HK40113402A - Methods for reducing packet loss during slow start stage of a session - Google Patents

Description

Methods to reduce packet loss during the slow start phase of a session

Technical Field

This invention generally relates to mobile wireless networks, and more specifically to systems and methods for modifying network protocol performance using mobile wireless network information.

Background Technology

Wireless connectivity is unpredictable, and packet loss is a frequent occurrence. The Transmission Control Protocol (TCP) is designed to assume that packet loss is a result of network congestion. To prevent network congestion, a series of congestion control algorithms are introduced, including a slow start phase and a congestion avoidance phase. According to the algorithms, in the slow start phase, the congestion window increases exponentially. When packet loss occurs, the congestion control process enters the congestion avoidance phase, where the congestion window increases linearly, rather than exponentially. To accelerate TCP, this invention employs a scheme utilizing multiple tunnels to increase the probability of successful data transmission at the beginning of a TCP session. This can be achieved by using duplicate data packets.

This avoids significant packet loss and maximizes the duration of the exponential increase.

Invention Summary

This invention discloses a method and system for transmitting data packets on a network device. According to the disclosure of this invention, the network device determines whether a received data packet belongs to a new session. When the data packet belongs to a new session, the network device selects a first connection according to a first selection strategy and selects at least one second connection according to a second selection strategy. Then, the network device transmits an Original Encapsulated Packet (OEP) through the first connection. The network device also transmits at least one Duplicate Encapsulated Packet (DEP) corresponding to each OEP through at least one second connection until a specific time period is reached. When the specific time period is reached, each OEP is transmitted through a third connection, and the transmission of at least one DEP is stopped.

The data packets received by the network device of the present invention may initially originate from a host connected to the network device. In one implementation of the present invention, the specific time period may be a time period within the slow start phase of the TCP data session establishment phase.

There is no limit to the number of DEPs transmitted for each OEP. Similarly, there is no limit to the number of connections used by the network device to transmit DEPs. In one variant, the OEP and the DEP are transmitted via aggregated connections.

Brief description of the attached figures

Figure 1 is a schematic block diagram of an exemplary network device 100 according to an embodiment of the present invention.

Figure 2 is a network diagram according to various embodiments of the present invention.

Figure 3 is a network diagram illustrating tunnels established between network devices according to various embodiments of the present invention.

Figure 4 is a flowchart illustrating the process performed at the network device during the slow start phase according to an embodiment of the present invention.

Figure 5A illustrates the relationship between the congestion window size and the time it takes for a TCP session to transmit data packets, as implemented according to a conventional congestion control process.

Figure 5B illustrates the relationship between the congestion window size and the time to transmit data packets in implementing a TCP session according to this congestion control procedure.

Figure 6A illustrates the structure of an OEP according to various embodiments of the present invention.

Figure 6B illustrates the structure of a DEP according to various embodiments of the present invention.

Specific Implementation

The present invention provides a system and method for reducing packet loss during the slow start phase to improve network performance, which helps to address the above-mentioned needs.

The following description provides only preferred and exemplary embodiments and is not intended to limit the scope, applicability, or configuration of the invention. Rather, the following description of preferred and exemplary embodiments will provide those skilled in the art with a description of feasible implementation of the preferred exemplary embodiments of the invention. It will be understood that various changes can be made to the function and arrangement of the elements without departing from the spirit and scope of the invention as set forth in the appended claims.

The processing unit executes program instructions or code segments for implementing embodiments of the present invention. Furthermore, embodiments can be implemented using hardware, software, firmware, middleware, microcode, hardware description languages, or any combination thereof. When embodiments are to be implemented using software, firmware, middleware, or microcode, program instructions for performing necessary tasks can be stored in a computer-readable storage medium. The processing unit can be implemented via virtualization and can be a virtual processing unit, including virtual processing units in cloud-based instances.

The program instructions constituting various embodiments may be stored in a storage medium. Furthermore, as disclosed herein, the term "computer-readable storage medium" can refer to one or more devices for storing data, including read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), random access memory (RAM), magnetic RAM, core memory, floppy disk, flexible disk, hard disk, magnetic tape, CD-ROM, flash memory device, memory card, and/or other machine-readable media for storing information. The term "computer-readable storage medium" may also include, but is not limited to, portable or fixed storage devices, optical storage media, magnetic media, memory chips or magnetic tape cassettes, wireless channels, and various other media capable of storing, containing, or carrying instructions and/or data. Computer-readable storage media can be implemented through virtualization and may be virtual computer-readable storage media, including virtual computer-readable storage media in cloud-based instances.

