CN115134307B - Load balancing method based on packet loss rate coding in cloud computing - Google Patents
Load balancing method based on packet loss rate coding in cloud computing Download PDFInfo
- Publication number
- CN115134307B CN115134307B CN202210740895.7A CN202210740895A CN115134307B CN 115134307 B CN115134307 B CN 115134307B CN 202210740895 A CN202210740895 A CN 202210740895A CN 115134307 B CN115134307 B CN 115134307B
- Authority
- CN
- China
- Prior art keywords
- packets
- packet
- packet loss
- loss rate
- coded
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 38
- 239000011159 matrix material Substances 0.000 claims description 27
- 230000005540 biological transmission Effects 0.000 claims description 8
- 230000008569 process Effects 0.000 claims description 8
- 238000004364 calculation method Methods 0.000 claims description 2
- 230000007246 mechanism Effects 0.000 abstract description 14
- 238000012937 correction Methods 0.000 abstract description 5
- 238000012545 processing Methods 0.000 abstract description 4
- 238000007418 data mining Methods 0.000 description 9
- 238000010586 diagram Methods 0.000 description 4
- 238000002474 experimental method Methods 0.000 description 4
- 238000004088 simulation Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000011056 performance test Methods 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- 241000976924 Inca Species 0.000 description 1
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 230000008092 positive effect Effects 0.000 description 1
- 239000007921 spray Substances 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L47/00—Traffic control in data switching networks
- H04L47/10—Flow control; Congestion control
- H04L47/28—Flow control; Congestion control in relation to timing considerations
- H04L47/283—Flow control; Congestion control in relation to timing considerations in response to processing delays, e.g. caused by jitter or round trip time [RTT]
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L1/00—Arrangements for detecting or preventing errors in the information received
- H04L1/004—Arrangements for detecting or preventing errors in the information received by using forward error control
- H04L1/0056—Systems characterized by the type of code used
- H04L1/0057—Block codes
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L47/00—Traffic control in data switching networks
- H04L47/10—Flow control; Congestion control
- H04L47/12—Avoiding congestion; Recovering from congestion
- H04L47/125—Avoiding congestion; Recovering from congestion by balancing the load, e.g. traffic engineering
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L47/00—Traffic control in data switching networks
- H04L47/10—Flow control; Congestion control
- H04L47/25—Flow control; Congestion control with rate being modified by the source upon detecting a change of network conditions
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/50—Reducing energy consumption in communication networks in wire-line communication networks, e.g. low power modes or reduced link rate
Abstract
The invention discloses a load balancing method based on packet loss rate coding in cloud computing, and relates to the technical field of data processing. The switch randomly scatters the encoded packet to all parallel paths. The receiving end receives a certain number of coded packets and decodes the coded packets to obtain source packets. The invention carries out forward error correction coding on the source packet by detecting the packet loss probability, randomly scatters the coded packet to all paths, and can effectively reduce the probability of TCP retransmission timeout caused by packet loss in a random packet scattering mechanism while balancing the load.
Description
Technical Field
The invention relates to the technical field of data processing, in particular to a load balancing method based on packet loss rate coding in cloud computing.
Background
Modern cloud computing applications have stringent requirements for low latency and high throughput, such as data mining, big data analysis, web searching, and deep learning. To meet the demands of low latency and high throughput of applications, edge cloud computing extends the capabilities of cloud computing to network edges, i.e., to provide services closer to users by placing storage and computing resources in distributed edge data centers that are connected to large data centers through high-speed networks while being connected to customers through various low-speed networks. Edge data center networks typically deploy massive equivalent parallel paths between a sending end and a receiving end to provide high bisected bandwidth, supporting efficient transmission. Data centers widely deploy network topologies such as fat tree structures and leaf-spine structures.
