The 2nd QIG meeting, on January 26th, 1999, was hosted by MCNC. The meeting was opened by Ben Teitelbaum who addressed the need to clarify modest, short-term deployment goals for the QBone. The QBone architecture document makes good progress on the architecture, but we need to focus on deployment plans as well. The BB group is very active. However, inter-domain BB implementation is not expected for the initial operational phase of the QBone. QBone network planning will start before the next QIG meeting. While most participants are interested in exchanging information on diverse traffic engineering and operational aspects, it is clear that the QIG will not work on binding guidelines in these areas. The DiffServ vendor survey or layer 2 interoperability issues are examples of related topics. Updates on the QBone implementation status were given by several QIG members. The QBone measurement group received their own mailing list shortly after the meeting. Their aim is to identify a set of required DiffServ QoS measurement parameters to be collected and made generally available by all participating QBone networks. Finally, the old QoS "chicken and egg" problem of application vs. network service development was highlighted. To get started, at least some application engineers should participate in the work of the QIG or cooperate closely with it. This may help both sides to understand each other's requirements. In this sense, the joint DVN/QIG meeting was a good start.
Version 03 of the Draft QBone Architecture was made available the day before the meeting. The main change of contents was the inclusion of a token bucket model to describe a QBone Premium Service packet flow. Such a flow should be consistent with a peak rate R (token rate) and a service MTU size M (maximum bucket depth). Replacing "button-pushing verbiage" resulted in some editorial changes. In particular, it was suggested to use the term "service level specification" (SLS) instead of "service level agreement" (SLA) for future versions.
It was decided that the measurement infrastructure section would be distilled into a set of required metrics and performance parameters, and that the discussion of specific tools that can be used to collect these data would be put in an informational document. Still missing from the QBone architecture draft is a discussion of a phased set of common operational practices. A goal was established to stablize the Draft QBone Architecture by the end of February.
Sue Hares summed up the results reached by the QBone BB group up to the QIG meeting. As the BB group is pretty active, interested readers may refer to the QBone BB homepage for a complete overview. Consensus on the inter-domain BB requirements now is sufficient to draft a document. The draft inter-domain BB requirements version 0 is expected for mid-February. While all members agree on the necessity of only one protocol for the inter-domain information exchange between BB's, no common view was reached on the issue yet. Multicast support of BBs was not discussed during the BB meeting the evening before. As many QBone applications will use multicast connections, Sue Hares's presentation on this issue was welcome. The discussion following stated a requirement of bi-directional connections using the same routes and communication of sender and receiver connectivity limits before connection setup. Sue invited all to continue this discussion on the BB mailing list.
The issue was discussed based on a presentation on DiffServ to ATM interoperability by Rüdiger Geib. DiffServ to Layer 2 QoS interoperability has aspects of both pure protocol interoperability and traffic engineering. While the former is often relatively straightforward to handle, the latter is definitely a sensitive area. Any binding guidelines in the area of traffic engineering are beyond the scope of the QIG. There is however a huge interest in an exchange of information on all aspects of traffic engineering, and the QBone itself is a traffic engineering playground. As this was clearly expressed by the QIG members during the meeting and during later discussions on the mailing list, the QIG should work on ways to organize this exchange of information.
Layer 2-specific results of the discussion: mapping of qualitative DiffServ services to ATM QoS was not part of Rüdigers presentation. It will require some more efforts, as ATM as yet supports only quantitative services. Phil Chimento suggested also tracking the Quantum projects discussion on the issue, as their testbed is based on ATM.
Steve Blake distributed information via the QIG mailing list on the planned QoS differentiation features of MPLS. A variety of mechanisms, including DiffServ, have been proposed, but not standardized yet.
