WO2024085281A1

WO2024085281A1 - Network interface device, operating method thereof, and server device including same

Info

Publication number: WO2024085281A1
Application number: PCT/KR2022/016096
Authority: WO
Inventors: 부준혁
Original assignee: 주식회사 망고부스트; 서울대학교산학협력단
Priority date: 2022-10-20
Filing date: 2022-10-20
Publication date: 2024-04-25

Abstract

Disclosed are a TOE-based network interface device, operating method thereof, and server device including same. The present invention may comprise: an event scheduler that generates connection information for a requested event and outputs a first control signal by performing scheduling on the connection information; a first TCP controller and a second TCP controller that each receive the connection information from the event scheduler and control a calculation operation for the connection information; and a memory control logic that transmits the connection information to a first memory or stores the connection information in a second memory, in response to the first control signal.

Description

Network interface device, method of operation thereof, and server device including the same

The present invention relates to network interfaces, and particularly to a TOE (TCP/IP Offload Engine)-based network interface device, its operating method, and a server device including the same.

The areas where big data or artificial intelligence are applied are expanding, and the use of OTT (Over-The-Top) services is becoming routine. Accordingly, the data required for various devices connected through the network, such as user terminals, web servers, web application servers (WAS), storage servers, and database servers, etc. The amount of processing is rapidly increasing and the processing speed is also accelerating.

Accordingly, when data communication is performed based on the TCP/IP protocol (Transmission Control Protocol/Internet Protocol) in a high-speed network environment of several tens or hundreds of Gbps (Gigabits per second) or more, enormous system resources may be required. For example, when a web server performs data communication with a user terminal through a network, a significant load may be incurred when the web server's central processing unit, that is, the host CPU (Central Processing Unit), performs TCP/IP operations. .

In this case, performance degradation of the entire web server may occur. Additionally, processing of network communications performed through a web server may be delayed.

The present invention is intended to solve the above-described problems, and seeks to provide a TOE-based network interface device that can reduce the load on the host CPU and accelerate data processing in a high-speed network, a method of operating the same, and a server device including the same.

A network interface device according to an embodiment of the present invention for solving the above technical problem includes an event scheduler that generates connection information for a requested event, performs scheduling on the connection information, and outputs a first control signal; a first TCP controller and a second TCP controller, respectively, receiving the connection information from the event scheduler and controlling calculation operations on the connection information; and memory control logic that transmits the connection information to a first memory or stores it in a second memory in response to the first control signal.

The event scheduler may transmit the connection information to at least one of the first TCP controller, the second TCP controller, and the memory control logic.

The first memory may be included in the first TCP controller and the second TCP controller.

The second memory may include a dynamic random access memory (DRAM) located outside the first chip equipped with the first TCP controller and the second TCP controller.

The memory control logic may be provided on a second chip together with the second memory.

The memory control logic may be provided in the same first chip as the first TCP controller and the second TCP controller.

The connection information stored in the first memory may be moved to the second memory if a corresponding event does not occur for a first time.

The memory control logic may generate a second control signal indicating whether the connection information is moved to the first memory when an event regarding the connection information stored in the second memory occurs.

The connection information stored in the second memory may be moved to one of the first memories of the first TCP controller and the first memory of the second TCP controller.

A first TCP operation that performs an operation by receiving the connection information and event information corresponding to the connection information stored in a first memory from a corresponding TCP controller among the first TCP controller and the second TCP controller, respectively. logic and second TCP operation logic; may be further included.

third to nth TCP controllers, respectively, controlling calculation operations on the connection information and the event information in response to the first control signal; and first TCP operation logic to perform an operation by receiving the connection information and the event information stored in the first memory included from the corresponding TCP controller among the third TCP controller to the nth TCP controller, respectively. Operation logic may be further included.

The n and the m may be different.

The event scheduler may include a mapping table for a storage location of the connection information.

a reception processing module that parses a data packet received through a network interface, outputs user data of the data packet through a host interface, and transmits metadata of the data packet as an event to the event scheduler; and a transmission processing module that generates a header corresponding to the connection information in data received through the host interface and outputs it as a data packet.

The method of operating a network interface device according to an embodiment of the present invention to solve the above technical problem is a method of operating a TOE (TCP/IP Offload Engine)-based network interface device, and is a method of operating a network interface device that transmits connection information corresponding to an event to the TCP controller. 1 storing in memory; Checking occurrence of an event corresponding to connection information stored in the first memory; and, when an event corresponding to the connection information does not occur for a first time, moving the connection information stored in the first memory to a second memory.

Storing the connection information in the first memory may include selecting one of the plurality of TCP controllers; and storing the connection information in the first memory of the selected TCP controller.

When an event corresponding to the connection information moved to the second memory occurs, determining whether to transmit the connection information moved to the second memory to the first memory; selecting one of the plurality of TCP controllers; and transmitting the connection information moved to the second memory to the first memory of the selected TCP controller.

