JP2019033351A - Packet processor and packet processing method - Google Patents

Abstract

To improve processing efficiency of a packet, by measuring the band width of packet flow, and changing the allocation capacity of a lookup table dynamically.SOLUTION: A packet processor for applying at least one function, out of predetermined multiple functions, to a received packet by using a CPU and an accelerator has a receiving part for receiving a packet, a band width measurement function part for measuring the band width of the flow of received packets for each flow and each function to be applied, an allocation calculation function part for calculating the allocation capacity of the lookup table, registering the flows to be processed on the accelerator for each function, based on the measured band width of flow, and a lookup table capacity allocation change function part for changing the allocation capacity of the lookup table, based on the calculated allocation capacity of lookup table.SELECTED DRAWING: Figure 2

Description

  The present invention relates to a packet processing device and a packet processing method, and more particularly to a packet processing device and a packet processing method that perform processing using a server and an accelerator.

  Network virtualization technology (NFV: Network Function Virtualization) that realizes higher scalability economically by virtualizing computer resources and network resources in carrier networks is being studied. In NFV, network functions and service functions can be softwareized and run on a general-purpose server.

  These network functions and service functions are normally processed by a CPU (Central Processing Unit) of the server, but a method for realizing high-speed processing using an accelerator in addition to the CPU has been proposed. For example, it has been proposed to improve performance by using an accelerator such as an FPGA (Field Programmable Gate Array) in combination with a CPU (see Non-Patent Documents 1 to 3).

Zvika Bronstein, et. Al., "Uniform Handling and Abstraction of NFV Hardware Accelerators", Internet, http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=7113221&tag=1, acquired July 7, 2017 Koji Yamazaki, et. Al., "Accelerating SDN / NFV with transparent offloading architecture", Internet, https://www.usenix.org/system/files/conference/ons2014/ons2014-paper-yamazaki.pdf, 2017/7 Acquired on 7th of May Leonhard Nobach, "Open, Elastic Provisioning of Hardware Acceleration in NFV Environments", Internet, http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=7089057, acquired July 7, 2017

  FIG. 1 is a configuration example of a device that implements various processing functions such as a network function and a service function using an accelerator in the prior art.

  In the apparatus shown in FIG. 1, an accelerator such as an FPGA is used in combination with a CPU. In the CPU, various functions can be realized as necessary, and a plurality of containers for providing an independent execution environment is set for each function. Also in the accelerator, various functions can be provided as necessary. Whether the received packet flow is processed in the container or on the accelerator is defined in a lookup table on the accelerator. In this way, the CPU load can be reduced by cut-through in which some processing is performed on the accelerator.

  When a CPU and an accelerator have a plurality of functions divided, the number of CPU cores allocated to each function and the allocation capacity of the lookup table are fixedly determined. For example, when a DPDK (Data Plane Development Kit) is used to realize a network function, a device driver called PMD (Poll Mode Driver) is provided. This driver is designed on the premise that it basically occupies and operates the CPU core, and acquires packets from the CPU core to the NIC (Network Interface Card) at a fine cycle.

  For example, when one CPU core is allocated to the container A that provides the function A, one CPU core is used even if the number of packets that require the function A increases or decreases. In addition, when the lookup table C is defined so that packets to a specific server are processed by the function C on the accelerator, if there are few packets to the specific server, resources on the accelerator are left over, The allocated capacity of the lookup table C is not used effectively.

  In this way, there is a temporal fluctuation in the bandwidth of the packet flow. If the number of CPU cores allocated and the allocation capacity of the lookup table are fixedly determined, the packet processing efficiency may decrease. is there.

  An object of the present invention is to improve packet processing efficiency by measuring the bandwidth of a packet flow and dynamically changing the allocation capacity of a lookup table.

A packet processing apparatus according to an aspect of the present invention is provided.
A packet processing device that applies at least one of a plurality of predetermined functions to a received packet using a CPU and an accelerator,
A receiver for receiving the packet;
A bandwidth measurement function unit that measures the bandwidth of the flow of the received packet for each flow and for each applied function;
An allocation calculation function unit that calculates an allocation capacity of a lookup table that registers, for each function, a flow to be processed on the accelerator based on the measured flow bandwidth;
A lookup table capacity allocation change function unit for changing the allocation capacity of the lookup table based on the calculated allocation capacity of the lookup table;
It is characterized by having.

