CN106878192A - A kind of data dispatching method of adaptive M PTCP - Google Patents

Abstract

The invention discloses a kind of data dispatching method of adaptive M PTCP, including：Congestion window according to each subflow sets up Markov model；The average congestion window size of each subflow is calculated according to Markov model；The current two-way time value of each subflow is calculated according to timestamp field；The average effective throughput in each subflow is obtained according to the average congestion window size of each subflow and the prediction of two-way time value；The congestion window size of minimum subflow is dynamically adjusted according to the average effective throughput in each subflow, minimum subflow is avoided congestion；The size of ssthresh is adjusted according to the congestion window size after adjustment so that the subflow after adjustment is still in congestion avoidance phase.The data dispatching method of described adaptive M PTCP overcomes the deficiency of existing MPTCP data dispatch and Research of Congestion Control Techniques, enable that MPTCP is adaptively more reasonably dispatched for network fluctuation and the situation of network isomerization, can finally improve the efficiency and reliability of multi-path data transmission.

Description

Data scheduling method of self-adaptive MPTCP

Technical Field

The invention relates to the technical field of data scheduling and congestion control, in particular to a data scheduling method of self-adaptive MPTCP.

Background

In recent years, with the more and more extensive application of intelligent terminals and the continuous development of internet communication, the demand of network application on traffic is also more and more increased; network applications represented by big data applications, mobile televisions, video calls, and the like, are in urgent need of a high-reliability fast data transmission technology as a support. The existing internet uses the TCP/IP protocol suite, of which the transport layer is most widely studied and used, represented by the TCP protocol. However, the conventional TCP protocol cannot simultaneously utilize multiple communication interfaces (such as an ethernet interface and a WIFI interface) of the intelligent terminal, only one transmission path can be established by using one interface during end-to-end connection, and cannot meet the requirement of increasing throughput, more importantly, the single-path transmission reliability is poor, if interference or transmission congestion occurs, the transmission performance is sharply reduced, and if the transmission congestion is serious, transmission is interrupted. The Multipath-TCP (MPTCP) was developed to make up for the TCP deficiency. The protocol is a multipath parallel transmission protocol based on the traditional TCP, can be compatible with the existing middleware, and has the characteristics of high reliability, high fault tolerance, high throughput, high safety, compatibility with the TCP and the like.

The end-to-end equipment places data on multiple paths through the MPTCP for parallel transmission, and the throughput and the robustness of data transmission can be improved by jointly utilizing multiple interfaces. When data distribution among multiple paths is performed, MPTCP needs to perform path selection in advance among multiple available links, a transmission path with poor link quality needs to be abandoned, and a transmission path with better link quality is selected, so that difference of transmission quality among paths is reduced. On one hand, if a transmitting end does not abandon a poor transmission path in the multipath transmission process, data transmission on the path is easy to cause the problems of timeout, errors, packet loss, retransmission and the like of data packets, a receiving end sequences and combines the data packets received on each path, even if the data transmission on other paths is normal, incomplete data caused by delayed arrival of the data packets on the path can cause delayed delivery to an upper layer application, and further the overall transmission performance of the MPTCP between the transmitting end and the receiving end can be influenced. On the other hand, if the difference in transmission quality between the paths is too large, the data packet carried by the poor path is lost, and the data packet received on the better path will quickly fill the data buffer of the receiving end, thereby causing the problems of data transmission suspension and overtime retransmission of a large amount of data on the better path. Therefore, in MPTCP data transmission, it is very important to schedule data onto reasonable paths and to handle congestion that may occur.

In the existing MPTCP protocol, an evaluation index of path quality mainly utilizes transmission delay of data on the path, the transmission delay is one of visual indexes representing the path quality, and an evaluation method for the index is to measure a Round-Trip Time (RTT) value of the path. RTT on a path represents the total time delay from the start of sending a packet on the path by the sender to the time the sender receives an acknowledgement for the datagram from the receiver on the path (the receiver sends an acknowledgement immediately after receiving the packet). The smaller the RTT, the higher the transmission quality of the path, and vice versa.