Embodiments of the present invention relate to implementing the techniques described herein using a computer system. In one embodiment, the processing unit of the present invention may reside on a machine such as a computer platform. According to one embodiment of the present invention, the techniques described herein are executed by a computer system in response to the processing unit executing one or more sequences of one or more instructions contained in the volatile memory. Such instructions may be read into the volatile memory from another computer-readable medium. Execution of the sequence of instructions contained in the volatile memory causes the processing unit to perform the processing steps described herein. In alternative embodiments, the present invention may be implemented using hardwired circuitry instead of or in combination with software instructions. Therefore, embodiments of the present invention are not limited to any particular combination of hardware circuitry and software.

Furthermore, it is worth noting that the embodiments described can be depicted as processes shown in flowcharts, data flow diagrams, structure diagrams, or block diagrams. Although a flowchart may describe the operations as a sequential process, many of the operations can be executed in parallel or simultaneously. Moreover, the order of the operations can be rearranged. When an operation is completed, the process terminates, but may have additional steps not included in the diagram. A process can correspond to a method, function, procedure, subroutine, subroutine, etc. When a process corresponds to a function, its termination corresponds to the function returning to the calling function or the main function.

Furthermore, as disclosed herein, the terms "auxiliary storage" and "main storage" can refer to one or more devices used for storing data, including ROM, RAM, magnetic RAM, core memory, disk storage media, optical storage media, flash memory devices, and/or other machine-readable media used for storing information. The term "machine-readable media" includes, but is not limited to, portable or fixed storage devices, optical storage devices, wireless channels, and various other media capable of storing, containing, or carrying instructions and/or data. Machine-readable media can be implemented through virtualization and can be virtual machine-readable media, including virtual machine-readable media in cloud-based instances.

Furthermore, embodiments can be implemented using hardware, software, firmware, middleware, microcode, hardware description languages, or any combination thereof. When implemented as software, firmware, middleware, or microcode, the program code or code segments used to perform the necessary tasks can be stored in a machine-readable medium, such as a storage medium. The processing unit can perform the necessary tasks. The processing unit can be a CPU, ASIC semiconductor chip, semiconductor chip, logic unit, digital processor, analog processor, FPGA, or any processor capable of performing logic and arithmetic functions. Code segments can represent any combination of procedures, functions, subroutines, programs, routines, subroutines, modules, software packages, classes, or instructions, data structures, or program statements. Code segments can be coupled to another code segment or hardware circuitry by passing and/or receiving information, data, parameters, arguments, or memory contents. Information, arguments, parameters, data, etc., can be passed, forwarded, or transmitted in any suitable manner, including memory sharing, message passing, token passing, network transmission, etc. The processing unit can be implemented via virtualization and can be a virtual processing unit, including virtual processing units in cloud-based instances.

End-to-end connections can use connection-oriented protocols, such as TCP, or connectionless protocols, such as User Datagram Protocol (UDP), to transmit data packets. Well-known protocols for deploying end-to-end connections include Layer 2 Tunneling Protocol (L2TP), Secure Shell (SSH), Multiprotocol Label Switching (MPLS), and Microsoft's Point-to-Point Tunneling Protocol (PPTP). Connections to network interfaces can take the form of fiber optics, Ethernet, ATM, Frame Relay, T1/E1, IPv4, IPv6, wireless technologies, Wi-Fi, WiMax, High-Speed Packet Access (HSPA), 3GPP, and Long Term Evolution (LTE).

Figure 1 is a schematic block diagram of an exemplary network device 100 according to an embodiment of the present invention. The network device 100 includes at least one storage unit 101, at least one I/O interface 102, at least one memory 103, at least one processing unit 104, and a plurality of network interfaces 105. The storage unit 101 may contain computer program instructions executable by the processing unit 104 to implement one or more aspects of the present disclosure. The processing unit 104 and the memory 103 are directly interconnected. A system bus 106 directly or indirectly connects the processing unit 104 to at least one storage unit 101, at least one I/O interface 102, and the plurality of network interfaces 105.

In another embodiment, the network device 100 may further include at least one SIM interface 107 and at least one wireless communication module (WCM) 108 with at least one antenna 109. A SIM or electronic SIM can be housed in or coupled to the SIM interface of at least one SIM interface 107, and establish a wide area network (WAN) connection through at least one WCM 108 with at least one antenna 109.

