QBone Premium | Scavenger | Deployment
News | Calendar | QoS Working Group

First Do No Harm

The utter simplicity of the best-effort service model is one of the essential reasons for the success of the Internet. It has allowed IP to be implemented over every conceivable link-layer and has made for extremely simple network management and inter-provider peering. Together with the inherent strengths of connectionless networking and the end-to-end design principle, the best-effort service model has enabled a fast, dumb, cheap, and wildly scalable internet.

The greatest challenge for internet QoS is to find lightweight traffic differentiation schemes that add value without adding significant additional operational complexity or endangering the principles that have made the Internet so successful.

The goal of the QBone is to test and deploy scalable QoS mechanisms in the Internet2 environment.

Premium Troubles

After several years of experience attempting to deploy an interdomain, EF-based, virtual wire service in the Internet2 environment, the Internet2 QoS working group has concluded that any reservation-based form of QoS faces prohibitively difficult deployment obstacles, including:

Moreover, within the Internet2 environment very few application performance problems can be traced to network congestion. Instead, end-to-end performance is often hampered by faults on or near end-systems including: broken TCP stacks (e.g. inadequate socket buffering), Ethernet duplex mismatch, and crummy cabling (e. g. CAT3, shared media, or physical damage).

Non-Elevated Services

In this environment, only the simplest forms of interdomain QoS are deployable. Currently, we are focusing on non-elevated flavors of best-effort service that deploy incrementally, with no need for policing, accounting, or significant change to operational practices.

One example of a non-elevated service is the QBone Scavenger Service (QBSS), which allows users, applications, and campus networks to mark traffic for potentially degraded treatment at congested downstream interfaces. QBSS is designed for bulk TCPs that are currently run voluntarily during periods of low-utilization (e.g. large nightly transfers of scientific datasets, network backups, CDN content pushing). It is also, however, gaining traction with Internet2 universities who want to downgrade the treatment of non-mission traffic, such as recreational file-sharing by students.

In addition, we are looking at non-elevated services like Alternative Best Effort (ABE) that could provide interactive applications with a low-latency best-effort service class.

Contact the webmaster or View the colophon