The inventor finds that, in the process of implementing the invention, the RTT is used to evaluate the path quality, and then the path selection is performed, although the measurement method is relatively simple and the accuracy is relatively high, the following disadvantages are at least present: 1) when measuring RTT, it needs to send redundant data packet or start actual data transmission, the former measures RTT by sending special data packet and occupies bandwidth resource of path, the latter is equivalent to actually using the path to transmit data, measures RTT by sending data packet in transmission process, if path quality is poor, it can directly cause data loss of receiving end; 2) RTT measurement relates to a process of sending a data packet, a process of receiving and processing a data packet at a receiving end, a process of sending an acknowledgement packet at a receiving end, and a process of receiving an acknowledgement packet and processing an acknowledgement packet at a sending end, and measurement time at least includes round trip time of a data packet, so that there is a certain time delay in measurement. The worse the path quality, the longer the corresponding measurement delay, which will reduce the efficiency of MPTCP path selection.

Disclosure of Invention

In view of this, the present invention is to provide a data scheduling method for adaptive MPTCP, which can improve the efficiency and reliability of multipath data transmission.

Based on the above purpose, the present invention provides a data scheduling method for adaptive MPTCP, which includes:

establishing a Markov model according to the congestion window of each sub-flow;

calculating to obtain the average congestion window size of each sub-flow according to a Markov model;

calculating to obtain the current round trip time value of each sub-flow according to the timestamp field in the data packet;

predicting to obtain the average effective throughput rate on each sub-flow according to the average congestion window size and the round trip time value of each sub-flow;

dynamically adjusting the size of a congestion window of the minimum sub-flow according to the average effective throughput rate on each sub-flow, so that the minimum sub-flow is prevented from being congested;

and adjusting the size of the slow start threshold according to the adjusted size of the congestion window, so that the adjusted sub-flow is still in a congestion avoidance stage.

Optionally, the markov model comprises: the sending state set of the equipment at each moment and the one-step transition probability set of the sending state conversion; wherein, the sending state set comprises the size of the congestion window of each sub-flow; the transition probability set is related to the packet loss rate of each sub-stream.

Optionally, the step of calculating the average congestion window size of each substream according to the markov model includes:

obtaining a transition probability matrix of the sending state of the equipment according to the packet loss rate of each sub-flow;

calculating to obtain ultimate stable distribution according to the probability transition rule of the transition probability matrix;

and calculating the average congestion window size of each sub-flow according to the limit stable distribution.

Optionally, the extreme stationary distribution is recorded as a feature vector μ; and the feature vector meets the following constraints:

μ＝(μ₁,μ₂…μ_v)

μP＝μ

μ₁+μ₂+…+μ_v-1+μ_v＝1

μ_j≥0,j∈{1,2…v}

wherein, the Maximum congestion window limited by the size of the receiving buffer is W MSS (Maximum segment) units, and the minimum congestion window is 2 MSS units, so the value range of the sub-flow congestion window is [2, W]. Let k be the number of substreams, cw_i(i∈[1,k]And I ∈ Z) is the congestion window size of substream I, the transmission state of the system is represented by each substream congestion window set, and it can be seen that the transmission state space is a k-dimensional discrete state space and is represented by I { (cw)₁,cw₂,…cw_i…,cw_k)|cw_i∈[2,W]And cw_i∈ Z, the known state space I contains (W-1)^kOne possible element, for convenience of presentation, remember v ═ W-1)^k；μ_jIs the stationary distribution probability on the jth state in the state space; p is a transition probability matrix, and the transition probability matrix is a sparse matrix in (v x v) dimension;

the calculation expression of the average congestion window size is as follows:

wherein I is the sequence number of the sub-stream, I is the sending state space, and j is the state sequence number in the state space I; max is the number of elements of the feature vector, i.e., (W-1)^k；The congestion window size of the sub-stream I in the jth state in the state space I; mu.s_jThe probability of the stationary distribution corresponding to the jth state;is the average congestion window size for sub-stream i.

Optionally, the step of calculating the current round trip time value of each sub-stream according to the timestamp field in the data packet further includes:

smoothing the round trip time value;

the expression of the smoothing process is:

RTT_t＝(1-α)RTT_new+αRTT_t-1

0<α<1

wherein RTT_newIs the round trip time value of the current moment; RTT (round trip time)_t-1Is the round trip time value of the previous moment; RTT (round trip time)_tWhich is the smoothed round trip time value, α is the factor by which the smoothed round trip time value is affected by the current round trip time value.

Optionally, the influence factor α is 0.8

Optionally, the calculation expression of the average goodput rate over each sub-stream is:

wherein p is_iIs the packet loss rate on sub-stream i; MSS_iIs the maximum transmission segment size on substream i;is the average congestion window size over sub-stream i; RTT (round trip time)_iIs the round trip time value on sub-stream i;is the average goodput over substream i.

