NM-WG GGF17 (10-12th May 2006, Tokyo) Summary ============================================= MANY THANKS to the people who provided content! NM-WG Sessions -------------- NM-WG held two sessions at GGF17, with the following agendas: NM-WG #1: Thursday 11th May, 13:45-15:15 Mark Leese presented the following topics: - NM-WG introduction and history - Motivation: Why do this? - V1 Schemas: where we started from, inc. details of EGEE-JRA4 (a v1 user) - V2 Schemas: where we are now, inc. details of perfSONAR (a v2 user) - Details of groups using the Schemas, including demonstrations - Questions NM-WG #2: Thursday 11th May, 15:45-16:30 Discussion session: The NM-WG schemas are gaining popularity in the research and educational network domain within Europe and the US, for example with EGEE, DANTE and Internet2. What are the barriers to their adoption (or trials) in Asia? This was largely a dissemination and fact-finding mission for NM-WG. We presented an overview and update on its work, including coverage of trial implementations, before attemped to ascertain why there is little interest in our work in Asia-Pac. Present: Mark Leese (CCLRC, Daresbury Lab), Horst Dumcke and Masum Hasan (Cisco Systems), Raju Shah (Force10 Networks), Tomohiro Kudoh (GTRC, AIST), Kenichi Tanaka (IBM Japan Systems Engineering), Michiaki Hayashi (KDDI R&D Laboratories Inc.), Haruyuki Iida and Norio Nishihara (NEC Soft, Ltd.), Yukio Tsukishima (NTT), Akira Hirano (NTT Network Innovation Labs) and In Soon Cho (Seoul National University). Contents: - Mark gave a 90 minute presentation (http://nmwg.internet2.edu/meetings/ggf17/nmwg-ggf17AllSlides.ppt) covering NM-WG basics: group history, our previous work (version 1 schemas including lessons learned) and the current V2 work. V1 and V2 schema work was reinforced by describing the EGEE-JRA4 and perfSONAR projects, complete with simple demonstrations of each. Background was also given as to why this is important work, from the point-of-view of both network operators and Grid users. - There were about 12 people present for the presentation session. Since Mark was the only "core" group member present, this represents a better attendance of non-core members at our general "outreach" session than held at other GGFs. It was especially pleasing given the somewhat disappointing turnout at this GGF. Several of the 12 were networking or telco professionals. - There was a lot of interest in using plots of network performance to analyse performance problems. Some examples showed how useful this can be, but it was noted that some kind of guide would be needed to help identify or rule out common problems by studying graphs of performance. There was a lot of interest in this type of guide. - Not for the first time, there was some scepticism about network operators wanting to share data. Operators have traditionally been very protective of their performance data, since it can be commercially and security sensitive. In reply, we have some research and educational operators like ESNet, Internet2 and DANTE who are ready to share data and actively want to share it. While this is an important issue, NM-WG's work is mainly about providing a mechanism for sharing the data, not persuading people to share it. Our best option here is likely to be to lead by example: complete development of the schemas and put them into service with enthusiastic adopters, so that the benefits can be demonstrated in real life scenarios. During the discussion session Mark spoke to a Japanese Grid academic, which was a very useful and revealing exercise: - He felt that the problem in Japan (at least) is that there is a lot of separation between Computer Scientists and Networking people, and that no one exists to bridge the gap. He also observes that local networking people don't feel a need to update their operations, e.g. they are happy to run a network using just a command-line interface, because it is sufficient for their needs. Something like Web Services would therefore be viewed as "too new", with unproven benefits, and garner little interest. - Also, he felt that while plots of performance are useful, the case for adaptive Grid apps and more inclusive/autonomous network operation is not yet strong enough, a view he believed would be shared by his networking colleagues. Hopefully the arrival of a major deployment of something like perfSONAR will correct this perception. - The contact's group would be happy to look at schema using software, but as always time and people are in short supply, so it must be "low hassle" exercise. This was defined as software that's easy to install on a PC. Mark said that this is something we can **potentially** look at. - The contact was not sure how monitoring infrastructures such as perfSONAR interrogated the underlying network infrastructure, but expressed the view that both Cisco and Juniper are popular in Asia, so any software should work that equipment. For clarification, Mark pointed out that there is much more to this work than simple SNMP polling of devices. - On a positive note, the situation is improving. He used an example of a South Korean team that used Web Services to link an in-house resource scheduler to software produced by another group(s). The Korean team were very surprised about how easy it was to us WS to plug disparate software together, and they're now more receptive of Web Services. This suggests that a more inclusive approach will be required to achieve an example deployment in Asia. NM-WG may need to capture a group's requirements and actively assist the group in building software which implements the schemas. Alternatively, a group may be open to deploying an instance/node of an exisiting infrastructure such as perfSONAR, if it meets the local requirements and a certain amount of "hand holding" is available. This is an issue for discussion during a future NM-WG meeting/phonecall. As a slight aside, some separate attendees though that Asia may only take the NM-WG work when it is complete, analyse it, and then adapt it to their own requirements. Other Info ---------- Mark also visited the Grid World Japan exhibition floor and spoke with several exhibitors, ranging from services companies such as Fujitsu and Korea National Grid to OEMs like HP. There is clearly more commerial adoption of Grids in Asia-Pac than US-Europe, yet while the people Mark spoke to knew that the network was essential to bandwidth dependant and large scale Grid deployments, the *impression* he received was that no one was doing anything special with the commodity IP networks they were using or the Grid middleware/apps themselves. The exception was Naregi, a Japanese academic Grid project, the Grid Network Group of which are working to produce a secure and high performance Grid network middleware. This includes protocol research and investigating management of network resources based on Virtual Organisation. NM-WG relevant work would include the real-time measurement of Grid application network traffic (flow-based measurement of RTT, thruput etc.). You can find out a little more here: http://www.naregi.org/papers/data/04_hpca_06.pdf, although this presentation is from 2004!