EGR-RG BOF session GGF10 Berlin, Thursday 4:00 PM-5:30 Attendance about 130 Proposed Chairs: Toshiyuki Nakata, Ravi Subramaniam Secretary: Satoshi Itoh Note takers: Hiro Kishimoto, Jem Treadwell, Ravi Subramaniam, Satoshi Itoh * Sign-up sheet * Background of this RG (Charlie Catlett) * Proposed Charter (Toshiyuki Nakata) (Slide for this presentation is uploaded on GridForge.) * Discussion Discussion log: Ravi Subramaniam (Intel): Prioritize standards – which other standards are important in this effort. Jeff Nick (IBM): Talking about a new RG; OGSA itself is working on scenarios those are not just science but business. The GGF needs to decide what it wants to be when it grows up… what gets done here has relevance to commercial computing – (narrower view is that Grid is about scientific computing) – we are trying to position Grid in commercial world – we are shaping distributed computing in the large – that’s the work that’s being done here. Wolfgang Gentzsch (SUN): This RG does not define standard spec. Is it about how building blocks based on standards are put together at customer sites etc. to build enterprise grids? Yes! OK, so this is about how existing grid users are implementing, e.g. on GT2.4 with additional elements such as portals, so it’s about scenarios, so that this group can structure the input and put it into some form… specific needs not yet reflected in the current set of standards? A: Goal, do a survey of grid deployments… Q: But I was asking about the contents, so we’re trying to get a list of requirements that can be brought back to this RG? Yes, we want to get a clearer view… Toshiyuki Nakata (NEC) (referred as Toshi): Commercial vendors are working on business grid but they don't have clear view of commonarity. Charlie Catlett (ANL, GGF Chair): I am scientific guy. But I try to give you two examples; (1) a friend of mine deploys condor into tons of machines. I would like to ask him what is missing from current OGSA usecases. (2) In my own situation I have big linux clusters in locations 2000 miles apart, I deploy Globus Toolkit into machines. It’s like people telling me if I install the apache web server on my pc then I’m an e-commerce site, when I’m only partway there – so I would like to find out what experiences others had? I would like to ask you what is the functionality grid should provide. Jay Unger (IBM): So is it the application profile, or just the fact that you’re installing a large-scale grid? Jeff: Here is confusion for me. My comments before is about relation between existing WG/RGs and this RG. I agree with what you want to achieve – but we need to articulate what we want to be doing that’s different from what we’re already doing. OGSA-WG thinks it’s doing this – thick set of usecases from HP, IBM, Hitachi & Fujitsu etc. that do cover enterprise scenarios, may be falling short of the objectives you’re looking for, but what is the difference? Charlie: OGSA-WG with its work profile is going to want to focus on fleshing out the architecture based on first wave of requirements… end user in this case not the financial analyst or the chemist, I mean the person taking and deploying the grid… Jeff: Got it… the OGSA WG in order to figure out … the work product is the set of functional definitions – you’re saying this group will focus on here’s what we want, here’s where we are, do a gap analysis between what we have and what customers want. What’s missing… more focused on how do we render it… Charlie: View this group as a sub-contractor – Jeff: In this case I think there’s a lot of value… Milan Milenkovic (Intel): Have a sub (interest) group to look at the scenarios; look at this and see if if this fits the enterprise area. You can look at it as separate RG. Q: Might be better not to have a separate group, but have a sub-group of the OGSA-WG – seems to be a way to expedite the work… technical observation – having seen scientific grids, different applications of grids, scientific and …more traditional application. Hiro Kishimoto (Fujitsu): Big concern, we just finished use case doc, and want to move on architecture work. WRT usecase issue. I see the RG has advantage in 2 dimensions: first is a depth, because it focuses on a specific areas, because OGSA usecases are broad; second is timeline. OGSA covers current and emerging usecase only, not future usecases. This RG should also focus on future application RG have a target for long range. Andrew Grimshow (AVAKI): I echo Hiro’s talk about time scale, the RG will work on future application and market analysis. OGSA just went through this – Q’s: what is the doc intended to be used for – who is the intended audience? Robert Cohen (ESI): One aspect that has not been addressed is why we are getting more technical users to attend. There is just beginning to see first commercial grids. We are trying to get commercial users to pay attention to GGF. Some of these papers if they address fundamental problems and lessons then people will find them extremely valuable in seeing roadblocks that have to be addressed…Also identify road blocks in the future and give cook-book to venders how to solve these issues are also valuable. - e.g. security issues encountered Jeff: 3 areas: 1) doing gap analysis, OGSA-WG is focused on here are the functions that are needed, and on interoperability; it doesn’t say anything about implementations exist – value to examining if there are real existing scenarios, and do a gap analysis for commercial consumers about what’s missing, ongoing, very valuable. 