Optionally, the step of dynamically adjusting the size of the congestion window of the minimum sub-stream according to the average goodput rate on each sub-stream includes:

calculating to obtain a throughput rate coefficient according to the average effective throughput rate on each sub-flow; wherein the throughput rate coefficient is the ratio of the maximum sub-stream throughput rate to the minimum sub-stream throughput rate;

and judging whether the throughput rate coefficient is larger than the throughput rate critical value, if so, correspondingly reducing the size of the congestion window of the minimum sub-flow.

Optionally, the step of calculating a throughput coefficient according to the average goodput rate on each sub-stream further includes: and smoothing the throughput rate coefficient, wherein the expression of the smoothing is as follows:

θ_t＝(1-β)θ_new+βθ_t-1

wherein β is a balance factor, theta_newThe current time throughput rate coefficient; theta_t-1Counting an average coefficient for the throughput rate of the previous moment; theta_tIs the throughput coefficient after the smoothing process.

Optionally, the calculation expression of the throughput threshold is:

(G_max-G_min)+1＜N_Buf＜(G_max-G_min)+2

so as to obtain the compound with the characteristics of,

wherein N is_BufFor the size of the receiving end buffer, G_maxFor maximum sub-stream throughput, G_minIs the minimum substream throughput rate.

From the above, it can be seen that, in the data scheduling method of the adaptive MPTCP provided by the present invention, a markov model is constructed based on the congestion window of each sub-stream, and then the average congestion window size of each sub-stream is obtained through calculation, and then the average effective throughput rate on each sub-stream is obtained through prediction based on the average congestion window size and the round trip time value, so that the size of the congestion window can be adjusted according to the predicted throughput rate, and finally data scheduling and congestion control are realized. On the basis of a heterogeneous network multipath parallel transmission technology, the congestion windows of substreams corresponding to each interface of the wireless mobile equipment and the data scheduling behaviors which can be adopted are researched, the probability model based on the random event accumulation times is established to analyze the state transition probability, a predictable throughput rate feedback model is obtained accordingly, and the data is scheduled to the optimal interface by adopting the scheduling strategy with the optimal throughput rate, so that the advantages of multipath parallel transmission high transmission performance are exerted, and the reliability of data transmission is enhanced. Therefore, the data scheduling method of the self-adaptive MPTCP overcomes the defects of the existing MPTCP data scheduling and congestion control technology, so that the MPTCP can adaptively perform more reasonable scheduling aiming at the conditions of network fluctuation and network isomerization, and finally the efficiency and reliability of multi-path data transmission can be improved.

Drawings

Fig. 1 is a flowchart of an embodiment of a data scheduling method of adaptive MPTCP according to the present invention;

fig. 2 is a flowchart of another embodiment of a data scheduling method of adaptive MPTCP according to the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to specific embodiments and the accompanying drawings.

It should be noted that all expressions using "first" and "second" in the embodiments of the present invention are used for distinguishing two entities with the same name but different names or different parameters, and it should be noted that "first" and "second" are merely for convenience of description and should not be construed as limitations of the embodiments of the present invention, and they are not described in any more detail in the following embodiments.

The embodiment of the invention mainly relates to a data scheduling and congestion control technology of a Multi-path transmission control Protocol (Multi-path transmission control Protocol), in particular to self-adaptive data scheduling and congestion control under the conditions of network fluctuation and network differentiation realized by an optimized MPTCP throughput prediction method, thereby realizing high reliability of Multi-path data transmission. Aiming at the defects that the traditional TCP transmission only can utilize a single network interface and the limitations of MPTCP transmission data scheduling and congestion control technology, the invention provides a Markov model based on the discrete state of each substream congestion window of MPTCP, establishes a Markov decision process and scientifically predicts the effective throughput rate of each substream, and then provides a self-adaptive congestion window adjusting scheme facing the conditions of network fluctuation and network differentiation, thereby indirectly realizing more reasonable data scheduling and improving the reliability of data transmission in MPTCP.

Specifically, referring to fig. 1, a flowchart of an embodiment of a data scheduling method of adaptive MPTCP according to the present invention is shown. Aiming at the defect that the traditional MPTCP simply carries out data scheduling according to RTT, the invention aims to reasonably schedule data while considering a congestion window by establishing a Markov prediction model of the congestion window of each substream of the MPTCP and combining real-time RTT to realize a dynamic throughput rate prediction algorithm, thereby reducing data packet retransmission, packet loss and congestion and realizing high transmission reliability. The data scheduling method of the self-adaptive MPTCP comprises the following steps:

step 101, establishing a Markov model according to a congestion window of each substream; firstly, on the basis of assuming that data are uniformly scheduled to each substream, combining the packet loss rate of each substream, and integrating the discrete states of the congestion windows of each substream of MPTCP to establish a Markov model.