A packet processing method according to an aspect of the present invention includes:
A packet processing method in a packet processing apparatus that applies at least one of a plurality of predetermined functions to a received packet using a CPU and an accelerator,
Receiving a packet;
Measuring the bandwidth of the received packet flow for each flow and for each applied function;
Calculating an allocation capacity of a lookup table for registering a flow to be processed on the accelerator for each function based on the bandwidth of the measured flow;
Changing the allocation capacity of the lookup table based on the calculated allocation capacity of the lookup table;
It is characterized by having.

  According to the present invention, it is possible to improve the packet processing efficiency by measuring the bandwidth of the packet flow and dynamically changing the allocated capacity of the lookup table.

It is an example of a structure which implement | achieves various processing functions using an accelerator in a prior art. It is a figure which shows the structure of the packet processing apparatus which concerns on the Example of this invention. It is a sequence diagram which shows the process in the packet processing apparatus concerning the Example of this invention. It is a figure which shows the structure of the packet processing apparatus which concerns on another Example of this invention.

  Embodiments of the present invention will be described below with reference to the drawings.

  FIG. 2 shows the configuration of a packet processing apparatus according to an embodiment of the present invention. The packet processing device is a device that applies at least one of a plurality of predetermined functions to a received packet using a CPU and an accelerator, and may be a transmission device such as an edge router, for example.

  In addition to the apparatus shown in FIG. 1, the packet processing apparatus shown in FIG. 2 has a function / flow band measurement function unit 101, a band distribution estimation function unit 103, an optimum allocation calculation function unit 105, and a lookup table capacity allocation. A change function unit 107 and a core allocation change function unit 109 are included. For example, the function / flow-by-flow bandwidth measurement function unit 101 and the lookup table capacity allocation change function unit 107 are realized on an accelerator, and the band distribution estimation function unit 103, the optimum allocation calculation function unit 105, and the core allocation change function unit 109 Realized on the CPU. Note that each of the functional units 101 to 109 may be realized by either an accelerator or a CPU. For example, each functional unit of the packet processing device may be realized by an accelerator or CPU executing data or a program stored in a storage device or a memory device (not shown).

  As described with reference to FIG. 1, in the packet processing apparatus, an accelerator such as an FPGA is used in combination with the CPU. The CPU can implement various functions (Function A, Function B, Function C), and multiple containers (Container A, Container B, Container C) that provide an independent execution environment for each function are set. . The container may be realized by a virtual machine. The accelerator can also provide various functions (function A, function B, function C) as necessary. Examples of function A, function B, and function C include NAPT (Network Address Port Translation), VLAN (Virtual Local Area Network) encapsulation, and encryption.

  Whether the packet flow received by the packet processing device is processed in the container or on the accelerator is defined in the lookup table (lookup table A, lookup table B, lookup table C) on the accelerator Is done. The lookup table registers a flow to be processed on the accelerator for each function. For example, a packet having an identifier registered in the lookup table A is processed by the function unit A set on the accelerator, and a packet having an identifier not registered in the lookup table A is stored in the CPU resource in the container A. Is processed using a CPU core that is logically divided. For example, 5tuple (destination IP address, source IP address, destination port, source port, protocol number) of the packet may be used as an identifier, and functions such as NAPT, VLAN encapsulation, and encryption are applied on the accelerator Whether or not to do so may be determined by the 5tuple of the packet. A plurality of functions may be applied to one packet.

  In the embodiment of the present invention, if the number of CPU cores to be allocated and the allocation capacity of the lookup table are fixedly determined, the accelerator is caused by temporal fluctuations in the packet flow bandwidth (for example, bandwidth for every 5 tuple flows). Since the above resources may be left over, the number of CPU cores allocated and the allocation capacity of the lookup table are dynamically changed.

  The function / flow band measurement function unit 101 measures the bandwidth of the flow of the packet received by the interface unit (I / F) for each flow and for each applied function.

  The band distribution estimation function unit 103 estimates a flow band distribution based on the flow band measured by the function / flow-by-flow band measurement function unit 101.

