Do you know how to build on-demand cloud networks based on programmable switching chips?

Fancy Wang 1309 2021

Traditional switching chips restrict upper-layer business software

Computer network has developed from the era of software forwarding to the era of ASIC (Application Specific Integrated Circuits) chips

It is a qualitative leap. In the era of ASIC chips, the message processing and forwarding logic of the network is all etched in the hardwareThe slice logic is completed, and the software no longer participates in these processes, so that the performance of the network is improved by leaps and bounds.

With the continuous development of IT services, the special needs of the upper-layer business and control software for the underlying network have gradually emerged. For example, if some Similar to NAT (Network Address Translation, network address translation) or SLB (Server Load Balancer, the server load is (Balance) and other commonly used functions are moved down to the underlying switching chip to achieve, which can greatly improve the processing performance of the entire business on the network, and greatly reduce。

The deployment complexity of the service itself is low, so that the overall efficiency of the system is improved.

Unfortunately, such a demand cannot be met in the era of traditional switching chips, because the message processing and forwarding logic of traditional chips is in production.Engraving is all solidified in the hardware. It is impossible for developers who use such chips to make any changes to its process and logic during the working process of the chip.

Moreover, because of the complexity and diversity of such requirements, it is impossible for traditional switching chips to integrate these functions into the chip. In this way, the upper level 。Business systems can only be designed and developed within the limits of chip capabilities, and cannot be adjusted on their own.The programmable switch chip came into being under this background.

Programmable switching chip unlocks upper-layer business software

As the name suggests, the biggest difference between a programmable switch chip and a traditional switch chip is its programmability. Processing and forwarding messages with traditional switching chips

Logic curing is different in the chip hardware. The message processing and forwarding logic of the programmable switching chip can be adjusted on demand through software. Exchange system.

The design developer of the programmable switch chip provides the high-level programming language to develop the message processing and forwarding of the chip according to the needs of the upper-level business system logic, and then download such logic to the processing pipeline of the programmable switching chip, the programmable switching chip can follow these customized processes to handle network traffic and messages.

The benefits of programmable switching chips are also obvious. In the switching system with programmable switching chip as the core, business and control software are no longer Limited by the ability of the underlying chip, it can be developed and customized according to its own needs; by customizing different reports for different needs and different scenarios

Document processing and forwarding logic, the various pipelines of the chip can work together, without losing the overall performance of the system, the requirements of these scenarios are in the core the message forwarding level of the slice is realized, which greatly improves the overall efficiency of the system.

Of course, the hardware carriers of programmable switching chips are still ASIC and FPGA (Field Programmable Gate Array)

The combination of hardware resources (on-chip memory, pipeline, etc.) integrated in the chip is limited after all, so it is impossible to infinitely

All network functions are programmed in the chip. However, in the current network application environment, some common network functions are transferred to the chip to be able to the form of programming is more than enough.

From the perspective of its implementation logic and the results achieved, programmable switching chips are not only a development of hardware technology of switching chips, but also SDN (Software Defined Network (Software Defined Network) is a great innovation at the level of hardware chips supporting the network. Programmable Switch Chip Let Net

On the premise of maintaining high performance, the network has approached the future development trend of software definition as never before.Build on-demand cloud network based on programmable switching chip

In the cloud network, the switches that make up the underlying network actually play different roles in different positions, and the cloud network actually plays different roles.

?Building on-demand cloud networks based on programmable switching chips puts forward different requirements.It is a typical Spine-Leaf architecture cloud network. In this cloud network, all switches are actually divided into three different roles

In the actual environment, because Leaf switches are often deployed on the top of each rack, they are also called ToR (Top of Rack) switches):

1. Spine switch, which forms the backbone switch of the cloud network.

2. Server ToR, located on the top of the physical server cabinet, is a switch that connects all servers to the cloud network.

3. Gateway ToR, a switch that connects the cloud and the outside world (Internet or other parts of the cloud) located on the top of the gateway equipment cabinet.

Obviously, the above three roles assume different tasks in different locations of the cloud network, the characteristics of the traffic they need to handle, and the cost

The proportions of software resources are different. In a network composed of traditional switching chips, we can only use a set of fixed processing logic to deal with these differences

However, in a network of programmable switching chips, the situation is different.

Programmable chips can prepare different forwarding logic and hardware resource allocation ratios for switches of different roles:

1. For Spine switches: The most critical task is to provide a simple, wire-speed three-layer multipath forwarding channel. At the same time, in Asterfusion in the wireless cloud network architecture of the protocol, it is required to be able to carry a large layer 2 forwarding entry.

2. For Server ToR: The most critical task is to provide the perception of the virtual computing node world, the offloading and forwarding of the virtual network, and at the same time, In Asterfusion's protocol unlimited cloud network architecture, it is only required to be able to carry local Layer 2 forwarding entries.

3. For Gateway ToR: The most critical task is to carry and handle complex control between gateways, forwarding strategies and large-capacity routing

?Based on programmable switching chips to build on-demand cloud network capabilities, at the same time, in Asterfusion's protocol unlimited cloud network architecture, its ability to forward entries at the second layer is very low.

Of course, in the above architecture, we can also use a centralized or distributed model to distribute some functions such as NAT and SLB to the Gateway.

ToR and/or Server ToR.

Asterfusion adopts programmable switch chip architecture cloud network switch products

Asterfusion's full line of cloud network switch products all use programmable chip architecture. Cloud network switch based on Asterfusion, cloud computing

Operators can easily build a cloud network with programmable software and change on demand, and build a flexible and agile network platform for cloud services.

Asterfusion Data Technologies?takes "architecting a new generation of network for cloud computing" as its main goal. Based on core patents such as PICFATM and vUnderlayTM and a number of software works, Asterfusion Data Technologies?has launched leading cloud data center network products and solutions, which have now won the industry Favored by many well-known public cloud and private cloud customers.