A server device according to an embodiment of the present invention for solving the above technical problem is a network including a TCP/IP hardware stack and a TCP/IP software stack that performs TCP/IP operations on received data packets. interface module; and a host module that receives and processes the data of the data packet from the network interface module, wherein the network interface module stores connection information corresponding to the data packet in the first memory of the TCP controller or controls the memory. Stored in a second memory controlled by logic.

The network interface module may include multiple TCP controllers.

The memory control logic may be provided on the same chip as the second memory.

According to a network interface device, a method of operating the same, and a server device including the same according to an embodiment of the present invention, a network adapter performs TCP/IP processing to reduce the load on the host CPU and provide information on requested events. By storing connection information in one of the first memory and the second memory and performing optimized scheduling, data processing in a high-speed network can be accelerated.

Alternatively, according to a network interface device, a method of operating the same, and a server device including the same according to an embodiment of the present invention, the load on the host CPU is reduced by performing TCP/IP processing on the network adapter, and the requested event is transmitted to the TCP controllers. By being distributed and processed in parallel, data processing in high-speed networks can be accelerated.

1 is a diagram showing a network interface device according to an embodiment of the present invention.

Figure 2 is a diagram showing data packets processed by a network interface device according to an embodiment of the present invention.

3 to 5 are diagrams showing memory control logic according to an embodiment of the present invention, respectively.

Figures 6 and 7 are diagrams showing a network interface device based on a TOE (TCP/IP Offload Engine) according to an embodiment of the present invention, respectively.

Figures 8A to 8C are diagrams showing the relationship between a TCP controller and TCP operation logic, respectively, according to an embodiment of the present invention.

Figure 9 is a flowchart showing a method of operating a network interface device according to an embodiment of the present invention.

FIG. 10 is a diagram illustrating a network interface device operating according to the operation method of FIG. 9.

Figures 11A and 11B are diagrams showing in more detail the operating method of the network interface device of Figure 9, respectively.

Figures 12A and 12B are diagrams for explaining an operation of retransmitting connection information to the first memory according to an embodiment of the present invention, respectively.

Figure 13 is a diagram showing a network interface device according to an embodiment of the present invention.

Figure 14 is a diagram showing a server device according to an embodiment of the present invention.

Hereinafter, specific embodiments of the present invention will be described with reference to the drawings. The detailed description below is provided to provide a comprehensive understanding of the methods, devices and/or systems described herein. However, this is only an example and the present invention is not limited thereto.

In describing the embodiments of the present invention, if it is determined that a detailed description of the known technology related to the present invention may unnecessarily obscure the gist of the present invention, the detailed description will be omitted. In addition, the terms described below are terms defined in consideration of functions in the present invention, and may vary depending on the intention or custom of the user or operator. Therefore, the definition should be made based on the contents throughout this specification.

The terminology used in the detailed description is only for describing embodiments of the present invention and should in no way be limiting. Unless explicitly stated otherwise, singular forms include plural meanings. In this description, expressions such as “comprising” or “including” are intended to indicate certain features, numbers, steps, operations, elements, parts or combinations thereof, and one or more than those described. It should not be construed to exclude the existence or possibility of any other characteristic, number, step, operation, element, or part or combination thereof.

Terms such as first, second, etc. may be used to describe various components, but the components should not be limited by the terms. The above terms may be used for the purpose of distinguishing one component from another component. For example, a first component may be referred to as a second component, and similarly, the second component may also be referred to as a first component without departing from the scope of the present invention.

Hereinafter, embodiments of the present invention will be described clearly and in detail so that a person skilled in the art can easily practice the present invention.

FIG. 1 is a diagram illustrating a network interface device 100 according to an embodiment of the present invention, and FIG. 2 is a diagram illustrating a data packet (DPK) processed by the network interface device 100 according to an embodiment of the present invention.

Referring to Figures 1 and 2, the network interface device 100 according to an embodiment of the present invention includes an event scheduler 120, a TCP controller 140, and a memory control logic 160, to store data in a high-speed network. Processing can be accelerated efficiently.

The event scheduler 120 generates connection information (ICT) for the requested event (EVT), performs scheduling on the connection information (ICT), and outputs a first control signal (XCT1). An event (EVT) may mean, for example, that a data packet (DPK) is received from a network or a transmission request for data is transmitted from an application. Alternatively, even if a timeout occurs in the operation of the TCP controller 140, etc., it may be processed as an event (EVT).

The TCP controller 140 may include a first TCP controller 142 and a second TCP controller 144. The first TCP controller 142 and the second TCP controller 144 each receive connection information (ICT) from the event scheduler 120 and control calculation operations for controlling the connection information (ICT). .

Connection information (ICT) may include information used to establish a connection between two end-points, a local node and a remote node, with which one wishes to communicate. Connection information (ICT) may include, for example, connection identifiers for IP addresses and port numbers of local nodes and remote nodes. This connection information (ICT) can be used to create or process a header for each layer of the TCP/IP stack. For each layer of the TCP/IP stack, a TCP packet containing a TCP header, an IP packet containing an IP header, and a MAC packet containing a MAC header may be generated. For example, the header of a TCP packet may include metadata such as a source address, destination address, sequence number, and acknowledgment number.