2) What do you do with it (Andrew) –useful in helping us to establish a roadmap – this kind of gap analysis would be useful in that. 3) keeps coming up: need to get more commercial buy-ins, ties, etc. – I agree, and this doesn’t go nearly far enough; GGF needs to get more engaged with commercial organizations, do marketing, evangelize the mission etc. Overwhelming perception that it’s not happening – need to consider more… Comment: Market research firms, IDC etc, already have done these research. They issues white papers. Whole things can be done by these firms. If we have the gap document then we could give this to people to start discussion… Hiro: I don’t agree. We’re developing new technology. Market research firm cannot do the right analysis on emerging technology. Technology development and setting the direction of it is GGF’s task. Robert Cohen: I second that. The firm cannot do sophisticated job….taking a group that’s developing standards and finding out what’s missing… a good forum for people who are using/developing products…It is a huge value for gap analysis. Ravi: Fundamentally is there a significant difference between users – research communities with labs across the world have the same kinds of problems, so why does enterprise computing have to be equated with commerce… Pramila Mullan (France Telecom): I speak for user community. There are many similarities with science grid. But there are many differences with telecom and commercial. … perhaps useful to do a level-setting in requirements. This is very good to have this RG. ----------------- Jay: Application model issue here. What is the potential difference between gridable (parallelizable) applications and traditional commercial (regular business) applications. SAP, OARCLE, IBM are trying to change their commercial middleware grid-aware. …It would be useful for people in GGF to start looking at commercial apps and consider how they could be decomposed into useful web services… important q’s – do we successfully support that type of utility computing model… we’ve done that for the scientific computing model, and tried to project it into the commercial space, but we don’t know how successful it is – don’t know what to do with oracle or db2 – how does it make it useful in big scalable applications… Hiro: Jay, this may be out side of this RG. This RG is just to list up prioritization and common requirement for the business commercial application. Ravi: I agree with Jay in doing that generates requirements that go both ways 1) what do application vender need to… what is the gap between OGSA… Jay: we’ve intuited that composability in OGSA is necessary… hopefully doing some research will validate that… but how do the pieces in commercial apps have to be refactored to exploit that… every vendor eg IBM, oracle, HP have experiences, need to explicitly state things and then measure our architecture against that – did it work? Hiro: Agree but, It has been done at OGSA-WG. Jay: That’s OK. But now what we need is people who have actual experience in building applications – if I split up my app is OGSA arch providing the composition points that give me what I need – we’ve guessed – great guesses, but they’re just guesses! Don’t have concrete examples of commercial apps decomposed into web services that prove I’m right… Ravi: What we need to do in the future is have people …think about that…generate OGSA requirement. Jay: I’d find an application that’s clearly monolithic and not factored well, what if I had the money/effort to lay this on a grid/utility computing model… try to re-implement the use case on a grid… Don’t know how accurate your analysis is until you try to slice/dice… picking one or two and going through the thought process is still useful – real development is beyond the scope of this RG. Andrew: I suspect the issue will be that the problem will be to aggregate the services Comment: You may have better with larger coarse grained (ie. facade) and then just doing the work and determine what need to be done. The hiding of the implementation may be a first good step. Grid is about the same computation being decomposed over multiple computers. (Most people see this that way). OGSA allows the representation of the resource and aggregate in a way that would be useful (the aggregation of the services may be more useful than decomposition of services). Comment: Organization are moving to Web services. The value that provides to the enterprise is the management of resources and this has to be communicated. ------- Toshi: Charter OK? Wolfgang: It may not be practical. How useful on this RG – GGF is always ahead of the crowd, and enterprises are very conservative in changing their current architectures to (say) a grid model – have distributed architectures but don’t consider it a grid, which is why they’re not here – so you’ll see a gap between what GGF wants and what the rest of the world wants, and so you may get less input than you expect, but it makes sense to have this group otherwise the work gets theoretical and this groups adds the missing users. this can only be a long term work that needs to be done. Jay: Why are they missing? Because we haven’t done a good job in explaining the value of grids in non-HPC environments (non-computing oriented aspects of the Grid further) – it’s almost never about aggregating MIPS – Customers are interested in issues about resiliency, cost reduction, utilization etc. – all the other things. Wolfgang: Do we have WG for marketing? (Charlie : Yes, GMAC. But it’s not WG in GGF.) This is one of reasons for the gap. This could be forum for developers and industrial end users in exchanging ideas. Ravi: Comes back to what is the definition of Grid? Maybe this group can help to define it… Toshi: Charter discussion will be continued on the ML. Hiro: Is there any ML? (Satoshi Itoh: Ask the staff to create our ML) Toshi: Milestone… Satoshi Matsuoka (Titech, GROC): Workshop in June is too aggressive. Workshop in GGF12 is better. Workshops need to be organized within a week… Jay: Focus of GGF11 is enterprise grid computing. Thus it is good starting point to have a panel at GGF11.… could have a panel session in which people describe their experiences… this group would learn… could be a plenary as it is the focus of GGF11. We need to seek out for a panel discussion seek out some of how they succeed is achieving their target. We went and this for a resilient. We need to this in plenary and this group can be kicked off from. Charlie: Agrees. Charlie: Set up mailing list for people interested in this discussion – take panel idea to program committee, and half-day workshop makes sense too. Toshi: Charter discussion will be done in the ML. Closing the session by Toshi.