In recent years, data centers have proposed a packet-based load balancing scheme, known as random packet scattering mechanism (RPS). The RPS randomly scatters packets in all parallel paths between the sender and receiver to fully utilize bandwidth resources. In this way, packets are transmitted to multiple paths simultaneously and the flow can be completed faster. However, RPS may exacerbate packet loss and even retransmission timeout in many-to-one bursty traffic scenarios, resulting in poor performance. Burst traffic is very common in edge data centers. For example, in distributed deep learning training, a large number of working nodes simultaneously transmit training gradients to an aggregation server to update global model parameters at the end of each iteration. In such sudden congestion situations, packet loss is very likely to occur. In the worst case, the loss of all packets of the congestion window of the TCP protocol or one of the last three tail packets of the flow may result in a Retransmission Timeout (RTO) of the TCP. The RTO defaults to 200 milliseconds, while the Round Trip Time (RTT) of a data center is only tens or hundreds of microseconds, meaning that a new packet cannot be sent for a significant amount of time before a lost packet is successfully retransmitted. That is, a timeout will result in a dramatic drop in throughput. Therefore, how to reduce packet loss retransmission and even timeout, thereby reducing the flow completion time and improving the application performance is a problem to be solved urgently.
Disclosure of Invention
In order to reduce the probability of TCP retransmission timeout caused by packet loss of a random packet scattering mechanism in a burst traffic scene in a data center network environment, the invention provides a load balancing method based on packet loss rate coding in cloud computing.
In order to solve the technical problems, the invention adopts the following technical methods: a load balancing method based on packet loss rate coding in cloud computing comprises the following steps:
1. the transmitting end operates according to the following steps:
step S11, initializing basic round trip delay RTT and packet loss rate update period threshold T th A packet loss rate P, a value of a start time t of a packet loss rate update period; let the number of encoded redundant packets be r, and the number of endogenous packets in the TCP congestion window CWND be w;
step S12, judging whether the difference between the current time and the start time T of the packet loss rate update period is greater than or equal to the packet loss rate update period T th If so, the packet loss rate P is updated to a value of the packet loss rate update period T th The ratio of the number of the acknowledgement packets which are not received in the packet transmission device to the total number of the transmitted packets, and setting the starting time t of the packet loss rate update period as the current time; otherwise, turning to step S13;
step S13, calculating the number r of coded redundant packets according to the packet loss rate P, and carrying out random linear combination coding on w source packets in a TCP congestion window CWND to generate w+r coded packets;
step S14, the coded packet obtained in the step S13 is sent to the switch, whether the packet is sent is judged, if yes, the process is finished, and if not, the process goes to the step S12;
2. the switch operates as follows:
step S21, receiving a coded packet;
step S22, randomly selecting a forwarding outlet port to forward the coded packet, and turning to step S21;
3. the receiving end operates according to the following steps:
step S31, receiving a coded packet;
step S32, judging whether the number of the received coded packets is equal to the number of the endogenous packets in a TCP congestion window CWND, if so, decoding the source packets and submitting the source packets to an upper TCP layer and an application layer, and turning to step S33; otherwise, turning to step S31;
step S33, an acknowledgement packet ACK is sent to the transmitting end, and the process goes to step S31.
Further, in step S11, at the time of initialization, the base round trip delay RTT is set to 50 μs; packet loss rate update period threshold T th Setting to 2RTT; the packet loss rate P, the start time t of the packet loss rate update period is set to 0.
Still further, in step S13, the calculation formula of the number r of encoded redundant packets is as follows:
r≥w/(1-P)-w (1)
to reduce the number of redundant packets transmitted, the minimum integer of w/(1-P) -w in equation (1) is taken upward as the value of the number of coded redundant packets r.