The event scheduler 120 may also generate event information (IEV) corresponding to connection information (ICT). Event information (IEV) may be information about the type of event (EVT) that occurs between the local node and the remote node indicated by the connection information (ICT). For the examples described above, the event information (IEV) is information about the reception of a data packet (DPK) from a remote node identified by the corresponding connection information (ICT), a data transmission request by the local node, or a timeout of an event (EVT), etc. may include.

Connection information (ICT) for a pair of local nodes and remote nodes may be transmitted to one of the first TCP controller 142 and the second TCP controller 144. For example, connection information (ICT) for a first pair of local nodes and remote nodes is communicated to the first TCP controller 142, and connection information (ICT) for a second pair of local nodes and remote nodes is transmitted to the first TCP controller 142. 2 may be transmitted to the TCP controller 144. Additionally, a plurality of connection information (ICT) may be stored in the first TCP controller 142 and the second TCP controller 144. For example, x pieces of connection information (ICT) may be transmitted to the first TCP controller 142, and y pieces of connection information (ICT) may be transmitted to the second TCP controller 144. The event information (IEV) may be transmitted to the TCP controller through which the corresponding connection information (ICT) is transmitted, among the first TCP controller 142 and the second TCP controller 144.

The memory control logic 160 moves the connection information (ICT) to the first memory (MEM1) or stores it in the second memory (MEM2) in response to the first control signal (XCT1). The memory control logic 160 may transmit the second control signal (XCT2) to the event scheduler 120 as a result of an operation performed in relation to the storage location or scheduling of the connection information (ICT).

The first memory (MEM1) may be SRAM (Static Random Access Memory). The first memory (MEM1) may be included in the first TCP controller 142 and the second TCP controller 144. The second memory (MEM2) may be DRAM (Dynamic Random Access Memory). However, the types of the first memory (MEM1) and the second memory (MEM2) are not limited.

If the event information (IEV) corresponding to the connection information (ICT) stored in the first memory (MEM1) is not updated for the first time, for example, it is identified by the connection information (ICT) stored in the first memory (MEM1). If communication between the local node and the remote node does not occur for a certain period of time, the connection information (ICT) stored in the first memory (MEM1) will be stored in the second memory (MEM2) in response to the first control signal (XCT1). You can.

As such, the network interface device 100 according to an embodiment of the present invention allows the first TCP controller 142 and the second TCP controller 144 to independently perform operations on separate connection information (ICT), thereby achieving high speed. Event processing may be possible. In addition, the network interface device 100 according to an embodiment of the present invention provides connection information (ICT) for a connection in which the waiting time for an event occurrence continues for a certain period of time or more to the first TCP controller 142 and the second TCP controller 144. ), and the first TCP controller 142 or the second TCP controller 144 performs control of the operation corresponding to the new connection information (ICT). By doing so, high-speed event processing may be possible.

3 to 5 are diagrams showing memory control logic 160 according to an embodiment of the present invention, respectively.

First, referring to FIG. 3, the second memory (MEM2) according to an embodiment of the present invention is located outside the first chip (CH1) equipped with the first TCP controller 142 and the second TCP controller 144. You can. There may be at least two second memories (MEM2). At this time, the memory control logic 160 may be provided as a separate chip from the first chip CH1 and the second memory MEM2. For example, the memory control logic 160 is an FPGA (Field Programmable Gate Array) chip different from the first chip (CH1), which is a core chip, on a board on which the first chip (CH1) and the second memory (MEM2) are mounted. It can be provided. The event scheduler 120 is shown as being located outside the first chip (CH1), but this only means that the event scheduler 120 can be located in various ways, and the event scheduler 120 is located outside the first chip (CH1). ) does not mean that it is excluded from being included in.

Next, referring to FIG. 4, the memory control logic 160 according to an embodiment of the present invention may be provided with a second memory (MEM2) and a second chip (CH2) separate from the first chip (CH1). there is. For example, the second chip CH2 may be provided as a Data Process Unit (DPU) chip. The first chip (CH1) and the second chip (CH2) can transmit and receive information and signals through a coherent interconnect. The first chip CH1 and the memory control logic 160 of FIG. 3 can also transmit and receive information and signals through a coherent interconnect.

Or, as shown in FIG. 5, the memory control logic 160 according to an embodiment of the present invention may be provided in the first chip (CH1) including the first TCP controller 142 and the second TCP controller 144. It may be possible. At this time, the memory control logic 160 may transmit and receive information and signals with the second memory MEM2 through a coherent interconnect.

As described above, the connection information (ICT) may be stored in one of the first memory (MEM1) and the second memory (MEM2) based on the status of the connection information (ICT). At this time, the memory control logic 160 according to an embodiment of the present invention determines whether to transmit the connection information (ICT) stored in the second memory (MEM2) to the first memory (MEM1) in response to the first control signal (XCT1). You can make decisions about whether or not to A more detailed description of the operation of the memory control logic 160 is described later.

