G. Watson, N. McKeown, and M. Casado. NetFPGA: A Tool for Network Research and Education. In 2nd workshop on Architectural Research using FPGA Platforms (WARFP), 2006.This paper talks about NetFPGA, a platform developed at Stanford which helps users to build real networking hardware using design tools such as Verilog that can be deployed and debugged in an operational network. FPGAs or Field-Programmable Gate Arrays are designed to be configured by the users after they are manufactured using standard hardware definition languages. It consists of various logic blocks that can be programmed to be wired together to perform complex functions. NetFPGA project leverages this technology to design Network specific FPGAs that can be used to build an Ethernet Switch or an Internet router (and probably other middleboxes too?). The NetFPGA project has released 2 versions at the time the paper was written:
NetFPGA-v1
This consisted of:
- 3 Altera EP20K400 APEX devices
- An 8 port Ethernet controller
- 3 1MB SRAMs
- Ancillary Logic
- Place and route tool: Altera Quartus
Further, a locally developed tool, Virtual Network System was used to map the NETFPGA-v1 ports onto the campus internet. The policy used was that of simulating first and debugging last. Due to the absence of a CPU on chip, the vendor's analysis tool couldn't be used which led the authors to develop there own simple logic analyzer. However, version 1 had various limitations. The PCB board format required an obscure rack that was difficult to obtain and required self-assembly, the rate was limited to 10Mb/s, there was no on-board CPU and finally the development environment was only limited to Linux and Solaris. These shortcomings led the authors to work on version 2 NetFPGA.
NetFPGA-v2
The main changes in v2 were:
- PCI Format was incorporated.This provided a familiar format to most users and also provided greater bandwidth benefits.
- Further tools were provided to enable development in Linux, Solaris and Windows XP
- Place and route tool: Xilinx ISE tool (Linux or XP)
Using a CPU within the Virtex devices are still being explored. Further the authors have developed 18 NetFPGA-v2 boards and also have a bus-master DMA driver for 4-port Ethernet port. At the time of writing this paper, The authors were also in the process of implementing RCP (Rate Control Procedure) using NetFPGA-v2.
Comments
NetFPGA is definitely a great technology. It will definitely help researchers to actually implement a system and test how does it work in a real environment taking into account the hardware issues which the software routers are unable to simulate. I have a few questions in mind though and it would be great to discuss them in class:
- As I see it, there are 3 class of network components, Software routers, FPGAs and ASICs. NetFPGAs are definitely better than software routers. It would be interesting to discuss how close are they to application specific ICs (ASICs).
- Aren't software routers sufficient to test the 'correctness' of a protocol? If that is the case, why are NetFPGAs an exciting tool for researchers as well?
Software routers may be unable to achieve the necessary wireline performance to prove the viability of an network idea. Also the FPGA approach gives a good approximation of hardware complexity to implement the function.
ReplyDeleteI totally agree and donot intend to question the merits of FPGAs over software routers. But my question is about the marginal benifit NetFPGAs provide to researchers. It totally makes sense for Cisco to test its routers on NetFPGAs before mass producing ASICs but are the researchers using it as well (since I donot remember any of the course paper implementations talking about NetFPGAs)?
ReplyDelete