Columbia UniversityIRT labColumbia University

SIGLITE

SIGLITE is an effort to study lightweight signaling for Internet resource reservation. Though it may be useful and insightful sometimes, here, we should try to avoid further debate on "do we need resource reservation". It might be helpful to assume that we might need it for "some" subset of Internet flows in some network, and maybe get some grasp on the scaling issues - i.e., what are reasonable assumptions on scaling needs. (In other words, scaling doesn't much matter if only 100 Mb/s flows need it or can get it, since you're not likely to have too many of them at any given time.) 

Overview:

Here are the technical issues that we are trying to address in SIGLITE:

  1. There is the perception at least that traditional QoS signaling is hard to implement and resource-intensive (in memory code or data space and/or CPU cycles), in particular for low-end end systems such as IP phones and Internet appliances, even though the task at hand is pretty simple: "May I have 100 kb/s to 128.59.16.3?" "Yes, please."
  2. There's a fair amount of disagreement as to scaling of per-flow or per-aggregate reservation. Some say that this doesn't scale, some say that it doesn't need to scale, some say that scaling problems are primarily issues of either implementation or protocol complexity, not necessarily the notion of reservation state in the network. Are there mechanisms to address these scaling issues? What are the trade-offs in terms of the type of services that can be provided (e.g., not all metrics may be amenable to being aggregated)?
  3. Also, we seem to be lacking a mechanism for negotiating and conveying prices for QoS. (The usual disclaimer: the IETF doesn't get involved in business models, so this topic may need to be suitably abstracted before it becomes BOF material, but since this mailing list is not a WG mailing list, this shouldn't keep us from discussing the topic, if there's interest.)
  4. Some have proposed that there's a wider class of problems that need signaling at the lower layers, such as tunneling through a bunch of firewalls and NATs or setting up lambda or label paths across some circuit-like network. Is there a common protocol or architecture infrastructure that can be re-used for a variety of signaling needs? Is there such a thing as signaling, which I'm defining loosely as protocols and mechanisms for establishing distributed state in the network?

For further understanding on Internet QoS, please refer to RFC2990: Next Steps for the IP QoS Architecture.

Main Proposals:

Over the years, there have been many studies on, signaling scaling and process simplification.

Another aspect in Siglite is to evaluate different Internet QoS pricing schemes:

In the past several years, there have been some studies on RSVP performance:

There are more papers on RSVP available at the official RSVP site.

Misc. Information:


Last updated by Ping Pan, Henning Schulzrinne