In this way, the network interface device 100 according to an embodiment of the present invention can improve data processing performance by optimizing the storage location of connection information (ICT) by the memory control logic 160. Furthermore, the network interface device 100 according to an embodiment of the present invention can perform optimized operation in a high-speed network environment by providing a memory control logic 160 to optimize the required resource limits and performance standards. there is.

Figures 6 and 7 are diagrams showing a TOE (TCP/IP Offload Engine)-based network interface device 100, respectively, according to an embodiment of the present invention.

First, referring to FIG. 6, the network interface device 100 according to an embodiment of the present invention may be a NIC (Network Interface Controller), a network adapter, a Smart NIC, etc. that independently processes the Internet standard protocol TCP/IP. In other words, the network interface device 100 according to an embodiment of the present invention may be a TOE protocol-based network interface device. The TOE protocol performs TCP/IP operations in the network interface device 100 rather than in the operating system of the host CPU. By doing so, the network interface device 100 according to an embodiment of the present invention can support a high-speed network environment by being equipped based on the TOE protocol.

To this end, the network interface device 100 according to an embodiment of the present invention includes a first TCP controller 142 and a second TCP controller 144, as well as a first TCP operation logic 182 and a second TCP operation logic 184. ) may further be included. The first TCP operation logic 182 and the second TCP operation logic 184 are stored in the first memory (MEM1) from the corresponding TCP controller among the first TCP controller 142 and the second TCP controller 144, respectively. TCP/IP operations can be performed by receiving connection information (ICT) and event information (IEV).

For example, the first TCP operation logic 182 and the second TCP operation logic 184 perform congestion window control, flow control, retransmission, etc. for the received connection information (ICT) and event information (IEV). , it is possible to perform calculation operations required to process the header of each layer. The first TCP operation logic 182 and the second TCP operation logic 184 may transmit the results of the operation to the first TCP controller 142 and the second TCP controller 144.

Next, referring to FIGS. 2 and 7, the network interface device 100 according to an embodiment of the present invention also includes a network interface 110, a reception processing module 130, a host interface 150, and a transmission processing module ( 170) may further be included.

The network interface 110 includes a Gigabit Ethernet controller and is responsible for the interface with the Gb Ethernet MAC/PHY chip to process transmission and reception of Ethernet packets. The reception processing module 130 may parse a data packet (DPK) received through the network interface 110 into user data and metadata. User data in the data packet (DPK) may be transmitted to the host module through the host interface 150. The host interface 150 may include, for example, a PCIe controller (Peripheral Component Interconnect express controller) and may perform an interface with a host module.

The reception processing module 130 may transmit metadata of the data packet (DPK) to the event scheduler 120. The event scheduler 120 may process metadata received from the reception processing module 130 into an event (EVT). When metadata transmitted from the reception processing module 130 indicates new connection information (ICT), the event scheduler 120 may generate new connection information (ICT). If the metadata transmitted from the reception processing module 130 is related to connection information (ICT) stored in the first memory (MEM1) or the second memory (MEM2), the corresponding connection information (ICT) may be updated.

The transmission processing module 170 may output a data packet (DPK) in response to a data transmission request received through the host interface 150. A data transmission request may be transmitted from an application running on the host module. A data transmission request may be transmitted directly to the host interface 150 without going through the TCP/IP protocol stack of the host operating system. This can be performed through API hooking operations for the socket API (socket application programming interface), such as the application's send() and recv().

The transmission processing module 170 may perform an operation of combining the TCP header, IP header, and MAC header corresponding to the data transmission request with data (DTA) to be transmitted to the remote node. The TCP header, IP header, and MAC header may be generated by the first TCP controller 142 and the first TCP operation logic 182, or the second TCP controller 144 and the second TCP operation logic 184. Data (DTA) to be transmitted to the remote node can be read from the main memory of the host module.

According to the transmission command of the TCP controller that processes the connection information (ICT) corresponding to the data transmission request among the first TCP controller 142 and the second TCP controller 144, the data packet (DPK) is sent to the transmission processing module 170. It may be transmitted to the network interface 110 and output.

Although not shown in FIG. 7, the network interface device 100 according to an embodiment of the present invention processes a data packet (DPK) to be processed by the reception processing module 130 or data processed by the transmission processing module 170. User data that is to be temporarily stored as a packet (DPK), processed by the reception processing module 130, and output through the host interface 150, or user data received through the host interface 150 in response to a data transmission request. A TCP buffer that temporarily stores may be further included. Alternatively, all or part of these TCP buffers may be provided in all or part of the second memory (MEM2).

As such, the network interface device 100 according to an embodiment of the present invention operates as shown in FIG. 6 or FIG. 7 and can improve data processing speed by performing TCP/IP operations on its own. At this time, if the connection information (ICT) is stored in the second memory (MEM2), the above operations can be performed after the connection information (ICT) is transferred from the second memory (MEM2) to the first memory (MEM1). there is. That is, the first TCP controller 142 and the second TCP controller 144 can control connections between more local nodes and remote nodes and operate adaptively to the occurrence of an event (EVT). Therefore, according to the network interface device 100 according to an embodiment of the present invention, high-speed networking can be efficiently supported.

