A Large-Capacity Accurate Packet Counting Method for 100G Interface

Abstract

A large-capacity accurate packet counting method for a 100G interface is characterized in that it includes the following steps: A) allowing the external memory to store high-bit statistics for table entries; a small amount of storage space is built in the logic device, and each table The low bit of the number of packets of the item and the number of bytes is used for statistics; B) after statistical preprocessing, the data is split into a plurality of sub-modules for operation, a kind of large-capacity accurate packet counting method for 100G interface of the present invention The key is to maintain two tables. Due to the 100G bandwidth problem, the read and write bandwidth required for statistics becomes very large. Therefore, in order to reduce frequent operations on off-chip storage, the same statistics are maintained internally, but only the lowest few are recorded. The bit width reduces the number of FPGA internal block memory (BRAM), reduces the cost of the device, and slows down the external storage read and write operations. It can be completed by using cheap synchronous dynamic random access memory (SDRAM), which is sufficient to support 100G bandwidth required for statistics.

Description

technical field [0001] The invention relates to the technical field of digital communication, in particular to a large-capacity and accurate packet counting method for a 100G interface. Background technique [0002] On June 17, 2010, the Institute of Electrical and Electronics Engineers (IEEE) officially approved the IEEE 802.3ba standard. In 2013, domestic operators began to fully deploy 100G transmission networks, and equipment connected to 100G networks relied on the After analyzing all kinds of traffic, the statistics of the set table items are obtained. The 100G bandwidth is 10 times the speed of the original common network interface, which brings the demand for statistical speed to increase by 10 times. The traditional large-scale static random access memory (SRAM) statistics method The interface rate is no longer sufficient. [0003] The front-end dedicated interface processing chip (or field programmable gate array (FPGA) internal 100G interface module is connected ...

AI Technical Summary

Problems solved by technology

[0002] On June 17, 2010, the Institute of Electrical and Electronics Engineers (IEEE) officially approved the IEEE 802.3ba standard. In 2013, domestic operators began to fully deploy 100G transmission networks, and equipment connected to 100G networks relied on the After analyzing all kinds of traffic, the statistics of the set table items are obtained. The 100G bandwidth is 10 times the speed of the original common network interface, which brings the demand for statistical speed to increase by 10 times. The traditional large-scale static random access memory (SRAM) statistics method The interface rate can no longer be satisfied

External storage at such a rate is no longer acceptable, or some people doubt that there is a static random access memory (SRAM) inside the logic device, which can be completed with internal storage, but the internal resources of the FPGA are limited, and it is currently used for core devices such as routers. The storage bit width of the byte and packet length statistics is 64 bits, and the current session capacity is more than 256K. After calculation, the storage capacity is 32M. It can be seen from the resource table of our mainstream XILINX Kintex 7 logic chip, the minimum The 7K70T has only 4M block access memory (BRAM)

[0007] Even if it is a relatively large device, if it is only used to complete the statistical function alone, the cost will be huge, which will directly affect the market competitiveness of the product

[0008] Moreover, it is well known that the price and storage capacity of static random access memory (SRAM) are not comparable to those of dynamic random access memory (DRAM). Even if the latest HMC (Hybrid Memory Cube, hybrid memory cube) is used, the price is extremely high and it takes Parallel-serial and serial-parallel converter (serdes) resources of logic chips, the main factors of the price of logic chips are parallel-serial and serial-parallel converters (serdes), internal wiring resources, followed by capacity and parallel IO interface Quantity, and the interface rate of 100G bandwidth According to the previous analysis, directly use the quadruple data rate static random access memory (QDR)-static random access memory (SRAM), the required interface rate brings circuit board wiring, power consumption, chip cost The increase in the accuracy of statistics is not worth the loss. Whether it is possible to use low-cost devices to complete the accurate statistics of large-capacity rule hits. In core functions such as device networking, security filtering, and route conversion, accurate statistics of interfaces are required to provide Analysis of Upward Strategies

Images