Optionally, the markov model comprises: the sending state set of the equipment at each moment and the one-step transition probability set of the sending state conversion; wherein, the sending state set comprises the size of the congestion window of each sub-flow; the transition probability set is related to the packet loss rate of each sub-stream. Specifically, let S represent a set of transmission states, and for a multipath parallel transmission device using k network interfaces, the transmission state S of the device at time t is represented by k elements, and the k elements record the congestion window sizes of k transmission sub-streams respectively. For example: CW_iIndicating the congestion window size of the ith sub-stream, the transmission state s of the device at time t_iDenoted as set S { CW₁,CW₂…CW_k}. Let P(s)_t|s_t-1) Representing the wireless mobile device transmitting state s at time t-1_t-1And at time t the transmission state is switched to s_tThe one-step transition probability. Since the transmission state depends on the size of the congestion window for all sub-streams, which is related to the packet loss rate of each sub-stream, P(s)_t|s_t-1) Indirectly limited by the packet loss rate of each sub-stream within an interval of Δ t, where Δ t represents the time interval between the sending end receiving two adjacent ACK acknowledgement packets.

102, calculating the average congestion window size of each substream according to a Markov model; based on the rule followed by the Markov model, probability distribution that all the sub-flows are finally stabilized in the size of the corresponding congestion window can be determined, so that ultimate stable distribution is obtained, and the size of the average congestion window can be calculated finally.

Optionally, the step of calculating the average congestion window size of each substream according to the markov model includes: obtaining a transition probability matrix of the sending state of the equipment according to the packet loss rate of each sub-flow; calculating to obtain ultimate stable distribution according to the probability transition rule of the transition probability matrix; and calculating the average congestion window size of each sub-flow according to the limit stable distribution. Specifically, in addition to the binary group model (S, P) in the Markov model, each subgroup is first identifiedThe flow packet loss rate deduces the transition probability matrix P of the equipment sending state, it is limited by the Maximum congestion window of the receiving buffer Size as W MSS (Maximum Segment Size) unit, and the minimum congestion window is 2 MSS unit, then the value range of the sub-flow congestion window Size is [2, W]. Let k be the number of substreams, cw_i(i∈[1,k]And I ∈ Z) is the congestion window size of substream I, the transmission state of the system is represented by each substream congestion window set, and it can be seen that the transmission state space is a k-dimensional discrete state space and is represented by I { (cw)₁,cw₂,…cw_i…,cw_k)|cw_i∈[2,W]And cw_i∈ Z, the known state space I contains (W-1)^kOne possible element, for convenience of presentation, remember v ═ W-1)^k. The transition probability matrix P is a sparse matrix in the dimension (v x v). From this, a limit stationary distribution is calculated, i.e. a probability distribution where all sub-streams eventually settle to the size of the corresponding congestion window, under the probability transition law following the P matrix. The size of the stationary congestion window for each substream can be obtained from the extreme stationary distribution of the Markov model, which is taken as the feature vector μ, where μ is a vector with v elements (μ₁,μ₂…μ_v). The feature vector μ satisfies three constraints:

μP＝μ

μ₁+μ₂+…+μ_v-1+μ_v＝1

μ_j≥0,j∈{1,2…v}

wherein v is the number of smoothly distributed elements, i.e., v ═ W-1)^k；μ_jIs the stationary distribution probability on the jth state in the state space I; p is the transition probability matrix.

Optionally, the stationary distribution can be derived through computer programming, and obviously, the transition probability matrix P is a sparse matrix, so that the time complexity of programming and solving mu is not high, and the feasibility of solving is achieved. After the smooth distribution mu is deduced, the congestion window size of each sub-stream is calculated according to the following expression:

wherein I is the sequence number of the sub-stream, I is the sending state space, and j is the state sequence number in the state space I; max is the number of elements of the feature vector, i.e., (W-1)^k；Represents the congestion window size, μ, of the sub-stream I in the j-th state in the state space I_jRepresenting the probability of the stationary distribution corresponding to the jth state;is the average congestion window size.