6 and 7 show an example in which the first TCP operation logic 182 and the second TCP operation logic 184 correspond one-to-one to the first TCP controller 142 and the second TCP controller 144. That is, the first TCP operation logic 182 processes the connection information (ICT) and event information (IEV) of the first TCP controller 142, and the second TCP operation logic 184 processes the connection information (ICT) and event information (IEV) of the first TCP controller 142. It can process connection information (ICT) and event information (IEV). However, it is not limited to this.

Figures 8A to 8C are diagrams showing the relationship between the TCP controller 140 and the TCP operation logic 180, respectively, according to an embodiment of the present invention.

First, referring to FIGS. 1 and 8A, the network interface device 100 according to an embodiment of the present invention may include n TCP controllers 140 and m TCP operation logic 180. Each of the n TCP controllers 140 may control calculation operations for connection information (ICT) and event information (IEV) in response to the first control signal (XCT1). The m TCP operation logic 180 may perform TCP/IP operation operations based on connection information (ICT) and event information (IEV) received from a corresponding TCP controller among the n TCP controllers 140.

Connection information (ICT) and event information (IEV) may be transferred from the first memory (MEM1) to the TCP operation logic 180. If the connection information (ICT) is stored in the second memory (MEM2), it is moved from the second memory (MEM2) to the first memory (MEM1) in response to the first control signal (XCT1), and then the TCP operation logic 180 ) can be output.

At this time, n and m may be the same. That is, the TCP controller 140 and TCP operation logic 180 may be provided in the same number. Additionally, the TCP controller 140 and the TCP operation logic 180 may have a one-to-one correspondence with each other. For example, the first TCP operation logic 182 performs an operation under the control of the first TCP controller 142, and the mth TCP operation logic 180m performs an operation under the control of the nth TCP controller 140n. Computational operations may be performed. Accordingly, the network interface device 100 according to an embodiment of the present invention can simultaneously perform parallel processing for connections of n remote nodes and local loads without other resource limitations.

Next, referring to FIGS. 1 and 8B, the network interface device 100 according to an embodiment of the present invention may include n TCP controllers 140 and m TCP operation logic 180. Similar to Figure 8A, n and m may be the same. That is, the TCP controller 140 and TCP operation logic 180 may be provided in the same number. However, n TCP controllers 140 in FIG. 8B may share m TCP operation logic 180.

For example, an operation operation is performed in one of the first TCP operation logic 182 to the mth TCP operation logic 180m under the control of the first TCP controller 142, and the nth TCP controller ( An operation may be performed in one of the first TCP operation logic 182 to the mth TCP operation logic 180m according to the control of 140n). However, it is not limited to this. Some of the n TCP controllers 140 may share some of the m TCP operation logic 180, and others among the n TCP controllers 140 may share the remainder of the m TCP operation logic 180.

The network interface device 100 according to an embodiment of the present invention is equipped as shown in FIG. 8B, so that not only can simultaneous parallel processing be performed on the connections of n remote nodes and local loads while optimal scheduling is performed, but also m Among the TCP calculation logics 180, a new event (EVT) can be processed through the calculation logic for which the calculation operation has been completed. Accordingly, the processing performance of the network interface device 100 according to an embodiment of the present invention can be improved.

Next, referring to FIGS. 1 and 8C, the network interface device 100 according to an embodiment of the present invention may include n TCP controllers 140 and m TCP operation logic 180. At this time, n and m may be different. That is, the TCP controller 140 and TCP operation logic 180 may be provided in different numbers. At this time, at least one TCP controller among the n TCP controllers 140 is shared by at least two TCP operation logics among the m TCP operation logics 180, or at least two TCP controllers among the n TCP controllers 140 are m. At least one TCP operation logic among the TCP operation logics 180 may be shared.

8C shows, as an example, an operation operation is performed in the first TCP operation logic 182 and the second TCP operation logic 184 under the control of the first TCP controller 142, and the n-1 TCP controller 140n- 1) and shows an example in which an operation is performed in the mth TCP operation logic 180m under the control of the nth TCP controller 140n. Accordingly, the network interface device 100 according to an embodiment of the present invention can improve its processing performance by performing optimized scheduling under resource constraints.

In FIGS. 8A to 8C, it is assumed whether the number of TCP controllers 140 and TCP operation logic 180 is the same, but the TCP controller 140 and TCP operation logic 180 are shown regardless of whether the numbers are the same. 8A to 8C or other connection relationships may be established.

FIG. 9 is a flowchart showing a method 900 of operating a network interface device according to an embodiment of the present invention, and FIG. 10 is a diagram showing a network interface device 100 operating by the operating method 900 of FIG. 9 .