Andy Adamson presented the actual DiffServ status of UMN/CITI/Merit. Andy is interested in application/network level measurement of QoS parameters. Chuck Song informed about vBNS experiences with Cisco's IOS12.0.3T. Chuck highlighted that vBNS will continue its efforts on RSVP as well as their DiffServ activities. Joe Lappa briefly reported a flood they had in the PSC labs, so there's no update this time. John Collins informed about the status and aims of MCNC. MCNC's Sonet research network is under test right now, and some more time will pass before DiffServ tests will start. Joe Mambretti, Linda Winkler, Phil Chimento and Doug Pearson explained the ongoing activities of MREN/iCAIR. While several key iCAIR applications are available today, the network architecture to support QBone services is still under development. Vishy Narayan updated the information from NREN and shared some of their experience with Cisco VIP 2.50 cards. The VIP 2.50 card was also one of the main issues of Becca Nitzan's report on ESnet status (given by phone). On request of the audience, Guy Almes briefly reported on the development of Abilene. While Cisco IOS 12.0S will surely provide some advanced features, it is not expected to fulfill the complete Abilene wish list.
A discussion of criteria for, and steps that would lead to, a successful deployment of the QBone resulted in a discussion of QBone network planning. Ben Teitelbaum suggested that the group consider very simple, mostly static provisioning options for getting a Phase 0 deployment off the ground. One suggestion was to provision a full mesh of QBone Premium Service reservations between participating campuses/labs. Each Premium reservation would be for a fixed amount of bandwidth and extend from campus demark to campus demark. This pragmatic approach makes it possible to accept a fixed number of users without having to worry about Connection Admission Control. However, as most participants know their far-end cooperation partners already or will only collaborate with one or two sites, a full mesh was not felt to be required for the QBone immediately. It was agreed that all QIG members planning application and measurement trials on the QBone would provide their connectivity requirements for Phase 0 of the QBone before the next meeting. This would allow the QIG to draw a first QBone network map, which will not be changed before Phase 1 of the QBone starts. Definitions of these phases and the other ideas mentioned above should form part of a straw man proposal on the initial QBone network and its planning.
Phil Emer reported on the status of the QBone vendor survey. The discussion showed that the survey should inform the QIG of all limitations a specific DiffServ/Premium implementation (card & software version) contains. These details may be made available by vendor development engineers. Good contacts with vendors will help to collect the information. The full truth on some piece of equipment may, however, only be discovered by testing it. While some QIG members are only interested in information on their own vendors' equipment, others would like to learn about competing products too. It was decided that the survey effort should shift its focus from surveying the vendors directly to surveying the implementation experiences of QBone implementors. A vendor workshop was suggested. Being represented by a development engineer or having such a person present for questions would be a condition for participation of a vendor.
Finally, the survey has to be built. Phil agreed to continue working on the issue with the information and ideas he has collected so far.
It was decided that the measurement infrastructure section of the QBone architecture draft be distilled into a set of required metrics and performance parameters, and that the discussion of specific tools that can be used to collect these data be should be put in an informational document. Application-specific performance metrics may also be considered. A QBone measurements subgroup was formed to draft the revision to this section and to create the separate information document. Nearly 15 QIG members expressed interest in participating in this effort, which will be headed by Matt Zekauskas. A mailing list for the QBone measurements subgroup has been created.
A well known deadlock in the development of new telecommunications services occurs when network engineers wait for application engineers to formulate their needs in terms of network service parameters, while the application engineers expect the network engineers to tell them what they can offer. A good way to break this deadlock is a close cooperation of application and network development during the design phase of new network services. In this sense, the joint QIG/DVN meeting provided a good opportunity for a start. Joel Mambretti presented the Digital Video Network (DVN) project. DVN plans to introduce a digital broadcasting testbed allowing trials with advanced digital video technology. Ben Teitelbaum gave an overview of the QBone initiative. Finally, Micah Beck informed about the Internet2 Digital Storage Initiative. A server architecture as suggested by DSI would allow a limitation of Premium connectivity to regional servers for multicast or broadcast applications. Long distance content transfers between servers may use connections with more relaxed QoS parameters.
The following action items were felt to be most important to proceed to the introduction of the QBone Premium service:
3rd QBone Interoperability Group (QIG) Planning Meeting
March 9th, 1999
NMSU, Las Cruces, NM
This meeting will be co-located with the joint NLANR/I2 workshop hosted by NMSU from March 7-10, 1999.