Congestion control also allocates available bandwidth to many flows, more or less reasonably. Since the deployment of TCP’s additive-increase, multiplicative decrease (AIMD) algorithm in 1988, the Internet has changed significantly, becoming much more heterogeneous in many areas including link speeds and characteristics, and types of application. Although there is no immediate crisis, TCP’s AIMD algorithm is showing its limits in a number of areas, including but not limited to:
- insufficient dynamic range to handle high-speed wide-area networks (especially with few sources using them)
- poor performance over links with unpredictable characteristics, such as some forms of wireless link.
- poor latency characteristics for competing real-time flows.
The network research community has been very active in designing modifications and alternatives to TCP’s basic AIMD algorithm, especially when it comes to high-speed congestion control. A non-exhaustive list of examples includes High-speed TCP, Scalable TCP, H-TCP, FAST, and XCP. However the network research community cannot, by itself, effectively evaluate any such solution for Internet-scale deployment. To do this requires the involvement of equipment vendors, network operators, and other IETF participants, in addition to the research community.
The key goal of the Internet Congestion Control Research Group (ICCRG) therefore is to move towards consensus on which technologies are viable long-term solutions for the Internet congestion control architecture, and what an appropriate cost/benefit tradeoff is.
To achieve this goal will require extensive evaluation of different mechanisms in a wide range of network conditions, using a mix of simulation, analysis, and experiment. It is unclear at this stage whether any single proposed solution is viable, or whether there needs to be a synthesis of ideas from many places.
Such an evaluation is especially difficult, given that some mechanisms involve changes to queuing algorithms, or to packet forwarding, whereas others only require end-system modifications. It is difficult therefore to compare like with like. One goal of the ICCRG is to improve our methods for evaluating transport protocols. To this end, a set of simulation (or other) test suites will be developed. There is an opportunity when making any widespread change to go further than the simplest incremental modifications, but such larger changes have costs, and so critical to the relevance of any recommendations from the ICCRG will be that any proposed solutions are considered to be economically viable. If router modifications are proposed, collecting them and the tradeoff underlying them would be an important service of the ICCRG.
There are many different aspects to congestion control that the ICCRG should consider. For example, different real-time media applications have (to a greater or lesser extent) a difficult time with dynamic congestion control mechanisms, especially if changes must be made on short timescales. The rise of Internet telephony and the expected rise of high-bandwidth Internet television will therefore significantly impact the congestion control landscape. The ICCRG should therefore consider such applications, both in terms of how to best support these various classes of application, and in terms of how this impacts the overall emerging congestion control architecture.
Also within scope for the ICCRG should be consideration of how congestion control interacts with QoS mechanisms, with traffic engineering, and with lower-layer technologies such as optical-burst-switching. Finally, of significant importance is the interaction between denial-of-service attacks targeted at bandwidth exhaustion, any mechanisms intended to prevent such attacks, and the congestion control architecture. Thus the context for the work of this group is necessarily wide-reaching, and touches on both architecture and mechanism.
Because congestion control is typically a key function of transport protocols, it is hard to separate other transport layer issues from congestion control. Considerations for deployment of transport protocols are therefore also considered within scope. The ICCRG may also consider congestion and protocol performance problems in general IP networks, i.e., not only on the global Internet. One example of such IP networks are multi-tenant, heterogeneous datacenters, which in some sense represent a microcosm of the larger Internet; changes to the transport protocols being considered in that space are fundamental and can have broader implications.