Step 103, calculating to obtain the current round trip time value of each sub-flow according to the timestamp field in the data packet; the Timestamp field of the packet may be used to calculate the current RTT value:

RTT＝Timestamp_Value–Timestamp_Echo_Reply

according to RFC1323, the Timestamp option in the TCP header includes Timestamp _ Value and Timestamp _ Echo _ Reply, where Timestamp _ Value indicates a Timestamp corresponding to the current time when the packet is sent, and Timestamp _ Echo _ Reply is a Timestamp _ Value of a packet that is received last time by a receiving end and recorded in a received ACK packet

In some preferred embodiments, in consideration of the dynamic variation of RTT, a smoothing process needs to be performed on the current RTT value, where the smoothing process is expressed as:

RTT_t＝(1-α)RTT_new+αRTT_t-1

0<α<1

wherein RTT_newIs the round trip time value of the current moment; RTT (round trip time)_t-1Is the round trip time value of the previous moment; RTT (round trip time)_tFor smooth-processed round-tripAnd α is a factor that the smooth round trip time value is influenced by the current round trip time value, so that the RTT value is more representative, and the accuracy and the stability of the data scheduling method of the self-adaptive MPTCP are higher.

More preferably, α is 0.8.

104, predicting the average effective throughput rate on each sub-flow according to the average congestion window size and the round-trip time value of each sub-flow by the queuing theory Little theorem; optionally, the calculation expression of the average goodput rate over each sub-stream is:

wherein p is_iIs the packet loss rate on sub-stream i; MSS_iIs the maximum transmission segment size on substream i;is the average congestion window size over sub-stream i; RTT (round trip time)_iIs the round trip time value on sub-stream i;is the average goodput over substream i.

Step 105, dynamically adjusting the size of a congestion window of the minimum sub-flow according to the average effective throughput rate on each sub-flow, so that the minimum sub-flow is prevented from being congested; the average effective throughput rate of each sub-flow is the predicted throughput rate, and whether packet loss is possible to occur in the current sub-flow can be reflected laterally, so that congestion is avoided in advance.

Optionally, the step of dynamically adjusting the size of the congestion window of the minimum sub-stream according to the average goodput rate on each sub-stream includes: calculating to obtain a throughput rate coefficient according to the average effective throughput rate on each sub-flow; wherein the throughput rate coefficient is the ratio of the maximum sub-stream throughput rate to the minimum sub-stream throughput rate; and judging whether the throughput rate coefficient is larger than the throughput rate critical value, if so, correspondingly reducing the size of the congestion window of the minimum sub-flow. Specifically, for the differentiated network, first, a throughput coefficient θ is defined as a ratio of the maximum sub-stream throughput to the minimum sub-stream throughput, that is:

secondly, a throughput coefficient critical value theta is preset_T，θ_TIs a critical value for predicting the difference between the maximum substream throughput rate and the minimum substream throughput rate. When theta > theta_TThis means that the predicted throughput difference between the sub-streams is too large, and the sub-stream with the smallest throughput is likely to have a packet loss, and appropriate congestion avoidance processing is performed, so that the congestion window of the sub-stream with the smallest throughput should be appropriately reduced. Here, exceeding this critical value will appropriately reduce the current congestion window of the sub-flow with the minimum predicted throughput rate, so as to address the upcoming congestion or packet loss phenomenon, which is referred to as a pseudo-congestion avoidance phase. Optionally, the congestion window is halved in the congestion avoidance phase with reference to classical congestion control algorithms, since θ > θ_TThe sub-stream with the lowest throughput rate enters a pseudo-congestion avoidance phase, so that the sub-stream congestion window should be reducedWithin the interval, the reduced congestion window size is

Optionally, in some embodiments of the present application, a dynamic change of θ is considered, and θ is smoothed, where an expression of the smoothing is: theta_t＝(1-β)θ_new+βθ_t-1；0<β<1, wherein β is a balance factor, theta_newThe current time throughput rate coefficient; theta_t-1Counting an average coefficient for the throughput rate of the previous moment; theta_tIs the throughput coefficient after the smoothing process.

Optionally, the critical value θ_TDepending on the receive buffer size, i.e., the maximum number of packets that the receiving end can accommodate. The calculation expression for obtaining the throughput threshold is therefore:

(G_max-G_min)+1＜N_Buf＜(G_max-G_min)+2

the two formulas are combined to obtain the compound,

wherein N is_BufFor the size of the receiving end buffer, G_maxFor maximum sub-stream throughput, G_minIs the minimum substream throughput rate.

And step 106, adjusting the size of the slow start threshold according to the adjusted size of the congestion window, so that the adjusted sub-flow is still in the congestion avoidance stage.