9 and 10, the method 900 of operating the network interface device 100 according to an embodiment of the present invention stores connection information (ICT) in the first memory (MEM1) of the TCP controller 140. A step (S920), a step of checking the occurrence of an event (EVT) corresponding to the connection information (ICT) stored in the first memory (MEM1) (S940), and an event (EVT) corresponding to the connection information (ICT) is generated. Data processing in a high-speed network can be accelerated, including the step (S960) of moving the connection information (ICT) stored in the first memory (MEM1) to the second memory (MEM2) if it does not occur for 1 hour. .

As described above, connection information (ICT) is information that the event scheduler 120 generates for a received event (EVT). The first memory (MEM1) may be the SRAM of the TCP controller 140. An operation of writing connection information (ICT) into the second memory (MEM2) may be performed by the memory control logic 160. The second memory MEM2 may be a DRAM located outside the chip on which the TCP controller 140 is installed. A plurality of second memories MEM2 are provided, and a memory into which connection information ICT is to be written may be selected by the memory control logic 160 among the plurality of second memories MEM2.

Accordingly, the network interface device 100 and its operating method 900 according to an embodiment of the present invention allow the limited TCP controller 140 to control connections between more local nodes and remote nodes, thereby improving performance. It can be.

FIGS. 11A and 11B are diagrams illustrating in more detail the method 900 of operating the network interface device of FIG. 9 , respectively.

First, referring to FIGS. 10 and 11A, in the method 900 of operating the network interface device 100 according to an embodiment of the present invention, when the event scheduler 120 receives an event (EVT) (S912), the event ( Connection information (ICT) corresponding to EVT can be created or already created connection information (ICT) can be updated (S914). The event scheduler 120 may include a mapping table (MTB) that stores the storage location of connection information (ICT).

If information regarding the newly created connection information (ICT) does not exist in the mapping table (MTB), the event scheduler 120 updates the mapping table (MTB) (S916) and sends the connection information (ICT) to the TCP controller. It can be delivered to (140) (S918). The TCP controller 140 may store connection information (ICT) in the first memory (MEM1) (S920).

The event scheduler 120 may periodically check the occurrence of an event (EVT) corresponding to the connection information (ICT) (S940). Multiple events (EVT) may occur continuously or discontinuously for connection information (ICT), which is an identifier for the connection between a paired remote node and a local node. If the event (EVT) corresponding to the connection information (ICT) does not occur during the first time (t1), the event scheduler 120 controls the operation related to the connection information (ICT) by using the first control signal (XCT1). It can be transmitted to the TCP controller 140 and memory control logic 160 (S942).

The TCP controller 140 may transmit the connection information (ICT) stored in the first memory (MEM1) to the memory control logic 160 in response to the first control signal (XCT1) (S944). The memory control logic 160 may write connection information (ICT) received from the TCP controller 140 to the second memory (MEM2) in response to the first control signal (XCT1) (S960).

The memory control logic 160 may transmit the second control signal (XCT2) indicating that the connection information (ICT) is stored in the second memory (MEM2) to the event scheduler 120 (S962). The event scheduler 120 may update the mapping table (MTB) in response to the second control signal (XCT2) (S964).

Next, referring to FIGS. 1 and 11B, the method 900 of operating the network interface device 100 according to an embodiment of the present invention includes an event scheduler ( Before 120) transmits the connection information (ICT), it is possible to select which TCP controller among the plurality of

TCP controllers

142 and 144 to transmit the connection information (ICT) (S915). The event scheduler 120 may select a TCP controller based on the number of connection information (ICT) controlled by each of the

multiple TCP controllers

142 and 144. For example, if the first TCP controller 142 stores two connection information (ICT) and the second TCP controller 144 stores five connection information (ICT), the event scheduler 120 newly The generated connection information (ICT) may be transmitted to the first TCP controller 142 (S918).

Figures 12A and 12B are diagrams for explaining an operation of retransmitting connection information (ICT) to the first memory (MEM1) according to an embodiment of the present invention, respectively.

First, referring to FIGS. 10 and 12A, when an event (EVT) is received by the event scheduler 120 (S912), the event scheduler 120 may search the mapping table (MTB) (S913). As a result, if the event (EVT) corresponds to the connection information (ICT) stored in the second memory (MEM2), the event scheduler 120 transmits the first control signal (XCT1) to the memory control logic 160. (S915). The signal value of the first control signal

The memory control logic 160 may determine whether to transmit the connection information (ICT) stored in the second memory (MEM2) to the first memory (MEM1) in response to the first control signal (XCT1) ( S970). For example, if the probability that an event (EVT) for connection information (ICT) stored in the second memory (MEM2) is performed within a second time is less than or equal to a certain value, the memory control logic 160) may maintain a standby state without transmitting to the first memory (MEM1). On the other hand, when the probability that an event (EVT) for the connection information (ICT) stored in the second memory (MEM2) is performed within the second time is higher than a certain value, the memory control logic 160 controls the connection information (ICT) can be moved to the first memory (MEM1) (S980).

The memory control logic 160 may transmit a second control signal (XCT2) indicating whether the connection information (ICT) has been moved to the first memory (MEM1) to the event scheduler 120 (S982). The event scheduler 120 may update the mapping table (MTB) with connection information (ICT) in response to the second control signal (XCT2) (S984).