Network device 100 can be connected to a wired or wireless access network via at least one network interface 105. The wired or wireless access network can carry one or more network protocol data. The wired access network can be implemented using Ethernet, fiber optic cable, cable, DSL, Frame Relay, Token Ring, serial bus, USB, FireWire, PCI, or any material capable of transmitting information. The wireless access network can be implemented using infrared, LTE, 5G, Low Earth Orbit (LEO) satellite, HSPA, HSPA+, WiMax, GPRS, EDGE, GSM, CDMA, Wi-Fi, CDMA2000, WCDMA, TD-SCDMA, Bluetooth, WiBRO, Evolved Data Optimized (EV-DO); Digital Enhanced Cordless Telecommunications (DECT); Digital AMPS (IS-136/TDMA); Integrated Digital Enhancement (iDEN); or any other wireless technology. A WAN connection can be established on the access network to connect two networks.

Figure 2 is a network diagram according to various embodiments of the present invention. Network devices 200 and 202 are similar to network device 100 shown in Figure 1. Network devices 200 and 202 are located at sites 205 and 206, respectively. Network devices 200 and 202 may be embodied as multi-WAN network devices supporting the aggregation of bandwidth for multiple WAN-to-WAN connections. Network devices 200 and 202 are connected via interconnection network 204. Interconnection network 204 is an interconnection network that may include a local area network (LAN), such as a corporate intranet, a metropolitan area network (MAN), or a WAN, such as the Internet or the World Wide Web.

At site 205, laptop computer 201a and desktop computer 201b are connected to network device 200 via the first network interface of network device 200. At site 206, server 203 is connected to network device 202 via the first network interface of network device 202.

In one embodiment, site 205 and site 206 are located in the same region or country, such as the United States.

In another embodiment, site 205 and site 206 are located in different regions or countries, such as the United States and Singapore, respectively.

Network device 200 includes at least one network interface capable of connecting to an access network, and network device 202 includes at least one network interface capable of connecting to an access network. The access network is a network that enables access to the interconnected network 204. Laptop 201a and desktop 201b can be connected to network device 200 via the first network interface of network device 200. Similarly, server 203 can be connected to network device 202 via the first network interface of network device 202. Server 203 may be located within the LAN of network device 202.

In one embodiment, each of the at least one network interface of network device 200 and each of the at least one network interface of network device 202 may be a WAN interface. In one embodiment, the WAN interface may be implemented using a WCM and an antenna, which enables connection to the base station of the Internet service provider.

In one variant, the antenna is a non-omnidirectional antenna used for LEO satellites.

In another embodiment, each of the at least one network interface of network device 200 and each of the at least one network interface of network device 202 may be a LAN interface. Each of network devices 200 and 202 is capable of connecting to the interconnected network via the LAN interface by configuring the LAN interface as a WAN interface.

Network device 200 may include three network interfaces capable of connecting to three access networks 207a-c, while network device 202 may include two network interfaces capable of connecting to two access networks 208a-b. Therefore, six connections can be created through access networks 207a-c and 208a-b. However, these configurations may vary depending on the required network devices and configurations. There is no limitation on the number of network interfaces for network devices 200 and 202; the specified three network interfaces for network device 200 and two network interfaces for network device 202 are for illustrative purposes only. As shown in Figure 3, the connection is established between network device 200 and network device 202.

Figure 3 illustrates the connection established between two network devices according to an exemplary embodiment of the present invention. For clarity, Figure 3 should be viewed in conjunction with Figure 2. At least one end-to-end connection can be established between hosts within the same LAN of network device 200 and hosts within the same LAN of network device 202 through at least one connection established between network devices 200 and 202. Each connection can be established through the network interfaces of network device 200 and network device 202. If network device 200 includes M network interfaces and network device 202 includes N network interfaces, then M*N connections are established between network devices 200 and 202.

For example, if network device 200 includes three network interfaces and network device 202 includes two network interfaces, then when establishing a connection between the network interfaces of network devices 200 and 202, six connections 209a-f can be created, as shown in Figure 3. The at least one end-to-end connection established between laptop 201a and server 203 can be established through at least one of these six connections.

The values of M and N can vary depending on the required device and configuration. There is no limit to the number of M and N; M and N can be any positive integer such that the sum of M and N is equal to or greater than three. In another example, M equals one and N equals two, thus creating two connections.

The access networks of network devices 200 and 202 are access connections used to transmit information within the interconnection network 204 between sites 205 and 206.

Connections 209a-f can have similar or different bandwidth capabilities. Furthermore, the access networks of network devices 200 and 202 can include different types of WAN connections, such as Wi-Fi, cable, DSL, 3G, 4G, LTE, 5G, satellite connections, etc. It should also be noted that sites 205 and 206 can be considered as the sender and the receiver, and discussions regarding the functionality of either site can be implemented at the other site. Connections 209a-f correspond to a unique arrangement of the access networks of site 205 and site 206.

