The speed of data center construction is getting faster and faster, the scale is getting bigger, the demand for data centers in the network is getting higher and higher, and the speed and capacity of single-link transmission have higher requirements. We will see from these years The popularity of 10G to 100G is much faster than from 1G to 10G. Not only is the bandwidth, but also the capacity of network equipment is also increasing. For example, the routing, ACL, and MAC specifications are very demanding and the network chip capacity cannot be satisfied. Therefore, many people use the external expansion method to implement the disguised form. Therefore, we will see some capacity-heavy equipment on the market. The capacity obviously exceeds the capacity that can be supported by existing private network chips. In fact, it is implemented by means of externally-expanded table entries. The expansion of network equipment mainly uses three types of external devices. achieve.
CAM (Content Addressable Memory) is a parallel comparison processor, also called NSE (Network Search Engine network search engine). CAM can read and write like RAM, the biggest advantage is the input data, return the address where the data is stored. It can quickly search from a huge database and return the best matching address. CAM can complete the exact matching search of keywords within one hardware clock cycle. More importantly, CAM can provide large storage space for table entries. When the rules of network chips are not large enough, they can be complemented by CAM to meet market competition needs.
Of course, the CAM capacity cannot be expanded as much as desired. On the one hand, it is limited by the CAM capability. On the other hand, the larger the CAM capacity, the higher the price is in the exponential level, and the requirements for the adapted CPU. The high-capacity CAM needs strong performance. The CPU can drive, so as to avoid the situation of the big horse-drawn car or the situation of the small horse-drawn cart, which will make the performance of the entire equipment unable to improve.
TCAMTCAM (Ternary Content Addressable Memory) is an extension of the CAM table. Using the index to find the result value in the CAM table, TCAM also uses the table lookup operation but is improved and can perform more complex operations. When CAM stores table entries, the length is fixed and not flexible enough. For example, before delivering a route, the specific distribution of the address prefix length is not accurately known. In order to ensure storage of multiple prefix table entries, each CAM needs more than one entry space, so the CAM storage space utilization is greatly reduced. TCAM is very flexible in terms of length requirements, and can store keyword entries of arbitrary length in the same TCAM chip. The status of each BIT in the general CAM memory is only two 0s and 1s. The TCAM has each BIT data except for 0 and 1 and there is a third status “don't careâ€, which is also referred to as three-state content of the TCAM. The reason for addressing the register.
The biggest advantage of TCAM is speed. The average speed of single-cycle search is six times that of SRAM. In the worst case, the search speed is increased by 128 times. However, TCAM also has obvious flaws. The first is that compared to the general RAM, the unit-bit TCAM is more expensive and the storage capacity is relatively smaller. The introduction of the TCAM device greatly increases the design cost of the entire network equipment; secondly, the TCAM adopts a parallel matching and comparison method. The consumption is large. When the number of TCAM chips is large, the heat dissipation problem of the device is highlighted. To introduce TCAM, the heat dissipation problem must be solved. In the search process, all keyword entries are compared. However, only a few keywords can be actually matched. Most of the comparison operations are wasted, and the operating efficiency is not high. Third, the TCAM needs to queue the entries. In order to achieve the longest mask match, this makes the update of the entry items more complicated, especially when there are frequent changes in the routing, it is easy to make mistakes.
FPGAFPGAs (Field-Programmable Gate Arrays) are products based on programmable devices such as PAL, GAL, and CPLD. FPGA emerges as a semi-custom circuit in the field of application-specific integrated circuits, which not only solves the deficiency of custom circuits, but also overcomes the shortcomings of the limited number of original programmable devices.
Using the TCAM method will increase the inter-chip delay, while reducing the available space on the device, thereby reducing the system performance of the circuit board, using the FPGA method is much better. FPGAs are much more flexible than CAMs and TCAMs. FPGAs can perform more complex functions than matching, so FPGAs can not only be used to store table entries, but also can implement some functions that cannot be implemented by CAM and TCAM, such as Netstream. Netstream filtering and statistical analysis of the entire device traffic enables the entire device to become smarter. The FPGA can implement message classification, filtering, keyword extraction, and high-speed search. It can also set up a special communication module to communicate with the main control unit.
FPGA has the characteristics of powerful performance, low latency, deep logic implementation, parallelism, low power consumption, etc., but it has not been used in network devices because of its high price. Nowadays, with the maturity of FPGA technology, the price factor of FPGA is not prominent. This has led many network vendors to begin to introduce FPGAs and use FPGAs to replace the existing mainstream TCAM solutions. This has led to a several-fold increase in the capacity of network devices. If the routing table entry is stored, it can be 512K. Now it is easy to use FPGA to reach several M capacity, and it is also more flexible. The delivery and matching of the table entry is just a unique feature of it. FPGAs complement the capacity rules that cannot be achieved with existing network chips to meet the needs of some specific data center users. Of course, FPGA can not cure all diseases, FPGA can do parallel operations, but the flexibility is very poor, so FPGA is most suitable for high-speed but repetitive work, this is exactly what network equipment needs, the combination of the two complementary The advantages and disadvantages.
In a short period of time, the capacity of network chip entries still cannot meet the needs of the existing market, and the demand for capacity expansion of network device requirements will not change. From CAM to TCAM to FPGA, it is the development history of the expansion of network equipment requirements. Now the implementation of FPGA is gradually becoming the mainstream, and FPGA will exist for a long period of time.
In the future, the capacity of data center network equipment will surely increase. This is the demand of the market and it is also the need of competition. Many people are worried that network equipment will be the same as PCs and servers. The last competition is performance and price. This concern is right. Now that the network equipment has begun to compete in performance and price, you can provide 1M routing table equipment, and I can provide 10M routing table equipment. Many times it depends on who has the largest capacity. At the same price, customers can buy larger-capacity equipment, and they always feel that it is cheap. Therefore, it is not only blindly providing large-capacity equipment. The key is cost-effective. The greater the capacity per unit area, the more competitive it becomes. .
Car Connector,Smt Patch Car Connector,Waterproof Patch Car Connector,Automotive Car Connector
Shenzhen Jinyicheng Electronci Technology Co.,Ltd. , https://www.jycconnector.com