Further, in step S13, w source packets within the TCP congestion window CWND are encoded into w+r encoded packets by using a random linear combination encoding method, where the encoding formula is:
a＝mH (2)
wherein, alpha is a code packet, m is a source packet, H is a generator matrix, and the generator matrix is expressed as:
wherein [ a ] 1 a 2 ... a w+r ]Is a matrix formed by w+r coded packets obtained by coding, [ m ] 1 m 2 ... m w ]Is a matrix of w source packets, h= (H i,j ) w×(w+r) A generator matrix of w× (w+r), h i,j Generating elements of the ith row and jth column in the matrix, i=1, 2, …, w; j=1, 2, …, w+r; define generator matrix h= (H i,j ) w×(w+r) When j is less than or equal to w, if i is less than or equal toj is h i,j =1, if i > j, h i,j =0; when j > w, h i,j =i-1+j-w, expressed as:
preferably, in step S32, after the receiving end receives any w encoded packets, the receiving end decodes the w source packets by using the following decoding formula:
wherein [ a ] 1 a 2 ... a w ]Is a matrix formed by arbitrary w coded packets received by a receiving end, h i,j The i element of the j-th column of the matrix H is generated by the transmitting end, and i is less than or equal to w; j=1, 2, w; j is less than or equal to w+r; when j is less than or equal to w, if i is less than or equal to j, h i,j =1, if i > j, h i,j =0; otherwise, when j > w, h i,j ＝i-1+j-w。
The load balancing method based on the packet loss rate coding in the cloud computing effectively reduces the probability of TCP retransmission overtime caused by packet loss of a random packet scattering mechanism in a burst flow scene in a data center network environment by detecting the packet loss probability, carrying out forward error correction coding on the source packet and randomly scattering the coded packet to all paths. Specifically, in the method, a transmitting end encodes a source packet using a forward error correction coding (FEC) technique, and redundancy of the encoded packet is dynamically adjusted according to a packet loss rate detected by the transmitting end. In order to adapt to different degrees of congestion, the CRPS increases redundant coded packets under the condition of high packet loss rate so as to reduce retransmission timeout probability, otherwise, the coded packets are reduced so as to reduce redundant traffic. The switch then randomly scatters the encoded packets in all available parallel paths to increase link utilization. Therefore, even if some coded packets are blocked or even lost, as long as a sufficient number of coded packets from any parallel path arrive at the receiving end, the source packet can be successfully decoded to avoid packet retransmission and timeout, thereby achieving the purpose of reducing the probability of retransmission timeout caused by packet loss in a random packet scattering mechanism.
Drawings
Fig. 1 is a flowchart of a load balancing method based on packet loss rate coding in cloud computing according to the present invention;
FIG. 2 is a topology diagram of a test scenario of an NS-2 network simulation platform in an embodiment of the invention;
FIG. 3 is a schematic diagram of performance test results under the workload of web searching and data mining in an embodiment of the present invention;
fig. 4 is a graph showing probability of occurrence of Incast in a scene of different concurrent streams and different cache sizes for RPS and CRPS in an embodiment of the present invention.
Detailed Description
The invention will be further described with reference to examples and drawings, to which reference is made, but which are not intended to limit the scope of the invention.
Before describing the present invention, the design concept of the present invention will be described in detail. In order to reduce the probability of retransmission timeout caused by packet loss in a burst traffic scene, the invention designs a load balancing method based on packet loss rate coding in cloud computing by adopting a random packet scattering mechanism based on coding, which is called CRPS, wherein the method mainly depends on coding to effectively reduce tail delay. Further, the transmitting end encodes the source packet using a forward error correction coding (FEC) technique, and dynamically adjusts redundancy of the encoded packet according to the packet loss rate detected by the transmitting end. To accommodate varying degrees of congestion, CRPS increases redundancy coded packets at high packet loss rates to reduce retransmission timeout probabilities, otherwise coded packets may be reduced to reduce redundancy traffic. The switch then randomly scatters the encoded packets in all available parallel paths to increase link utilization. Thus, even if some of the encoded packets are blocked or even lost, the source packet can be successfully decoded to avoid packet retransmissions and timeouts as long as a sufficient number of encoded packets from any parallel path arrive at the receiving end. Although redundant coded packets consume link bandwidth, the benefit of reducing the timeout probability is far greater than the redundant traffic overhead from coding. CRPS is implemented between the transport layer and the network layer of the end hosts (including the sender and receiver), without modifying the Transmission Control Protocol (TCP) or Internet Protocol (IP) stack. The present invention will be specifically described below.
Referring to fig. 1, a load balancing method based on packet loss rate coding in cloud computing mainly includes three processing procedures of a transmitting end, a switch and a receiving end, wherein the processing procedures are as follows:
first, the current operation subject type is judged.
If the terminal is a transmitting terminal, the method comprises the following steps:
step S11, initializing the setting of the basic round trip delay RTT to be 50 mu S; packet loss rate update period threshold T th Setting to 2RTT; the packet loss rate P and the start time t of the packet loss rate update period are both set to 0; let the number of encoded redundant packets be r, and the number of endogenous packets in the TCP congestion window CWND be w.