  The optimal allocation calculation function unit 105 calculates the allocation capacity of the lookup table based on the flow band measured by the function / flow-based band measurement function unit 101 or the flow band distribution estimated by the band distribution estimation function unit 103. calculate. The allocation capacity of the lookup table indicates the amount of resources allocated to the lookup table on the accelerator. When the lookup table is stored in a memory such as SRAM (Static Random Access Memory), the allocated capacity of the lookup table corresponds to the amount of memory resources that the lookup table can use in the memory. By changing the allocated capacity of the lookup table, it is possible to control the amount of offload target packets defined by the lookup table. For example, increasing the allocation capacity of the lookup table C increases the possibility that a packet that requires the function C corresponding to the lookup table C will be processed on the accelerator. On the other hand, since the lookup tables A, B and C share resources on the same accelerator, as a result of increasing the allocation capacity of the lookup table C, the allocation capacity of the lookup tables A and B is reduced. Packets that require functions A and B corresponding to lookup tables A and B are more likely to be processed in containers A and B.

  The optimum allocation calculation function unit 105 calculates the number of CPU cores allocated to each container based on the flow bandwidth measured by the function / flow-based bandwidth measurement function unit 101. The optimal allocation calculation function unit 105 estimates the bandwidth of traffic flowing into the container, increases the allocation of CPU cores to containers that require a large amount of CPU processing, and for containers that do not require a large amount of CPU processing Reduce CPU core allocation.

  The lookup table capacity allocation change function unit 107 changes the allocation capacity of the lookup table based on the allocation capacity of the lookup table calculated by the optimum allocation calculation function unit 105. Further, the lookup table capacity allocation change function unit 107 preferentially processes the flow having the larger bandwidth on the accelerator by registering the flow identifier in the lookup table in order from the flow having the larger bandwidth, for example. Change the lookup table to be

  The core assignment change function unit 109 changes the CPU core assignment based on the number of CPU core assignments calculated by the optimum assignment calculation function unit 105.

  FIG. 3 shows processing in the packet processing apparatus according to the embodiment of the present invention. The process shown in FIG. 3 may be repeated at a predetermined cycle, or may be repeated when a predetermined amount of packets is received.

  First, the bandwidth measurement function unit 101 for each function / flow measures the bandwidth of the received packet flow for each flow and for each applied function, and outputs the measured bandwidth to the bandwidth distribution estimation function unit 103 (S101). ).

  The band distribution estimation function unit 103 estimates the band distribution from the band information for each flow and for each function (S103), and outputs the band distribution to the optimum allocation calculation function unit 105 (S105).

  The optimal allocation calculation function unit 105 calculates the allocation number of the CPU core and the allocation capacity of the lookup table that maximize the packet processing efficiency from the bandwidth distribution (S107). For example, the optimum allocation calculation function unit 105 preferentially allocates the capacity of the lookup table to a function including many flows having a large bandwidth. Furthermore, since the amount of traffic flowing into the container changes by changing the allocation capacity of the lookup table, the optimal allocation calculation function unit 105 estimates the bandwidth of traffic flowing into the container and determines the number of CPU cores allocated. calculate.

  The optimal allocation calculation function unit 105 instructs the core allocation change function unit 109 to change the core allocation, and the core allocation change function unit 109 changes the CPU core allocation (S109).

  The optimal allocation calculation function unit 105 instructs the lookup table capacity allocation change function unit 107 to change the allocation capacity of the lookup table, and the lookup table capacity allocation change function unit 107 determines the allocation capacity of the lookup table. Change (S111). Also, the lookup table capacity allocation change function unit 107 changes the lookup table so that the flow with the larger bandwidth is preferentially processed on the accelerator.

  In the above embodiment, the container and the function on the accelerator have a one-to-one relationship. However, the container and the function on the accelerator are not limited to a one-to-one relationship, but are one-to-many and many-to-one. Or a many-to-many relationship.

  FIG. 4 shows the configuration of a packet processing apparatus according to another embodiment of the present invention. FIG. 4 shows an example in which the container A1 and the container A2 are used to provide the function A, and the container B1 and the container B2 are used to provide the function B.

  The packet processing device shown in FIG. 4 further includes a container inflow traffic band measurement function unit 111 in addition to the packet processing device shown in FIG.

  The container inflow traffic band measurement function unit 111 measures the band of traffic flowing into each container.

  The optimum allocation calculation function unit 105 calculates the number of CPU cores allocated to each container based on the traffic band measured by the container inflow traffic band measurement function unit 111.