If the sub-stream congestion window is reduced, comparing the reduced congestion windows cw'_iAnd a slow start threshold SSTHresh_iIf cw'_i＜SSThresh_iThen the slow start threshold is adjusted downward such that SSTHresh is set_i＝cw′_iThis ensures that the sub-streams remain in the congestion avoidance phase after the policy has been enforced.

It can be seen from the foregoing embodiments that, in the data scheduling method of the adaptive MPTCP provided in the present invention, a markov model is constructed based on the congestion window of each sub-stream, and then an average congestion window size of each sub-stream is obtained through calculation, and then an average effective throughput rate on each sub-stream is obtained through prediction based on the average congestion window size and the round trip time value, so that the congestion window size can be adjusted according to the predicted throughput rate, and finally data scheduling and congestion control are implemented. On the basis of a heterogeneous network multipath parallel transmission technology, the congestion windows of substreams corresponding to each interface of the wireless mobile equipment and the data scheduling behaviors which can be adopted are researched, the probability model based on the random event accumulation times is established to analyze the state transition probability, a predictable throughput rate feedback model is obtained accordingly, and the data is scheduled to the optimal interface by adopting the scheduling strategy with the optimal throughput rate, so that the advantages of multipath parallel transmission high transmission performance are exerted, and the reliability of data transmission is enhanced. Therefore, the data scheduling method of the self-adaptive MPTCP overcomes the defects of the existing MPTCP data scheduling and congestion control technology, so that the MPTCP can adaptively perform more reasonable scheduling aiming at the conditions of network fluctuation and network isomerization, and finally the efficiency and reliability of multi-path data transmission can be improved.

Referring to fig. 2, a flowchart of another embodiment of a data scheduling method of adaptive MPTCP according to the present invention is shown.

Firstly, establishing a Markov model;

assuming end-to-end communication between devices having two network interfaces and establishing an MPTCP connection comprising two sub-streams, the sending state at the last instant is expressed in the size of the congestion window of the two sub-streams, i.e. s_t-1＝{CW₁,CW₂}. When the sub-flow in the minimum transmission time interval has not lost the packet, the size of the congestion window is reduced by half, otherwise, the size of the congestion window is increased by 1, and s is the possible sending state s at the current time t obtained according to the sending state at the previous time_tThere are four possible situations:

{CW₁+1,CW₂+1}

according to the packet loss rates of sub-streams 1 and 2, the four state transition probabilities can be calculated and recorded asp₁₂。

The congestion window size is expressed quantitatively according to the maximum transmission message MSS, and the size range of the sub-flow congestion window is taken to be [2, W ]. The transmit state space is then:

I：{(CW₁,CW₂)|CW₁∈[2,W],CW₂∈[2,W]}.

as can be seen from the state space, a transmit state at a certain time may be (W-1)²In this case, the state transition probability matrix from the previous time to the current time is one (W-1)²*(W-1)²Of the matrix of (a). Recording the one-step transition probability matrix as P according to the state transition probabilityp₁₂A one-step transition probability matrix P can be calculated. Note that the ultimate plateau of this mahalanobis chain is:

then the conditions of the plateau from the limit:

μ＝μP

μ₁+μ₂+…+μ_(W-1) ²＝1

μ_i≥0,i∈{1,2,…(W-1)²}

the extreme stationary distribution characteristic vector of the Markov chain can be obtained

Secondly, solving the average congestion window size of each sub-flow;

the extreme stationarity vector represents the congestion window size at which the substreams are stable in each state. Deducing smooth distribution mu through computer programming, and solving the congestion window size of each sub-flow according to the following expression after deducing the smooth distribution mu:

wherein,represents the congestion window size, μ, of the sub-stream I in the j-th state in the state space I_jAnd (3) representing the steadily distributed characteristic vector element corresponding to the jth state, wherein max represents the number of the element of the characteristic vector, and the model can show that: max (W-1)²

The average congestion window size for subflow 1 can be derived from the above equation:

the same can be obtained.

Thirdly, calculating the average value of the statistical RTT of each sub-flow;

the Timestamp field of the packet may be used to calculate the current RTT value:

RTT＝Timestamp_Value–Timestamp_Echo_Reply

optionally, in consideration of a dynamic change of the RTT value, performing a smoothing process on the current RTT value:

RTT_t＝(1-α)RTT_new+αRTT_t-1

where 0< α <1, α is a factor reflecting that the smoothed RTT value is affected by the current RTT, and is determined by the network conditions.