Step S12, judging whether the difference between the current time and the start time T of the packet loss rate update period is greater than or equal to the packet loss rate update period T th If so, the packet loss rate P is updated to a value of the packet loss rate update period T th The ratio of the number of the acknowledgement packets which are not received in the packet transmission device to the total number of the transmitted packets, and setting the starting time t of the packet loss rate update period as the current time; otherwise go to step S13.
And S13, calculating the number r of coded redundant packets according to the packet loss rate P, and carrying out random linear combination coding on w source packets in the TCP congestion window CWND to generate w+r coded packets. In order to ensure that the receiving end can receive the number of encoded packets required for successful decoding, the following formula needs to be satisfied:
(1-P)(w+r)≥w；
namely: r is greater than or equal to w/(1-P) -w (1)
To reduce the number of redundant packets transmitted, the minimum integer of w/(1-P) -w in equation (1) is taken upward as the value of the number of coded redundant packets r.
When w source packets in the TCP congestion window CWND are encoded into w+r encoded packets by adopting a random linear combination encoding method, the encoding formula is as follows:
a＝mH (2)
wherein, alpha is a code packet, m is a source packet, H is a generator matrix, and the generator matrix is expressed as:
wherein [ a ] 1 a 2 ... a w+r ]Is a matrix formed by w+r coded packets obtained by coding, [ m ] 1 m 2 ... m w ]Is a matrix of w source packets, h= (H i,j ) w×(w+r) A generator matrix of w× (w+r), h i,j Generating elements of the ith row and jth column in the matrix, i=1, 2, …, w; j=1, 2, …, w+r; define generator matrix h= (H i,j ) w×(w+r) When j is less than or equal to w, if i is less than or equal to j, h i,j =1, if i > j, h i,j =0; when j > w, h i,j =i-1+j-w, expressed as:
it should be noted that, the present invention dynamically adjusts the number r of encoded redundant packets according to the packet loss rate P detected by the transmitting end, in order to adapt to congestion with different degrees, the CRPS increases the number r of encoded redundant packets under the high packet loss rate to reduce the retransmission timeout probability, otherwise, the number r of encoded redundant packets is reduced to reduce the redundant traffic.
Step S14, the coded packet obtained in the step S13 is sent to the switch, whether the packet is sent is judged, if yes, the process is ended, and if not, the process goes to the step S12.
If the switch is a switch, a forwarding outlet port is randomly selected to forward the coded packet after the coded packet is received, and then the coded packet is continuously received. The switch randomly scatters the encoded packets in all available parallel paths to improve link utilization. Thus, even if some of the encoded packets are blocked or even lost, the source packet can be successfully decoded to avoid packet retransmissions and timeouts as long as a sufficient number of encoded packets from any parallel path arrive at the receiving end.
If the received code packet is the receiving end, judging whether the quantity of the received code packet is equal to a TCP congestion window CWND, if not, continuing to receive the code packet, if so, decoding the source packet, submitting the source packet to an upper TCP layer and an application layer, sending an acknowledgement packet ACK to the sending end, and then continuing to receive the code packet. Here, after receiving any w encoded packets, the receiving end decodes the w source packets by using the following decoding formula:
wherein [ a ] 1 a 2 ... a w ]Is a matrix formed by arbitrary w coded packets received by a receiving end, h i,j The i element of the j-th column of the matrix H is generated by the transmitting end, and i is less than or equal to w; j=1, 2, …, w; j is less than or equal to w+r; when j is less than or equal to w, if i is less than or equal to j, h i,j =1, if i > j, h i,j =0; otherwise, when j > w, h i,j ＝i-1+j-w。
The invention can effectively reduce the probability of retransmission overtime caused by packet loss in a burst flow scene through the random packet scattering mechanism based on the coding, and particularly, the probability of Incast occurrence after the load balancing method is adopted in a data center network environment is P '' Incast It is calculated as follows.
1) Firstly, the packet loss probability P is calculated by adopting the following formula l ′；
Wherein C is the link bandwidth, RTT min For minimum round trip delay, MSS is the size of one TCP congestion window CWND packet, B is the buffer size,n is the number of flows of the TCP congestion window CWND.