Next, referring to FIGS. 1 and 12B, the method 900 of operating the network interface device 100 according to an embodiment of the present invention is as shown in FIGS. 12A and 12B when two or

more TCP controllers

142 and 144 are provided. Likewise, before the memory control logic 160 transfers the connection information (ICT) stored in the second memory (MEM2) to the first memory (MEM1) (S980), the event scheduler 120 configures a plurality of first memories ( It is possible to determine which of MEM1) to which the connection information (ICT) will be re-moved to the first memory of the TCP controllers 142 and 144 (S990).

Specifically, the memory control logic 160 determines whether to transmit the connection information (ICT) stored in the second memory (MEM2) to the first memory (MEM1) (S970) and sends the result to the event scheduler 120. It can be transmitted as a second control signal (XCT2) (S982). When the second control signal By searching the mapping table (MTB), a TCP controller, for example, the first TCP controller 142, that can store connection information (ICT) in the first memory (MEM1) can be selected (S990). The following assumes that the first TCP controller 142 is selected.

At this time, if the first memory (MEM1) of all

TCP controllers

142 and 144 cannot store the connection information (ICT), the event scheduler 120 stores the first memory (MEM1) of one of the plurality of

TCP controllers

142 and 144. TCP controller 142 can be selected. The event scheduler 120 may select a TCP controller based on a Least Recently Used (LRU) strategy. The event scheduler 120 may transmit the first control signal (XCT1) to the selected first TCP controller 142 (S992).

The first TCP controller 142 that receives the first control signal (XCT1) may transfer one of the connection information (ICT) stored in the first memory (MEM1) to the second memory (MEM2) (S994) ). The first TCP controller 142 may select connection information (ICT) transmitted to the second memory (MEM2) from among the stored connection information (ICT) based on the LRU strategy, etc.

The memory control logic 160 transmits the connection information (ICT) stored in the second memory (MEM2) to the first TCP controller 142, which delivered the connection information (ICT) stored in the first memory (MEM1). (S996). The memory control logic 160 transmits a second control signal (XCT2) indicating that the connection information (ICT) stored in the second memory (MEM2) is stored in the first memory (MEM1) to the event scheduler 120 ( S998), the event scheduler 120 may update the mapping table (MTB) in response to the second control signal (XCT2) (S999).

As such, according to the network interface device 100 and its operating method 900 according to an embodiment of the present invention, the storage location of the connection information (ICT) is determined through scheduling of the event scheduler 120 and the memory control logic 160. Data processing performance can be improved by optimizing and at the same time efficiently moving it back to the first memory (MEM1).

Figure 13 is a diagram showing a network interface device 100 according to an embodiment of the present invention.

Referring to FIG. 13, the network interface device 100 according to an embodiment of the present invention, unlike the embodiment described above, allows the event scheduler 120 to transmit connection information (ICT) to the memory control logic 160. You can. For example, when multiple connection pairs to a remote node and a local node are created at the same time or within close proximity, the event scheduler 120 first configures memory control logic (ICT) to store connection information (ICT) in the second memory (MEM2). 160).

In this case, connection information (ICT) stored in the second memory (MEM2) may be processed according to the operation of FIG. 12A or FIG. 12B. In this way, according to the network interface device 100 according to an embodiment of the present invention, as many connections as possible can be processed based on a certain resource.

Figure 14 is a diagram showing a server device 1400 according to an embodiment of the present invention.

Referring to FIG. 14, the server device 1400 according to an embodiment of the present invention includes a network interface module 1420 and a host module 1440. The server device 1400 according to an embodiment of the present invention may be one of various types of web servers, such as an asynchronous web server, a thread-based web server, and a forward proxy or reverse proxy-based web server. You can. Alternatively, the server device 1400 according to an embodiment of the present invention may be one of various types of servers such as a web application server (WAS), a storage server, and a database server.

The network interface module 1420 includes a TCP/IP hardware stack 1422 and a TCP/IP software stack 1424 that perform TCP/IP operations on data packets (DPKs). The TCP/IP hardware stack 1422 may include a network interface device 100 such as that shown in FIG. 1 or FIG. 10 .

The TCP/IP software stack 1424 is not shown in FIG. 1, etc., but as explained in FIG. 7, the TCP/IP software stack 1424 is used in the network interface module 1420 without going through the TCP/IP protocol stack of the operating system of the host module 1440. Communication can be performed between the TCP/IP hardware stack 1422 and the host module 1440 to perform IP operations. For example, when the host module 1440 generates a data transmission request, the TCP/IP software stack 1444 performs an API hooking operation for the socket API, such as send() and recv() of the application of the host module 1440. can be performed.

Various embodiments of the network interface device 100 described above may be applied to the network interface module 1420 of FIG. 14. For example, the network interface module 1420 may include at least two

TCP controllers

142 and 144 as shown in FIG. 1 . Alternatively, the network interface module 1420 may be provided as a DPU chip together with the second memory (MEM2) as shown in FIG. 4.

