Summary of Birds-of-Feather meeting on Grid Networking, Part I Global Grid Forum 2 Vienna, Virginia, USA July 16, 2001, 5-6PM Notes Scribe and Editor: Nagi Rao, Oak Ridge National Laboratory raons@ornl.gov Meeting was initiated and organized by: Thomas Ndousse Networking Program, Office of Science US Department of Energy tndousse@er.doe.gov The meeting was very well attended - 59 people signed up during the meeting. Originally planned presentations could not be made due to limitations of time. A follow-on meeting was scheduled on July 17, 2001 for the presentations and also for further discussion. 5:00PM Thomas Ndousse started the meeting with a presentation that discussed the overall intended theme of this planned Working Group. He presented a brief account of networking issues for the grid community. Grids need networking support, and in a way the network is an important and critical part of the grid environment. He asked if we are doing enough for providing adequate support for grid networking. Particularly, the network has to be an integral part of the grid research. He then stated the following assumption: The current commodity Internet is adequate to support all current and future networking needs of all grids environments. This assumption is referred to as the GridNet assumption subsequently in this report. This assumption has very far-reaching implications: 1. Incompleteness of grid architecture in terms of network support 2. Impact of networks on grids is not being adequately understood and addressed 3. Inadequate emphasis on grid networking by funding agencies Then the discussion focused on the services layout of grid architecture by Ian Foster, wherein the network is represented at the same level as the fabric. Thomas presented that the network appears at several other levels in the grid architecture. Indeed network performance manifests at all levels of the grid architecture. He presented the networking perspective by refining the various layers, and emphasized the performance at various levels: 1. Application level, 2. Grid services, 3. Resource performance, 4. Network level performance. He discussed the network layer more in detail in terms of IP/MPLS layer, and lambda layer with optical domain advances. He presented his vision statement followed by three goals. He also presented a list of potential grid networking areas. He emphasized the need for grid networking standard documents. Next BOF on Grid Networking has been scheduled at the Global Grid Forum 3 in Frascati, Italy (near Rome) during October 7-10, 2001. Floor was then open for discussions. There was a lot of interest and participation in the meeting. First, there was an over-whelming response by a number of attendees in openly stating that the above GridNet assumption is not valid. Indeed, there was not a single response that stated that this assumption is valid. List of intended presentations was presented, which are not presented due to anticipated time limitations. Edi Clark from London stated that the Europeans did not make the assumptions similar to GridNet assumption. In fact they identified various networking aspects of grids. There are several groups in Europe, especially in UK, that are working on various aspects this area. It was suggested that this working group work in cooperation with those groups. Steve Cobarb mentioned that he saw no explicit efforts on networks for grids in the main stream grid efforts. Warren Mathews asked if grid networking is just a subarea of grid performance. There were several responses. Most of them indicated that grid networking needs to be addressed separately but in cooperation with other grid working groups. Micah Beck commented if the people engaging in R&D in Networking are represented in grid forum, especially this meeting. The response seemed like there are a good number of them at the meeting. Debbie Agrawal wondered how the activities of this group relate to the IETF efforts. Thomas explained about the plans for cooperation between this group and IETFs through liaisons. John Crowcoft said that this group might replicate mistakes made in transport, and said that the traffic pattern from grids at any level must be addressed by suitable traffic engineering. He said that we may learn from IETFs a number of standard lessons. He also said that we need liaisons with the networking community to interface with other efforts. Bill Johnston said that there be will be several types of working groups in Grid Forum. He suggested that this WG must try to clearly layout its specific charter. He said that some of the interests of this WG will likely overlap with IETF's. But, this group is very interested in middleware, which has not been the prime focus of IETFs. Micah Beck cited the multiple TCP example and wondered if the mainstream network R&D people may not approve the grid versions of ftp or other custom transport mechanisms. Thomas Ndousse stated that there is a big gap between application and network for the grid environments. This gap must be bridged, for otherwise the entire performance of the grid might suffer. Grid applications cannot assume that network will be there in terms of the underlying infrastructure as well as the middleware together with the program and user interfaces. If two communities are separated, they will be isolated; they must work together. Rich Wolski stressed the middleware issues of grid applications. The grid applications can reach into the middleware, and match the grid processes with the network. John Crowcroft said that IETF is aware of the requirements of grid networking. We can also expect to see defensiveness with the ISPs in providing the grid functionality into the networks. He stressed the importance of applicability statement and the need to articulate it clearly. Brian Tierney mentioned that one of the biggest issues is the monitoring infrastructure needed for grids. Current efforts are very adhoc and do not seem to adequately cover all the aspects. Bill Johnston discussed more about the structure on the working groups. He said most of the information is also available in the grid forum website. Stuwart Carruthers, suggested that in the PCM30 reports, several of these are addressed, including the problem of bandwidths. He suggested that this group look at those reports. Bill Johnston said that the goal of working groups be as specific as possible. The Grid Forum wants to make it easier for a group of people to work together. Javad Boroumand asked "what exactly is the purpose of the working groups ?" Thomas discussed at length that the groups identify the issues, establish liaisons, contact others forums, bring the information for the grid forum, and also tackle some technical issues that are not of interest to general network researchers. We should identify a subset of network areas that are important to grids, which will not be developed otherwise. He said that a single document is not the end product. Someone must bring into grids end-to-end initiative. Grid community can give a problem to be solved to the working grids, and we could create a project to address it. Debbie Agrawal said that we may want to find some grid networking problems to solve, in stead of waiting for the grid people to formulate and give us the problems. Bill Johnston further elaborated on the groups Thomas mentioned the intended formation of the mailing list to which participants can sign on. John Dyer suggested that we should come up with a strawman version of the tasks of the WG, which should be circulated for comments and refinements. Bill Johnston mentioned that there seems to be enough interest in the area to have a "Working group in progress", for which the list server can be arranged from the gridforum site. The meeting concluded by 6:00PM in time for the bus trip to Georgetown.