2) And then the probability P 'of overtime caused by the loss of the last three packets at the tail part of one stream is calculated by adopting the following formula' tto ；
Wherein R is the number of rounds of data packet transmission of one stream, WR refers to the size of a TCP congestion window CWND transmitted by the R-th round, rR refers to the number of coded redundant packets corresponding to the data packets of the WR window, and W i Refers to the size, r, of a TCP congestion window CWND of the ith round of transmission i Refers to W i The number of coded redundant packets corresponding to the windowed packets.
3) Then, the overtime probability P 'caused by the data packet full window loss of the TCP congestion window CWND in one stream transmission process is calculated by adopting the following formula' wto ；
4) Then the timeout probability P 'of a stream is calculated by the following formula' to ；
P′ to as P′ to ＝P′ tto +P′ wto (6)
5) Finally, the probability P 'of the occurrence of Incast is calculated by adopting the following formula' Incast ；
P′ Incast ＝1-(1-P′ to ) n (7)。
To verify the effectiveness of the present invention, performance testing of the present invention is next performed using an NS-2 network simulation platform.
Fig. 2 is a topology diagram of a test scenario of an NS-2 network simulation platform, where the experiment adopts a leaf-spine network topology structure, and the whole network has 8 leaf switches, 8 spine switches and 40 end hosts (1 end host serves as 1 transmitting end and 1 receiving end at the same time), and each leaf switch communicates with 40 end hosts respectively. All links have a bandwidth of 40Gbps and each link has a network propagation delay of 5 microseconds. The switch caches 256 packets.
TCP was used in the experiments as the default transport protocol. Experiments generated two typical data center workloads, web page search and data mining. In a web search scenario, about 62% of the streams are less than 100KB, with an average stream size of 1.6MB. In a data mining scenario, about 83% less than 100KB of the streams provide less than 5% of the traffic. All streams under both actual workloads are generated between randomly selected pairs of servers, the arrival times of the streams obeying poisson distribution. The average network load was increased from 0.3 to 0.7 during the experiment.
Fig. 3 is a schematic diagram of performance test results under the workload of web page search and data mining, in which fig. 3 (a) is a web page search average stream completion time, fig. 3 (b) is a web page search 95 minute bit stream completion time, fig. 3 (c) is a web page search 99 minute bit stream completion time, fig. 3 (d) is a data mining average stream completion time, fig. 3 (e) is a data mining 95 minute bit stream completion time, and fig. 3 (f) is a data mining 99 minute bit stream completion time. In fig. 3, three conventional load balancing mechanisms are involved in RPS, DRILL, TLB, and a load balancing method CRPS based on packet loss rate coding provided by the present invention, where RPS represents (RandomPacket Spray) a random packet scattering load balancing mechanism, DRILL represents (DistributedRandomized In-network LocalizedLoad-balancing) a distributed random intra-network switch local load balancing mechanism, and TLB represents (Traffic-awareloadbalancing) a Traffic-aware load balancing mechanism. This is because CRPS effectively encodes and transmits a source packet, and as long as a receiving end receives a certain number of encoded packets from a non-congestion path, the receiving end can successfully decode the original packet, thereby avoiding retransmission and even timeout caused by packet loss. In addition, CRPS can detect changes in packet loss rate and adjust the redundant coded packets accordingly to reduce tail latency. While the flow size in the data mining workload is typically a heavy-tail distribution, CRPS also ensures minimal tail latency compared to the other three schemes. The reason is that even if the CRPS scatters the data packets to all parallel paths like other mechanisms, there will always be some encoded packets going through the non-congested path to the receiving end, thus successfully decoding the source packet. CRPS significantly improves latency performance in heavy load scenarios.