In one embodiment, an aggregated connection is formed between network devices 200 and 202. The multiple connections may be aggregated, combined, or bound together to form a single aggregated connection. Those skilled in the art will understand that there are numerous ways to aggregate, combine, or bind multiple connections to form a single aggregated connection. The aggregated connection is treated as a single connection by the session or application using it. An aggregated connection may be considered a tunnel, a Virtual Private Network (VPN) connection, or a connection-oriented connection.

For example, the aggregated connection is an aggregation of multiple connections (e.g., connections 209a-f) established between network devices 200 and 202. The aggregated connection can be created when a TCP session is created. For example, when laptop 201a attempts to connect to server 203, the aggregated connection is established between laptop 201a and server 203 via network devices 200 and 202.

In another example, the aggregated connection is a VPN tunnel, comprising multiple tunnels, each established between the network interfaces of network devices 200 and 202.

In one example, the aggregated connection is a TCP connection.

In another example, the aggregated connection is a UDP connection.

For ease of explanation, when laptop 201a attempts to establish a connection with server 203 through network devices 200 and 202, a TCP session is created. After the TCP session is created, network device 200 can transmit data packets through the network interface of network device 200 and the network interface of network device 202. At this point, the congestion control process begins.

In one embodiment, data packets are transmitted through connections in the aggregated connection according to the weights assigned to each connection, and a default weight is determined for each connection in the aggregated connection. The default weights are initially assigned to each connection by default before any modification or change to the weights of the connections.

In one example, a default weight is determined and/or assigned to each access network based on the uplink and downlink bandwidth data of the exchange for each access network.

In another example, default weights are determined and/or assigned to each access network based on criteria that may affect the performance of the access network.

Assume that the downlink bandwidth of access network 1 to a of site 205 is d1, d2, ..., dm, and the uplink bandwidth of access network 1 to b of site 206 is u1, u2, ..., un; the default weight of the connection between access network x of site 205 and access network y of site 104 can be defined as DW(x,y), where DW(x,y) = dx·uy.

The default weights can be calculated using the method described above. If the uplink/downlink bandwidths of the three access networks of network device 200 are 10M/6M, 8M/4M, and 6M/6M respectively, and the uplink/downlink bandwidths of the two access networks of network device 202 are 7M/5M and 9M/3M respectively, then the default weight for each connection is:

Network equipment 200 Network equipment 202 DW(1,1) = 6M * 7M = 42M DW(1,1) = 5M * 10M = 50M DW(1,2) = 6M * 9M = 54M DW(1,2) = 5M * 8M = 40M DW(2,1)＝4M*7M＝28M DW(1,3) = 5M * 6M = 30M DW(2,2)＝4M*9M＝36M DW(2,1)＝3M*10M＝30M DW(3,1)＝6M*7M＝42M DW(2,2)＝3M*8M＝24M DW(3,2) = 6M * 9M = 54M DW(2,3)＝3M*6M＝18M

There are many ways to calculate the default weights; the methods described above are for illustrative purposes only.

In another embodiment, data packets are transmitted through a group of connections of the aggregated connection. Each connection of the aggregated connection is classified into a different connection group, and data is transmitted through the connection at least in part based on the group to which the connection is classified. There is no limit to the number of groups to which a connection is classified, and a connection can be classified into multiple connection groups according to the criteria of each connection group.

Two main parameters are used to control congestion in established connections: the congestion window size (cwnd) and the slow start threshold (ssthresh). The established connections are controlled by the network device through four phases of the congestion control process, specifically by controlling the packet transmission rate of the connection.

(i) Slow start phase, (ii) Congestion avoidance phase, (iii) Fast retransmit phase, and (iv) Fast recovery phase.

In a preferred embodiment, the congestion control process begins after the session is established, prior to the three-way handshake.

In another embodiment, the congestion control process begins after a three-way handshake, once the session is established.

When network device 200 receives the first data packet destined for server 203 from laptop 201a, server 203 is locally connected to network device 202 via LAN. Network device 200 can then send the data packet to network device 202 through the established aggregated connection. At this time, the congestion window size becomes 1. Those skilled in the art will understand that after receiving the first data packet, network device 200 can receive more data packets. If network device 200 receives more data packets destined for server 203 from laptop 201a, network device 200 can send subsequent data packets after receiving the acknowledgment of the previous data packet, depending on the congestion window size.

