Very few of the innovations in deep networking have seen data center scale implementation. Because the Data Center network's extreme scale performance requires hardware implementation, which is only accessible to a few. However, the emergence of reconfigurable match-action table (RMT) paradigm-based switches have finally opened up the development life cycle of data plane devices. The P4 language is the dominant language choice for programming these devices. Now, Network operators can implement the desired feature over white box RMT switches. The process involves an innovator writing new algorithms in the P4 language and getting them compiled for the target hardware. However, there is still a roadblock. After designing an algorithm, the P4 program's compilation technology is not fully open-source. Thus, it is very difficult for an average researcher to get deep insight into the performance of his/her innovation when executed at the silicon level. There is no open-source compiler backend available for this purpose. Proprietary compiler backends provided by different hardware vendors are available for this purpose. However, they are closed-source and do not provide access to the internal mapping mechanisms. Which inhibits experimenting with new mapping algorithms and innovative instruction sets for reconfigurable match-action table architecture. This paper describes our work toward an open-source compiler backend for compiling P416 targeted for the V1Model architecture-based programmable switches.
翻译:由于数据中心网络的极端规模性能要求硬件的安装,只有少数人可以使用。然而,基于模式的开关终于打开了数据平面设备开发寿命周期。 P4 语言是设计这些设备的主导语言选择。 现在, 网络操作员可以在白色框 RMT 开关上实施所期望的功能。 这个过程涉及一个创新者以P4 语言撰写新的算法并将其编译为目标硬件。 但是, 仍然有一个路障。 在设计算法后, P4 程序编译技术不是完全开放的源。 因此, 普通研究者很难在 silicon 级别执行时深入了解数据平面装置的开发寿命周期。 没有可用的开源编译器后端。 不同硬件供应商为此提供了基于原始编译器的后端。 但是, 它们是一个封闭源, 并且不提供内部绘图机制的接入。 因此, 普通研究者很难深入了解他/ 她的创新性设计程序, 用于对新版本的版本的版本的系统化图纸型结构进行测试。