Fig. 4 is a probability of an RPS and CRPS to occur Incast in a scene of different concurrent streams and different cache sizes. As shown in fig. 4, by changing the number of streams and the buffer size, the probability of Incast occurrence increases with an increase in the number of streams. Wherein the probability of Incast occurrence under CRPS mechanism increases much lower than RPS. The reason is that when the timeout is mainly caused by the loss of the tail packet of a small number of streams, the positive effect of the encoded packet is dominant, because the encoded packet can recover the lost packet in time to avoid the timeout. However, as traffic continues to increase, the CRPS will increase in probability of Incast occurrence by a higher amount than RPS because the traffic overhead of redundantly encoded packets reduces the packet encoding benefit. Finally, when the flow exceeds 50, the incremental probabilities of both CRPS and RPS reach 1. In the case of a buffer size of 100 packets, incast occurs earlier in RPS and CRPS due to the smaller buffer space. In summary, the smaller the size of the buffer under RPS and CRPS, the earlier Incast occurs under the same streaming numbers. However, CRPS achieves a lower probability of Incas than RPS.
Compared with the traditional method, the load balancing method based on the packet loss rate coding in the cloud computing can effectively reduce the probability of TCP retransmission timeout caused by packet loss of a random packet scattering mechanism in a burst flow scene in a data center network environment by detecting the packet loss probability, performing forward error correction coding on the source packet and randomly scattering the coded packet to all paths.
The foregoing embodiments are preferred embodiments of the present invention, and in addition, the present invention may be implemented in other ways, and any obvious substitution is within the scope of the present invention without departing from the concept of the present invention.
In order to facilitate understanding of the improvements of the present invention over the prior art, some of the figures and descriptions of the present invention have been simplified, and some other elements have been omitted from this document for clarity, as will be appreciated by those of ordinary skill in the art.
Claims (5)
1. The load balancing method based on the packet loss rate coding in the cloud computing is characterized by comprising the following steps of:
1. the transmitting end operates according to the following steps:
step S11, initializing basic round trip delay RTT and packet loss rate update period threshold T th A packet loss rate P, a value of a start time t of a packet loss rate update period; let the number of encoded redundant packets be r, and the number of endogenous packets in the TCP congestion window CWND be w;
step S12, judging whether the difference between the current time and the start time T of the packet loss rate update period is greater than or equal to the packet loss rate update period T th If so, the packet loss rate P is updated to a value of the packet loss rate update period T th The ratio of the number of the acknowledgement packets which are not received in the packet transmission device to the total number of the transmitted packets, and setting the starting time t of the packet loss rate update period as the current time; otherwise, turning to step S13;
step S13, calculating the number r of coded redundant packets according to the packet loss rate P, and carrying out random linear combination coding on w source packets in a TCP congestion window CWND to generate w+r coded packets;
step S14, the coded packet obtained in the step S13 is sent to the switch, whether the packet is sent is judged, if yes, the process is finished, and if not, the process goes to the step S12;
2. the switch operates as follows:
step S21, receiving a coded packet;
step S22, randomly selecting a forwarding outlet port to forward the coded packet, and turning to step S21;
3. the receiving end operates according to the following steps:
step S31, receiving a coded packet;
step S32, judging whether the number of the received coded packets is equal to the number of the endogenous packets in a TCP congestion window CWND, if so, decoding the source packets and submitting the source packets to an upper TCP layer and an application layer, and turning to step S33; otherwise, turning to step S31;
step S33, an acknowledgement packet ACK is sent to the transmitting end, and the process goes to step S31.
2. The load balancing method based on packet loss rate coding in cloud computing according to claim 1, wherein: in step S11, at the time of initialization, the base round trip delay RTT is set to 50 μs; packet loss rate update period threshold T th Setting to 2RTT; the packet loss rate P, the start time t of the packet loss rate update period is set to 0.
3. The load balancing method based on packet loss rate coding in cloud computing according to claim 2, wherein: in step S13, the calculation formula of the number r of encoded redundant packets is as follows:
r≥w/(1-P)-w (1)
to reduce the number of redundant packets transmitted, the minimum integer of w/(1-P) -w in equation (1) is taken upward as the value of the number of coded redundant packets r.