When server 203 successfully receives the first data packet, network device 200 can then receive an acknowledgment of the first data packet from server 203 via network device 202, and the size of the congestion window increases until the slow start threshold is reached. For ease of explanation, after receiving the acknowledgment, the size of the congestion window is increased by doubling the original size (new cwnd = cwnd * 2).

When network device 200 successfully receives the acknowledgment of the first data packet from server 203 via network device 202, network device 200 may attempt to transmit the second and third data packets to server 203 via network device 202, and the size of the congestion window becomes 2. The second and third data packets are received from laptop 201a. When network device 200 successfully receives the acknowledgment of the second and third data packets from server 203 via network device 202, network device 200 may attempt to transmit the fourth, fifth, sixth, and seventh data packets to server 203 via network device 202, and the size of the congestion window becomes 4. Therefore, the size of the congestion window increases exponentially until the slow start threshold (i.e., cwnd = ssthresh) is reached.

The congestion window size does not necessarily have to increase exponentially; any other suitable algorithm for increasing the congestion window size can be applied. For example, upon receiving an acknowledgment, the congestion window size can be increased by 2 (new cwnd = cwnd + 2).

When the congestion window size reaches the slow start threshold, packet transmission transitions from the slow start phase to the congestion avoidance phase to repair network failure. During the congestion avoidance phase, the congestion window size no longer increases exponentially but linearly. After a period of time, the TCP transmission rate exceeds the network connection capacity, leading to packet loss. However, packet loss or timeouts may occur before the congestion avoidance phase, i.e., during the slow start phase.

Figure 4 is a flowchart illustrating a process performed at the network device during the slow start phase according to an embodiment of the present invention.

When the network device receives data packets from a host, such as a laptop 201a, the process will begin for each received data packet.

In process 401, network device 200 can determine whether a new session has been established. The session can be a TCP session or a UDP session.

In one embodiment, the network device determines whether a new session has been established by identifying whether the received data packet belongs to a new session. For example, to determine whether the received data packet is the first data packet of a new session, the processing unit of the network device 200 may read one or more of the following information contained in the header of the received data packet: source address, source port, destination address, destination port, and protocol.

When the received data packet is not the first data packet of a new session, the network device routes the received data packet through a pre-selected connection and continues to apply determination process 401 to the next received data packet. Once the first data packet of the data session to which the received data packet belongs is determined, the pre-selected connection can be selected. In one variant, when the received data packet is not the first data packet of a new session, the network device determines whether the received data packet is the last data packet of the existing session. If the received data packet is the last data packet of the existing session, then obviously the next data packet to be received will be the first data packet of the new session. Therefore, the network device can perform processes 402 and 403 in advance before receiving the first data packet of the new session. Therefore, when the next data packet is received, the network device will not need to perform processes 401-403, but can skip to process 404. This will significantly facilitate the present invention by accelerating the establishment of data sessions and reducing the processing load during the new data session establishment phase. Whether the received data packet is the last data packet of the existing session can be determined based on one or more of the source address, source port, destination address, destination port, and/or protocol of the received data packet.

On the other hand, when the data packet is the first data packet of a new session, the network device can execute process 402 and subsequent processes.

In one variant, network device 200 may determine whether the received data packet is the second or third data packet of a new session, rather than the first data packet. When the received data packet is the second or third data packet of a new session, the network device may execute process 402 and subsequent processes.

In process 402, network device 200 may select a first connection to transmit data packets belonging to a new session. The first connection is a connection established between the network interfaces of network device 200 and network device 202. The first connection is selected according to a first selection strategy.

In process 403, network device 200 may select at least one second connection according to a second selection strategy. The at least one second connection is at least one connection other than the first connection established between the network interface of network device 200 and the network interface of network device 202.

The first and second selection strategies are chosen from one or more of the following: weight balancing, least usage, lowest latency, and priority. Therefore, data packets can be transmitted in a manner that reduces the likelihood of packet loss during the slow start period.

In one embodiment, the first selection strategy is the same as the second selection strategy.

In another embodiment, the first selection strategy is different from the second selection strategy.

In one example, if "weight balancing" is applied at network device 200 to select the connection, the data packets to be transmitted through the connection are allocated according to the weight of each connection in the aggregated connection. The weights may be the default weights calculated for each connection as described above. For illustrative purposes, the weights of each connection in the aggregated connection are related to bandwidth, and the default weights are calculated based on the bandwidth of each connection. During the transmission of data packets through multiple connections from network device 200 to network device 202, the bandwidth of each of the multiple connections may be different, and therefore the weights may also be different. The network device can transmit data packets through various connections based on the changes in the weights of each of the multiple connections.

