transmitted in data packets. The parameters of the transmission target bandwidth B_T, p, and wait time t_w are input to the computer 10, as explained above. The computer computes the packet sizes as described with respect to <figref idref="f0001 f0002 f0003 f0004">Figs. 1A-1D</figref> and operates to control the transmission 1 6 of packets 18 (black bars), shown as having varying sizes.

The packets 1 8 are received at a computer 20 that has a receiving application program installed. This program reassembles the received packets into a file that corresponds to the content file 12 that was sent. This is conventional.

Specific features of the invention are shown in one or more of the drawings for convenience only, as each feature may be combined with other features in accordance with the invention. Alternative embodiments will be recognized by those skilled in the art and are intended to be included within the scope of the claims.