4. The load balancing method based on packet loss rate coding in cloud computing as claimed in claim 3, wherein: in step S13, w source packets in the TCP congestion window CWND are encoded into w+r encoded packets by using a random linear combination encoding method, where the encoding formula is:
a＝mH (2)
wherein, alpha is a code packet, m is a source packet, H is a generator matrix, and the generator matrix is expressed as:
wherein [ a ] 1 a 2 ...a w+r ]Is a matrix formed by w+r coded packets obtained by coding, [ m ] 1 m 2 ...m w ]Is a matrix of w source packets, h= (H i,j ) w×(w+r) A generator matrix of w× (w+r), h i,j Generating elements of the ith row and jth column in the matrix, i=1, 2, …, w; j=1, 2, …, w+r; define generator matrix h= (H i,j ) w×(w+r) When j is less than or equal to w, if i is less than or equal to j, h i,j =1, if i > j, h i,j =0; when j > w, h i,j =i-1+j-w, expressed as:
5. the load balancing method based on packet loss rate coding in cloud computing as claimed in claim 4, wherein: in step S32, after receiving any w encoded packets, the receiving end decodes the w source packets by using the following decoding formula:
wherein [ a ] 1 a 2 ...a w ]Is a matrix formed by arbitrary w coded packets received by a receiving end, h i,j The i element of the j-th column of the matrix H is generated by the transmitting end, and i is less than or equal to w; j=1, 2, …, w; j is less than or equal to w+r; when j is less than or equal to w, if i is less than or equal to j, h i,j =1, if i > j, h i,j =0; otherwise, when j > w, h i,j ＝i-1+j-w。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210740895.7A CN115134307B (en) | 2022-06-27 | 2022-06-27 | Load balancing method based on packet loss rate coding in cloud computing |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210740895.7A CN115134307B (en) | 2022-06-27 | 2022-06-27 | Load balancing method based on packet loss rate coding in cloud computing |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115134307A CN115134307A (en) | 2022-09-30 |
| CN115134307B true CN115134307B (en) | 2024-01-26 |
Family
ID=83380375
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210740895.7A Active CN115134307B (en) | 2022-06-27 | 2022-06-27 | Load balancing method based on packet loss rate coding in cloud computing |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115134307B (en) |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101686106A (en) * | 2008-09-28 | 2010-03-31 | 华为技术有限公司 | Self-adaptive forward error correction (FEC) method, device and system |
| CN107508655A (en) * | 2017-07-19 | 2017-12-22 | 西南交通大学 | A kind of adaptive end to end network coding and transmission method |
| US9992088B1 (en) * | 2014-11-07 | 2018-06-05 | Speedy Packets, Inc. | Packet coding based network communication |
| CN109992419A (en) * | 2019-03-29 | 2019-07-09 | 长沙理工大学 | A kind of collaboration edge calculations low latency task distribution discharging method of optimization |
| CN110198273A (en) * | 2019-05-31 | 2019-09-03 | 中南大学 | Multi-path transmission method in data center network based on network code |
| US10491329B1 (en) * | 2016-12-08 | 2019-11-26 | Amazon Technologies, Inc. | Transfer of data-redundancy encoded data via unreliable, connectionless protocol |
| CN113438055A (en) * | 2021-06-24 | 2021-09-24 | 西安电子科技大学 | Convolutional network coding transmission method based on unequal redundancy insertion |
| CN113612580A (en) * | 2021-08-03 | 2021-11-05 | 四川大学 | Screen updating transmission method based on fountain code coding strategy and redundancy self-adaption |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10320526B1 (en) * | 2014-11-07 | 2019-06-11 | Strong Force Iot Portfolio 2016, Llc | Packet coding based network communication |
| US10999012B2 (en) * | 2014-11-07 | 2021-05-04 | Strong Force Iot Portfolio 2016, Llc | Packet coding based network communication |
| EP3281413B1 (en) * | 2015-04-09 | 2021-01-20 | Dejero Labs Inc. | Systems, devices and methods for distributing data with multi-tiered encoding |
| US11507064B2 (en) * | 2016-05-09 | 2022-11-22 | Strong Force Iot Portfolio 2016, Llc | Methods and systems for industrial internet of things data collection in downstream oil and gas environment |