In another example, if "least used" is applied at network device 200 to select the connection, then the data packets to be transmitted through the connection are allocated based on the traffic of each connection in the aggregated connections. If multiple connections are available, the connection with the lowest traffic in the aggregated connections is preferentially selected to transmit more data packets.

In another example, if "latency" is used to select the connection at network device 200, then the data packets to be transmitted through the connection are allocated according to the latency of each connection in the aggregated connection. If two connections are available with latencies of 100ms and 40ms respectively, the connection with the 40ms latency is preferred, meaning the connection with the lower latency transmits more data packets.

In another example, if a "priority" is used to select the connection at network device 200, then data packets to be transmitted through the connection are allocated according to the priority of each connection in the aggregated connections. The priorities can be assigned by a user or administrator as required or as needed. During the transmission of data packets from network device 200 to network device 202 through the plurality of connections, the connection with the highest priority is preferentially selected for data packet transmission. If the highest priority connection is unavailable at any given time, the next highest priority connection is preferentially selected.

The user or administrator can configure the first selection policy and the second selection policy of the network device 200 locally or remotely by sending configurations through a web interface, application programming interface (API), command line interface or console.

In one variant, multiple first selection strategies and/or multiple second selection strategies can be provided simultaneously. For example, when the packet is transmitted to a specific destination port, priority is applied to select the connection, while when the packet is transmitted from another specific destination port, weight balancing is applied. The user or administrator can configure the priority of the strategies to be applied so that there are no conflicts between the applied strategies.

In process 404, network device 200 can transmit Original Encapsulated Packets (OEPs) through the first connection. Each OEP encapsulates the data packet and has an Original Encapsulated Packet Global Sequence Number (OEP-GSN).

In process 405, network device 200 may transmit at least one duplicate encapsulation packet (DEP) via at least one second connection within a specific time period. The at least one DEP encapsulates at least one of the data packets. Each DEP has a DEP Global Sequence Number (DEP-GSN), which is stored in the header of the DEP. Details of the OEP and the DEP will be discussed below.

Those skilled in the art will understand that in wireless communication, the quality of data packet transmission can be unpredictable, and latency can be high in some cases. The packet loss rate can also be high. Transmitting a DEP increases the chances of the data being received by network device 202 in case the OEP is lost. However, transmitting the DEP can significantly increase bandwidth utilization. Therefore, the data packets will only be copied and transmitted within a specific time period.

In process 406, after a specific time period, network device 200 may continue to transmit the OEP through the third connection, but no longer transmit the at least one DEP through the at least one second connection.

Typically, processes 404 and 405 are executed after the slow start phase begins. During the slow start phase, the number of data packets transmitted through the connection increases exponentially. By transmitting the data packets through the aggregated connection during the slow start phase, the likelihood of packet loss and latency can be reduced, and the rate of data packet transmission through the aggregated connection established between network devices 200 and 202 can be maximized.

However, if packet loss occurs during the slow start phase, the number of packets that can be interrupted during a specific time period increases exponentially, and the congestion control process moves to the congestion avoidance phase, as shown in Figure 5B.

After a specific period of time has elapsed since the start of the slow start phase, the data packets will no longer be transmitted through the one or more connections established between network device 200 and network device 202 using the first selection strategy and/or the second selection strategy. Instead, the data packets will be transmitted through the third connection established between network device 200 and network device 202.

In one embodiment, the third connection is the same as the first connection selected in step 402.

In another embodiment, the third connection is different from the first connection selected in step 402. The network device 200 can select another connection to transmit the data packets in any way, such as the second selection strategy.

Factors determining the specific time period include one or more of the following: the duration of the slow start phase, bandwidth, latency, and any factors that may affect data transmission. There are no restrictions on how the specific time period is determined. The specific time period can be any time between the start and end of the slow start phase.

In one example, the specific time period is 2 to 4 seconds from the start of the slow start phase.

In another example, the specific time period is 3 seconds, calculated from the start of the slow start phase.

Figure 5A illustrates the relationship between the congestion window size and the transmission time of data packets in a TCP session, implemented according to a conventional congestion control process. The data packets are transmitted through connections among the plurality of connections established between network devices 200 and 202. No policy is applied to control the transmission of data packets through the aggregated connections; the data packets can be transmitted through the aggregated connections. As previously described, the congestion control process comprises four phases, but for the purpose of illustrating an embodiment of the invention, Figure 5A only shows the slow start phase 501a and the congestion avoidance phase 502a.