Thirdly, predicting the effective throughput rate of each sub-flow;

and combining the average congestion window size and the smooth RTT value to obtain the effective throughput rate of each sub-flow:

finally, dynamically adjusting a congestion window and a slow start threshold;

the predicted goodput rates of the two sub-streams are denoted by size G_max、G_min. The current time throughput coefficient θ is obtained by the following two equations:

θ_t＝(1-β)θ_new+βθ_t-1

wherein theta is_t-1Is the statistical average coefficient at the previous time, and β is an influence factor determined according to the network condition.

Assuming that the predicted throughput difference of the two substreams reaches a certain threshold θ_TSo that the number of data packet accumulations approaches the buffer size, based on the transmit buffer size N_BufLimited by the relationship of the maximum substream throughput rate and the minimum substream throughput rate.

(G_max-G_min)+1＜N_Buf＜(G_max-G_min)+2

And is composed of

Obtaining a throughput rate coefficient critical value theta_T：

Comparison of θ_tAnd theta_TAnd according to the magnitude relation between the throughput rate coefficient and the critical value, corresponding operations are adopted:

when theta > theta_TReducing the congestion window of the least throughput sub-stream toComparing the reduced congestion windows cw'_iAnd a slow start threshold SSTHresh_iIf cw'_i＜SSThresh_iThen the slow start threshold is adjusted downward such that SSTHresh is set_i＝cw′_i；

When theta is less than or equal to theta_TAnd in time, the size of the congestion window is adjusted only according to the packet loss condition of each sub-stream.

Compared with the existing MPTCP data scheduling and congestion control technology, the invention has at least the following advantages:

1. the smooth RTT value is obtained based on the Timestamp field, and the condition of RTT jitter when the network fluctuates can be effectively handled.

2. And establishing a Markov decision process to dynamically predict the average congestion window size of each sub-flow, and accurately predicting the effective throughput rate of each sub-flow.

3. And the size of a congestion window of the slowest sub-flow is properly reduced according to the predicted throughput rate, and more scientific data scheduling and congestion control are performed aiming at different differential transmission networks with different sub-flow performances.

4. The influence of slow start and quick retransmission is considered, prediction deviation is reduced, and the slow start threshold value is scientifically adjusted aiming at the condition that the congestion window is reduced to be lower than the slow start threshold value.

Those of ordinary skill in the art will understand that: the discussion of any embodiment above is meant to be exemplary only, and is not intended to intimate that the scope of the disclosure, including the claims, is limited to these examples; within the idea of the invention, also features in the above embodiments or in different embodiments may be combined, steps may be implemented in any order, and there are many other variations of the different aspects of the invention as described above, which are not provided in detail for the sake of brevity.

In addition, well known power/ground connections to Integrated Circuit (IC) chips and other components may or may not be shown within the provided figures for simplicity of illustration and discussion, and so as not to obscure the invention. Furthermore, devices may be shown in block diagram form in order to avoid obscuring the invention, and also in view of the fact that specifics with respect to implementation of such block diagram devices are highly dependent upon the platform within which the present invention is to be implemented (i.e., specifics should be well within purview of one skilled in the art). Where specific details (e.g., circuits) are set forth in order to describe example embodiments of the invention, it should be apparent to one skilled in the art that the invention can be practiced without, or with variation of, these specific details. Accordingly, the description is to be regarded as illustrative instead of restrictive.

While the present invention has been described in conjunction with specific embodiments thereof, many alternatives, modifications, and variations of these embodiments will be apparent to those of ordinary skill in the art in light of the foregoing description. For example, other memory architectures (e.g., dynamic ram (dram)) may use the discussed embodiments.

The embodiments of the invention are intended to embrace all such alternatives, modifications and variances that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, substitutions, improvements and the like that may be made without departing from the spirit and principles of the invention are intended to be included within the scope of the invention.

Claims (10)

1. A data scheduling method of adaptive MPTCP is characterized by comprising the following steps:

establishing a Markov model according to the congestion window of each sub-flow;

calculating to obtain the average congestion window size of each sub-flow according to a Markov model;

calculating to obtain the current round trip time value of each sub-flow according to the timestamp field in the data packet;

predicting to obtain the average effective throughput rate on each sub-flow according to the average congestion window size and the round trip time value of each sub-flow;

dynamically adjusting the size of a congestion window of the minimum sub-flow according to the average effective throughput rate on each sub-flow, so that the minimum sub-flow is prevented from being congested;

and adjusting the size of the slow start threshold according to the adjusted size of the congestion window, so that the adjusted sub-flow is still in a congestion avoidance stage.