According to the server device 1400 according to an embodiment of the present invention, the network interface module 1420 performs TCP/IP operation to reduce the load on the CPU of the host module 1440 and provides connection information for requested events. By storing data in one of the first memory (MEM1) and the second memory and performing optimized scheduling, data processing in a high-speed network can be accelerated. In addition, according to the server device 1440 according to an embodiment of the present invention, it includes a plurality of TCP controllers and can perform processing for connection pairs of remote nodes and local loads in parallel, so the server device 1440 Data processing performance can be improved.

Although representative embodiments of the present invention have been described in detail above, those skilled in the art will understand that various modifications can be made to the above-described embodiments without departing from the scope of the present invention. . For example, although not described, the network interface device 100 and the server device 1400 according to an embodiment of the present invention use other protocols such as Internet Control Message Protocol (ICMP) and Address Resolution Protocol (ARP) in addition to TCP/IP. A separate module to support a network protocol may be further included, and link layer routing, etc. may be performed through the corresponding processing module. Therefore, the scope of the present invention should not be limited to the described embodiments, but should be determined by the claims and equivalents of the claims as well as the claims described later.

Claims

an event scheduler that generates connection information for a requested event, performs scheduling on the connection information, and outputs a first control signal;

a first TCP controller and a second TCP controller, respectively, receiving the connection information from the event scheduler and controlling calculation operations on the connection information; and

A network interface device comprising: memory control logic that transmits the connection information to a first memory or stores it in a second memory in response to the first control signal.
In claim 1,

The event scheduler is,

A network interface device that transmits the connection information to at least one of the first TCP controller, the second TCP controller, and the memory control logic.
In claim 1,

The first memory is,

A network interface device included in the first TCP controller and the second TCP controller.
In claim 1,

The second memory is,

A network interface device including a dynamic random access memory (DRAM) located outside a first chip equipped with the first TCP controller and the second TCP controller.
In claim 4,

The memory control logic is,

A network interface device provided on a second chip together with the second memory.
In claim 1,

The memory control logic is,

A network interface device provided on the same first chip as the first TCP controller and the second TCP controller.
In claim 1,

The connection information stored in the first memory is,

If a corresponding event does not occur for a first time, the network interface device is moved to the second memory.
In claim 1,

The memory control logic is,

A network interface device that generates a second control signal indicating whether to move the connection information to the first memory when an event regarding the connection information stored in the second memory occurs.
In claim 8,

The connection information stored in the second memory is,

A network interface device that is moved to a first memory of one of the first memory of the first TCP controller and the first memory of the second TCP controller.
In claim 1,

A first TCP operation that performs an operation by receiving the connection information and event information corresponding to the connection information stored in a first memory from a corresponding TCP controller among the first TCP controller and the second TCP controller, respectively. A network interface device further comprising: logic and second TCP operation logic.
In claim 10,

third to nth TCP controllers, respectively, controlling calculation operations on the connection information and the event information in response to the first control signal; and

First TCP operation logic to m-th TCP operation logic that performs an operation by receiving the connection information and the event information stored in the first memory included from the corresponding TCP controller among the third TCP controller to the n-th TCP controller, respectively. A network interface device further comprising logic.
In claim 11,

The n and the m are different network interface devices.
In claim 1,

The event scheduler is,

A network interface device including a mapping table for a storage location of the connection information.
In claim 1,

A reception processing module that parses a data packet received through a network interface, outputs user data of the data packet through a host interface, and transmits metadata of the data packet as an event to the event scheduler. ; and

A transmission processing module that generates a header corresponding to the connection information in data received through the host interface and outputs the header as a data packet.
In the method of operating a TOE (TCP/IP Offload Engine)-based network interface device,

storing connection information corresponding to the event in a first memory of the TCP controller;

Checking occurrence of an event corresponding to connection information stored in the first memory; and

If an event corresponding to the connection information does not occur for a first time, moving the connection information stored in the first memory to a second memory.
In claim 15,

The step of storing the connection information in the first memory includes:

selecting one of the plurality of TCP controllers; and

A method of operating a network interface device comprising: storing the connection information in a first memory of the selected TCP controller.
In claim 15,

When an event corresponding to the connection information moved to the second memory occurs, determining whether to transmit the connection information moved to the second memory to the first memory;

selecting one of the plurality of TCP controllers; and

A method of operating a network interface device further comprising transmitting the connection information moved to the second memory to the first memory of the selected TCP controller.
A network interface module including a TCP/IP hardware stack and a TCP/IP software stack that performs TCP/IP operations on received data packets; and

A host module that receives and processes the data of the data packet from the network interface module,

The network interface module is,

A server device that stores connection information corresponding to the data packet in a first memory of a TCP controller or in a second memory controlled by memory control logic.
In claim 18,

The network interface module is,

A server device comprising a plurality of the TCP controllers.
In claim 18,

The memory control logic is,

A server device provided on the same chip as the second memory.