During the slow start phase 501a, the size of the congestion window continues to grow exponentially from its initial size. For clarity, when the packet transmission time of the TCP session reaches 505a, the size of the congestion window continues to grow exponentially from its initial size of congestion window 503a to the slow start threshold 504a. However, packet loss may occur during this slow start phase.

When packet loss occurs during the slow start phase, the congestion control process moves from the slow start phase 501a to the congestion avoidance phase 502a due to detected congestion, even if the congestion window size has not reached the slow start threshold 504a. For example, when packet loss occurs at 507a, the congestion control process moves from the slow start phase 501a to the congestion avoidance phase 502a, even if the congestion window size only reaches 508a instead of 504a.

During congestion avoidance phase 502a, the congestion window size can only increase linearly until there is indication that duplicate acknowledgments (ACKs) have been received. In one embodiment, a congestion avoidance algorithm can be applied to control the congestion.

Therefore, network device 200 can send subsequent data packets based on the applied congestion avoidance algorithm. The congestion avoidance algorithm can be selected from one or more of the following: TCP Reno, TCP Vegas, TCP-Fast, CompoundTCP, TCP Veno, TCP Westwood, TCP Cubic, and TCP BIC.

There are no restrictions on the congestion avoidance algorithm used to control the congestion; any congestion avoidance algorithm, including various aspects of the additive increase/multiplicative decrease (AIMD) scheme, can be applied.

For clarity, at 506a, network device 200 has received a duplicate ACK. Therefore, the size of the congestion window is reset to the initial size of congestion window 503a, and the process of increasing the congestion window size restarts (e.g., the congestion window size is reset to 1), and the slow start threshold is reduced (e.g., ssthresh = current cwnd/2). The size of the congestion window after the reset is not limited and may vary depending on the applied congestion avoidance algorithm.

To reduce packet loss during the slow start phase shown in Figure 5A, according to an embodiment of the invention, the method proposed in this invention is applied during the slow start phase. Therefore, the relationship between the congestion window size and the packet transmission time of the TCP session shown in Figure 5B differs from that shown in Figure 5A.

Figure 5B illustrates the relationship between the congestion window size and the packet transmission time of the TCP session, implemented according to the congestion control process, in the event of packet loss during the slow start phase. Figure 5B should be viewed in conjunction with Figure 2 and compared with Figure 5A. The packets are transmitted through the multiple connections established between network devices 200 and 202. Similar to Figure 5A, Figure 5B only shows the slow start phase 501b and the congestion avoidance phase 502b to illustrate embodiments of the invention.

In one variant, the multiple connections are aggregated into a single aggregate connection.

When the transmission of a session's data packets begins, the congestion control process starts and enters the slow start phase 501b. Once the first data packet of the session is determined, network device 200 can transmit the OEP and the DEP through the multiple connections established between network devices 200 and 202.

During the slow start phase 501b, the size of the congestion window continues to grow exponentially from the initial size of the congestion window, similar to the slow start phase 501a shown in FIG5A.

For clarity, when the time for transmitting TCP session packets reaches 505b, the size of the congestion window continues to grow exponentially from the initial size of 503b to the slow start threshold of 504b. However, packet loss may occur during the slow start phase.

Because the method shown in Figure 4 is applied through the multiple connections established between network devices 200 and 202, packet loss can be reduced. For example, when data packets are transmitted through the multiple connections at network device 200, the data packets are encapsulated as an OEP and at least one DEP, respectively. Network device 200 can transmit the OEP through the first connection and the at least one DEP through the at least one second connection within a specific time period. Therefore, transmitting OEP and DEP simultaneously through the multiple connections increases the chances of data being received by network device 202.

For example, data packets are transmitted from laptop 201a to server 203 via connection 209a for OEP and via connection 209c for DEP. Packet loss may occur when transmitting the OEP via connection 209a. When transmitting the DEP via connection 209c, network device 202 may successfully receive the DEP. Therefore, network device 202 receives an acknowledgment of the data packet from server 203. As a result, the size of the congestion window will continue to grow exponentially, and the congestion control process will not switch from slow start phase 501b to congestion avoidance phase 502b. This will prolong the duration of slow start phase 501b.

Therefore, compared to the conventional technique disclosed in Figure 5A, the congestion avoidance phase will begin with a higher congestion window size according to the technique disclosed in Figure 5B. Achieving a higher congestion window size during the slow start phase increases the probability that the congestion window size will reach the slow start threshold; whereas according to the conventional technique shown in Figure 5A, the congestion control process may trigger the congestion avoidance phase before reaching the slow start threshold.

