A platform to re-invent the Internet?

Download as PDF

Many experts today propose enhancements for the Internet. They claim that the 30-year old concept, originally designed for a handful of military bases and universities, cannot cope with today's challenges of more than a billion of participants, and new features such as VoIP, VoD, IP-TV, videoconferencing etc. Many ideas and concepts to improve the Internet have been proposed. Among the many trends which can be observed is the re-introduction of connection-oriented communication based on flows or streams, like the flow transfer mode (FTM), which has been proposed by van As.[1]

However, there is a deadlock situation concerning the implementation. The reason is that the economy of almost the whole world relies on the Internet. No network operator or network administrator would allow any modification which would risk the correct operation of its network. The situation was different when the Internet was developed. At that time it was a playground, where new concepts could easily be deployed and improved.

This problem has been recognized and several trial networks have been deployed world-wide for research and experimentation, e.g., GENI (http://www.geni.net/) and Stanford University’s Clean-slate lab (http://cleanslate.stanford.edu/) in the US, FIRE (http://cordis.europa.eu/fp7/ict/fire/) in Europe, and many others.

Hence test networks are available, but the next problem is the availability of platforms. Researchers want to quickly implement their ideas in powerful nodes with throughputs in the multi-Gigabit per second range. However, there is also a lack of suitable platforms, which is outlined below. Possible candidates for experimental platforms are

• Routers  
• Ethernet Switches    
• PCs      
• Network processors

None of them are fully suitable for exploring new Internet concepts.

Commercial routers with multi-Gbps throughputs do not allow modifications deep inside the protocol stack. They are not made for experimentation. Any programmability could affect existing functions which would have to be verified by lengthy regression tests. No router manufacturer would invest this effort just to have an experimental platform with uncertain profit expectations.

Ethernet switches offer high throughput, but have limited functionality. Basically, they support learning, flooding and aging, most of it hardwired, not suitable for the addition of new layer 3 and layer 4 functions.

The PC is a versatile platform and with the Linux operating system a state-of-the-art PC can achieve more than one Gigabit per second throughput. But this is far too low for studies on node applications with throughputs of 10s to 100s of Gigabits per second.

At the chip level dedicated network processors exist, which do allow any modification of protocol stack headers of packets at speeds up to 10 Gigabit per second and more. Typically network processors use custom made processors optimized for data plane processing. They have special instruction sets, which are not supported by standard tools for the compilation and debugging of code written in higher programming languages. Assembly language has to be used and in-depth knowledge of the processor subsystem is required as well. Only a handful of experts within the network processor companies can do this work, so these companies are reluctant to provide this support for academic work.

How to overcome this deadlock?

One approach has been taken by the OpenFlow initiative (http://www.openflow.org). In their whitepaper (http://www.openflow.org/documents/openflow-wp-latest.pdf) they describe the platform problem, too. Their solution is to ask node manufacturers to add a flow routing interface to their routers and switches. Via a secure interface the programmer can specify special routes for newly defined packet types. While this idea has been taken up with much interest by operators - expecting open platforms with less software cost - the big manufacturers are obviously reluctant.

Another approach has been taken by the startup company InnoRoute (http://www.innoroute.de). Their goal is to provide a scalable platform to the academic community that is easy to program. To do this in an efficient way they took a pragmatic approach, building on available components. Ethernet switches solve the scalability issue, as they are available up to hundreds of Gigabit. Equiped with InnoRoute’s FlowEngine researchers are able to trade hardware cost for performance at a fine-grained level. For programming they rely on standard processors with, e.g., a Linux operating system.

The figure on the left shows the InnoRoute FlowEngine between control plane processor and data plane Ethernet switch. From the point of view of the data plane the FlowEngine adds intelligent packet processing, while the programmers of the control plane processor use it as a programmable hardware accelerator.

InnoRoute intends to provide a modular demo system in 2011 for the academic community.

European Framework 7 projects would be the ideal framework to accelerate demonstrator development and explore the platform with Future Internet experts. Foreseen applications in this context include but are not limited to:

• Protocol development for overlay networks
• Investigation of new routing paradigms
• Implementing IPv6-optimized routers
• ATM-like OAM functions adapted for IP networks

Research will show, if it’s InnoRoute’s platform that researchers were waiting for to re-invent the Internet.