  Thus, by measuring the traffic band for each container, it is possible to more appropriately allocate CPU cores when the container and the function on the accelerator are not in a one-to-one relationship.

<Effect of the embodiment of the present invention>
In general, the change of the cut-through traffic amount with respect to the change of the allocated capacity of the lookup table depends on the bandwidth for each flow, the bandwidth of the flow is not uniform in each flow, and the distribution is unknown. Therefore, if the number of CPU cores allocated and the allocation capacity of the lookup table are fixedly determined, the packet processing efficiency decreases.

  According to the embodiment of the present invention, it is possible to improve packet processing efficiency by measuring the bandwidth of a packet flow and dynamically changing the number of CPU cores allocated and the allocation capacity of a lookup table. Become. As a result, when the packet processing apparatus according to the embodiment of the present invention is used in a transmission apparatus such as an edge router, the packet transfer efficiency can be improved.

  Even when the container and the function on the accelerator are not in a one-to-one relationship, the CPU core can be allocated more appropriately by measuring the traffic bandwidth for each container.

<Supplement>
For convenience of explanation, the packet processing device according to the embodiment of the present invention is described using a functional block diagram. However, the packet processing device according to the embodiment of the present invention may be hardware, software, or a combination thereof. It may be realized. For example, the embodiment of the present invention includes a program that causes a computer to realize the functions of the packet processing device according to the embodiment of the present invention, a program that causes a computer to execute each procedure of the method according to the embodiment of the present invention, and the like. May be realized. In addition, the functional units may be used in combination as necessary. In addition, the method according to the embodiment of the present invention may be performed in an order different from the order shown in the embodiment.

  As described above, the method for improving the packet processing efficiency by measuring the bandwidth of the packet flow and dynamically changing the allocation capacity of the lookup table has been described. However, the present invention is not limited to the above embodiment. Without being limited, various modifications and applications are possible within the scope of the claims.

101 Bandwidth measurement function unit for each function / flow 103 Band distribution estimation function unit 105 Optimal allocation calculation function unit 107 Lookup table capacity allocation change function unit 109 Core allocation change function unit 111 Container inflow traffic band measurement function unit

Claims (6)

A packet processing device that applies at least one of a plurality of predetermined functions to a received packet using a CPU and an accelerator,
A receiver for receiving the packet;
A bandwidth measurement function unit that measures the bandwidth of the flow of the received packet for each flow and for each applied function;
An allocation calculation function unit that calculates an allocation capacity of a lookup table that registers, for each function, a flow to be processed on the accelerator based on the measured flow bandwidth;
A lookup table capacity allocation change function unit for changing the allocation capacity of the lookup table based on the calculated allocation capacity of the lookup table;
A packet processing apparatus. When a flow that is not registered in the lookup table is processed in at least one container among a plurality of containers set in the packet processing device,
The allocation calculation function unit calculates the number of CPU cores allocated to each container based on the measured flow bandwidth,
The packet processing device according to claim 1, further comprising a core assignment changing function unit that changes the assignment of CPU cores based on the calculated number of assigned CPU cores. When a flow that is not registered in the lookup table is processed in at least one container among a plurality of containers set in the packet processing device,
The packet processing apparatus further includes a container inflow traffic band measurement function unit that measures a band of traffic flowing into each container,
The allocation calculation function unit calculates the number of CPU cores allocated to each container based on the measured traffic bandwidth,
The packet processing device according to claim 1, further comprising a core assignment changing function unit that changes the assignment of CPU cores based on the calculated number of assigned CPU cores. A band distribution estimation function unit for estimating a flow band distribution based on the measured flow band;
4. The packet processing device according to claim 1, wherein the allocation calculation function unit calculates an allocation capacity of a lookup table based on the estimated bandwidth distribution of the flow. 5.   The lookup table capacity allocation change function unit changes the lookup table so that a flow having a larger bandwidth is preferentially processed on the accelerator. Packet processing device. A packet processing method in a packet processing apparatus that applies at least one of a plurality of predetermined functions to a received packet using a CPU and an accelerator,
Receiving a packet;
Measuring the bandwidth of the received packet flow for each flow and for each applied function;
Calculating an allocation capacity of a lookup table for registering a flow to be processed on the accelerator for each function based on the bandwidth of the measured flow;
Changing the allocation capacity of the lookup table based on the calculated allocation capacity of the lookup table;
A packet processing method.

Images