When the congestion window size reaches the slow start threshold 504b, the congestion control process switches from the slow start phase 501b to the congestion avoidance phase 502b. Since a higher congestion window size can be achieved according to the present invention, the slow start threshold 504b can also be set at a higher level. This will help to increase the data transmission bandwidth very quickly.

During congestion avoidance phase 502b, the congestion window size increases linearly until there is indication that a duplicate ACK has been received. Compared to Figure 5A, more data packets can be transmitted during this congestion avoidance phase.

In order to monitor the bandwidth of various connections 209a-f, some embodiments of the present invention encapsulate each transmitted IP packet with various information.

Figure 6A illustrates the structure of an OEP according to various embodiments. OEP 600 is an encapsulation packet that encapsulates data packets along with various information for packet transmission. OEP 600 includes an IP header 601, an OEP-GSN 602, a tunnel sequence number 603, other information 604, and an encapsulated data packet 605. OEP-GSN 602 stores the OEP-GSN of OEP 600.

Each tunnel sequence number 603 can be a 32-bit field representing the sequence number assigned to each packet routed to a specific tunnel. The payload field of the IP packet can be encrypted using the Advanced Encryption Standard (AES) to transmit the payload. AES encryption can be applied to enhance the security of the payload and prevent attacks from third parties.

The IP header 601 may include one or more of the following: version field, protocol type field, tunnel ID field, and global sequence number field. Other information 604 may include an AES initialization vector field and a tunnel ID field.

The version field may contain information about the protocol version being used, while the protocol type field may contain the protocol type of the payload packet. Typically, the value of this field will correspond to the Ethernet protocol type of the packet. However, other values may be defined in other documents. The tunnel ID field may be a 32-bit field and may contain an identifier for identifying the current tunnel of the IP packet.

The AES initialization vector field may be a 32-bit field and may contain an initialization vector for AES encryption. The global sequence number field may be a 32-bit field and may contain a sequence number for reordering each packet of each session to the correct order as it emerges from its respective tunnel.

In one variant, each tunnel sequence number 603 and the tunnel ID field can be used to compare tunnel performance and reallocate packets based on the tunnel performance.

In another variant, each tunnel sequence number 603 can be a session number or a stream number.

In another variant, each tunnel sequence number 603 can be optional. When each tunnel sequence number 603 and the tunnel ID field are not included, it may be necessary to maintain a database to store information about which tunnel is used to transmit encapsulated packets for which GSN. Tunnel performance can be compared by performing a lookup in that database. Therefore, including each tunnel sequence number 603 and the tunnel ID field in the OEP can help determine latency without maintaining and looking up a database for all GSNs.

In one variant, OEP-GSN 602, each tunnel sequence number 603, and other information 604 are included in the options field of the IP header 601, and the encapsulated packet 605 is included in the AES-encrypted payload field. The advantage of including OEP-GSN 602, each tunnel sequence number 603, and other information 604 in the options field of the IP header 601 is that the information contained in these fields can be retrieved without decapsulating the OEP 600, which can significantly reduce processing time. However, the IP header 601 may change as the OEP 600 is transmitted through the tunnels, routers, and devices, so the information in these fields may be lost if it is included in the options field of the IP header 601. Therefore, another option is that OEP-GSN 602, each tunnel sequence number 603, other information 604, and encapsulated packet 605 can all be included in the AES-encrypted payload field 612 so that the information is preserved when the OEP 600 passes through routers and devices that may alter the IP header 601.

Figure 6B illustrates the structure of a DEP according to various embodiments. DEP 610 is an encapsulated data packet that encapsulates data packets together with various information for data packet transmission. Similar to OEP 600, DEP 610 includes an IP header 611, a DEP-GSN 612, a tunnel sequence number 613, other information 614, and an encapsulated data packet 615.

OEP-GSN 602 and DEP-GSN 612 can be identical. Encapsulated data packets 605 and 615 can also be identical, such that OEP 600 and DEP 610 encapsulate the same data packets, which become encapsulated data packets 605 and 615, respectively. When network device 200 generates at least one DEP 610 for transmission, the payload, i.e., the encapsulated data packet, is the same as the payload of OEP 600. Since OEP 600 and DEP 610 have the same content, their global sequence numbers are also the same. Because DEP 610 contains the same encapsulated data packet as OEP 600, if OEP 600 is lost, dropped, or delayed during transmission, and DEP 610 is received before OEP 600 is received, the information in OEP 600 can be recreated using DEP 610.