| US10630315B2 (en) * | 2017-09-29 | 2020-04-21 | Intel Corporation | Technologies for applying a redundancy encoding scheme to segmented network packets |
-
2022
- 2022-06-27 CN CN202210740895.7A patent/CN115134307B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101686106A (en) * | 2008-09-28 | 2010-03-31 | 华为技术有限公司 | Self-adaptive forward error correction (FEC) method, device and system |
| US9992088B1 (en) * | 2014-11-07 | 2018-06-05 | Speedy Packets, Inc. | Packet coding based network communication |
| US10491329B1 (en) * | 2016-12-08 | 2019-11-26 | Amazon Technologies, Inc. | Transfer of data-redundancy encoded data via unreliable, connectionless protocol |
| CN107508655A (en) * | 2017-07-19 | 2017-12-22 | 西南交通大学 | A kind of adaptive end to end network coding and transmission method |
| CN109992419A (en) * | 2019-03-29 | 2019-07-09 | 长沙理工大学 | A kind of collaboration edge calculations low latency task distribution discharging method of optimization |
| CN110198273A (en) * | 2019-05-31 | 2019-09-03 | 中南大学 | Multi-path transmission method in data center network based on network code |
| CN113438055A (en) * | 2021-06-24 | 2021-09-24 | 西安电子科技大学 | Convolutional network coding transmission method based on unequal redundancy insertion |
| CN113612580A (en) * | 2021-08-03 | 2021-11-05 | 四川大学 | Screen updating transmission method based on fountain code coding strategy and redundancy self-adaption |
Non-Patent Citations (5)
| Title |
|---|
| A new reliability -based incremental redundancy hybrid ARQ scheme using LDPC codes;Saber H et al;《Information Theory》;全文 * |
| Network coding meets TCP: theory and implementation;SUNDARARAJAN J K;《Proceedings of the IEEE》;全文 * |
| 基于网络编码的数据通信技术研究;董赞强;《中国博士学位论文全文数据库》;全文 * |
| 无线多播中基于网络编码的高效重传方法研究;邵鹏飞;《中国优秀硕士学位论文全文数据库》;全文 * |
| 适合TCP的网络编码重传机制;陈静等;《通信学报》;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115134307A (en) | 2022-09-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11018701B2 (en) | Reliable data transmission method based on reliable UDP and fountain code in aeronautical ad hoc networks | |
| Amir et al. | Efficient TCP over networks with wireless links | |
| US8462624B2 (en) | Congestion management over lossy network connections | |
| KR100785293B1 (en) | System and Method for TCP Congestion Control Using Multiple TCP ACKs | |
| US8873385B2 (en) | Incast congestion control in a network | |
| US9007905B2 (en) | System to improve an Ethernet network | |
| KR101610715B1 (en) | One-way data transmission and reception system, and one-way data transmission and reception method | |
| CN108282402B (en) | Packet scattering method based on coding in data center network | |
| WO2016045332A1 (en) | Method and device for encoding parameter adjustment and feedback information processing | |
| JP2007028623A (en) | System and method for adjusting ber/per for accelerating transmission speed of network stream base | |
| Luo et al. | FRUDP: A reliable data transport protocol for aeronautical ad hoc networks | |
| Lundqvist et al. | TCP with end-to-end FEC | |
| CN110198273B (en) | Multi-path transmission method based on network coding in data center network | |
| US9887925B2 (en) | Network performance enhancement | |
| CN110324255B (en) | Data center network coding oriented switch/router cache queue management method | |
| CN115134307B (en) | Load balancing method based on packet loss rate coding in cloud computing | |
| CN114401208B (en) | Data transmission method and device, electronic equipment and storage medium | |
| US11502986B2 (en) | Reducing transmission delay of transmitting data in Wi-Fi | |
| Abdelmoumen et al. | Analysis of TCP latency over wireless links supporting FEC/ARQ-SR for error recovery | |
| Altahir et al. | Performance evaluation of TCP congestion control mechanisms using NS-2 | |
| Sreekumari et al. | A simple and efficient approach for reducing TCP timeouts due to lack of duplicate acknowledgments in data center networks | |
| Sato et al. | Improving TCP throughput using forward error correction | |
| Chen et al. | Effective retransmission in network coding for TCP | |
| Ha et al. | Making TCP/NC Adjustable to Time Varying Loss Rates. | |
| Ha et al. | TCP with network coding meets loss burstiness estimation for lossy networks |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |