Dear All,
I got the date (month) wrong in the date. It is April. Thanks to Jim for seeing this -- he would have had a lot of visitors with nothing to do!
-Rob
This is posted here:
Generally a quiet and "working" meeting -- people were decompressing from report writing and the after effects of SC01 I suppose. The newest things were the Tutorial Beta Test conducted by David Bernholdt and the Birds Of a Feather sessions during the morning of the regular meeting. From reactions I have received, both went well. Below is a summary of the proceedings:
David's work was greatly appreciated though it was generally recognized as only a beginning. This is complicated by the fact that users are asking for material now. The following is his description and analysis of it:
CCA Tutorial presented
It is worth noting that on Wednesday afternoon we presented the first official CCA Tutorial to about 10 attendees with varying degrees of familiarity with CCA.
I began with a talk introducing CCA concepts and also a little about how the CCA and CCTTSS work. Craig Rasmussen then presented a general intro to object and component programming, emphasizing that they are little different. Based on audience interest, we got Scott Kohn to give an off-the-cuff presentation on Babel. Finally, Lori Freitag and I walked through some live code and demos to give people a feel for what they'll see and do if they download the SC01 demo software distribution.
Though it was pretty unrefined, people seemed to appreciate it and get something out of it. We got some helpful feedback, which has been incorporated into some of the ideas presented below.
Finally the web site reared its head again:
Lois Curfman McInnes gave the post mortem. Generally it was very well received and people (even DOE) were impressed. Lois gives an overview of the scope triumphs and soft spots. Lois's viewgrafs appear here.
Scott Kohn spoke briefly about this. This is everything that is necessary to build a component outside of the code and the CCA specification itself. Craig Rasmussen, Matt S., Rob Armstrong expressed interest.
This was reserved for the second day morning part of the meeting, but we also had some time the first day afternoon. Generally this was well received. People did more than use this as an excuse to leave early :-). There were several concurrent subject areas:
Jim has provided the following viewgrafs from his MxN group: here.
This is an idea to create a GUI as a component (Dennis already has one). Some of the interesting ideas are allowing the component to construct its own presentation in the GUI using XML. Steve and Boyana are on this.
As usual, David has provided a thorough review of the proceedings:
This is to summarize discussions held over the course of the latest CCA Forum meeting in Santa Fe. We had a fair amount of breakout time, and although only a few people were focusing on tutorial stuff, we had some really productive conversations.
Kate's idea for regular pre-meeting tutorials seems to have been pretty good (based on the first one), and we want to continue to present these.
We discussed a variety of other types and venues for tutorials. We should look for opportunities to present tutorials at conferences, etc. SC02 is an obvious choice, but we should not focus exclusively on it. It would be good to find tutorial (or talk/poster) opportunities are large domain-science meetings (as opposed to computer science). The ACTS workshop (or whatever it evolves into) is another target. Tony Drummond seemed ammenable to hosting workshops even if they weren't specifically associated with ACTS.
The idea of including a hands-on component in tutorials or "bring your own code" workshops was discussed. We decided that this needs at least a full day, and preferrably more, to be effective. Facilities are also an important factor (preparation of the software environment, etc.) that would have to be dealt with. We decided to defer serious consideration of a hands-on component for at least a year or two while we develop real presentation materials and gain some practical experience with early adopters. Lois touted the idea of "bring your own code" workshops as being very effective with PETSc, and probably a good way to do hands-on CCA stuff.
The next major demo opportunity we know of is SC02; see below.
We would like to be able to demo 2-3 significant SciDAC science applications using CCA. Having a number of small mini-apps as demos (also useful in tutorials) was also thought to be a good idea. Jaideep Ray has promised a methane combustion application. And the Chemistry folks plan to have a quantucm chemistry+TAO demo ready by then. If these come through, we should be in good shape, but we should also look for other opportunities as the apps groups progress in their work. These two apps represent a small piece of the SciDAC apps pie in terms of level of funding.
Both ANL and ORNL are willing to host the booth. Questions were raised as to whether there would be a SciDAC booth and what would look like. The answers (discovered after the meeting) are that there _will_ be a separate SciDAC booth. It will be coordinated by ORNL's Betsy Riley (who has for many years coordinated our own booth). It will be 20x20 feet, and its primary job will be to point to SciDAC-related content at other booths. It is not clear if there will be posters, demos, or other opportunities/mandates on us for the SciDAC booth, but we have a good connection to Betsy, and so should be able to find out as soon as decisions are made.
In terms of hardware, people generally thought the setup in the SC01 booth worked well. The big SmartBoard was considered a good thing. It might be good to have one additional LCD display. People weren't comfortable switching between Linux and Windows on the display machines, so it wasn't done. Some interest in having separate machines and a KVM switch. I am concerned about the space required and cost of such a solution, though I wouldn't rule it out.
Some people are still very interested in being able to have a small cluster on-site for more computationally-intensive demos, a la the SC01 Tiny Cluster. Others feel that it is not absolutely necessary to actually run parallel demos on the show floor (as long as we can honestly say that it does work in parallel), and that the show network is generally good enough that we could use a remote machine for this purpose. This seemed to be generally acceptable. However the tiny cluster idea is also interesting/useful when thinking about hands-on tutorials, and is still being pursued by several groups.
Between the feedback session from the tutorial and other discussions, we made significant progress in outlining a set of tutorial modules we want to develop. The idea is that we can mix and match modules to the time available and audience interests. We need to develop mini-demos and code examples to go with this stuff. Lois commented that in her experience interspersing live demos into presentations helps maintain audience interest. Modules should also include "static" versions of the demos (screenshots, etc.) for situations where it isn't practical to quickly switch back and forth between presentation and demo (probably requires two machines or VMWare, unless everything is ported to Windows in the near future).
Modules and (rough) presentation order:
Notes:
Lois is close to having the basics of one three-way reusability demo using function minimization and dF/dx as the two solvers.
Three-way reusability demos will be most effective if all three can be shown in succession in the same framework/GUI session. This requires that framework & GUI must be able to disconnect links between ports. Rob and Ben have been informed.