2. The method of claim 1, wherein the markov model comprises: the sending state set of the equipment at each moment and the one-step transition probability set of the sending state conversion; wherein, the sending state set comprises the size of the congestion window of each sub-flow; the transition probability set is related to the packet loss rate of each sub-stream.

3. The method of claim 1, wherein the step of calculating the average congestion window size for each substream based on a markov model comprises:

obtaining a transition probability matrix of the sending state of the equipment according to the packet loss rate of each sub-flow;

calculating to obtain ultimate stable distribution according to the probability transition rule of the transition probability matrix;

and calculating the average congestion window size of each sub-flow according to the limit stable distribution.

4. The method according to claim 3, characterized in that the extreme stationary distribution is denoted as eigenvector μ; and the feature vector meets the following constraints:

μ＝(μ₁,μ₂…μ_v)

μP＝μ

μ₁+μ₂+…+μ_k-1+μ_v＝1

μ_j≥0,j∈{1,2…v}

wherein, mu_jThe probability of a stationary distribution on the jth state in the state space I; p is a transition probability matrix, and the transition probability matrix is a sparse matrix in (v x v) dimension;

the calculation expression of the average congestion window size is as follows:

$\stackrel{\&OverBar;}{{\mathrm{cw}}_i}=\underset{j=1}{\overset{max}{\&Sigma;}}{x}_{{\mathrm{iI}}_j}{\mathrm{\&mu;}}_j$

wherein I is the sequence number of the sub-stream, I is the sending state space, and j is the state sequence number in the state space I; max is the number of elements of the feature vector;the congestion window size of the sub-stream I in the jth state in the state space I; mu.s_jThe probability of the stationary distribution corresponding to the jth state;is the average congestion window size.

5. The method according to claim 1, wherein the step of calculating the current round trip time value of each sub-flow according to the timestamp field in the data packet further comprises:

smoothing the round trip time value;

the expression of the smoothing process is:

RTT_t＝(1-α)RTT_new+αRTT_t-1

0<α<1

wherein RTT_newIs the round trip time value of the current moment; RTT (round trip time)_t-1Is the round trip time value of the previous moment; RTT (round trip time)_tWhich is the smoothed round trip time value, α is the factor by which the smoothed round trip time value is affected by the current round trip time value.

6. The method of claim 5, wherein the impact factor α is 0.8

7. The method of claim 1, wherein the average goodput rate over each substream is calculated by the expression:

$\stackrel{\&OverBar;}{{\mathrm{Goodput}}_i}=\frac{(1-p_i){\mathrm{MSS}}_i\stackrel{\&OverBar;}{{\mathrm{cw}}_i}}{{\mathrm{RTT}}_i}$

wherein p is_iIs the packet loss rate on sub-stream i; MSS_iIs the maximum transmission segment size on substream i;is the average congestion window size over sub-stream i; RTT (round trip time)_iIs the round trip time value on sub-stream i;is the average goodput over substream i.

8. The method of claim 1, wherein the step of dynamically adjusting the congestion window size of the smallest sub-stream according to the average goodput rate over each sub-stream comprises:

calculating to obtain a throughput rate coefficient according to the average effective throughput rate on each sub-flow; wherein the throughput rate coefficient is the ratio of the maximum sub-stream throughput rate to the minimum sub-stream throughput rate;

and judging whether the throughput rate coefficient is larger than the throughput rate critical value, if so, correspondingly reducing the size of the congestion window of the minimum sub-flow.

9. The method of claim 8, wherein the step of calculating a throughput coefficient based on the average goodput over each substream is further followed by: and smoothing the throughput rate coefficient, wherein the expression of the smoothing is as follows:

θ_t＝(1-β)θ_new+βθ_t-1

wherein β is a balance factor, theta_newThe current time throughput rate coefficient; theta_t-1Counting an average coefficient for the throughput rate of the previous moment; theta_tIs the throughput coefficient after the smoothing process.

10. The method of claim 8 wherein the throughput threshold is calculated as:

(G_max-G_min)+1＜N_Buf＜(G_max-G_min)+2

${\mathrm{\&theta;}}_T=\frac{G_{max}}{G_{min}}$

so as to obtain the compound with the characteristics of,

$\frac{(N_{Buf}-2)}{G_{\min }}+1<{\mathrm{\&theta;}}_T<\frac{(N_{Buf}-1)}{G_{\min }}+1$

wherein N is_BufFor the size of the receiving end buffer, G_maxFor maximum sub-stream throughput, G_minIs the minimum substream throughput rate.