The EOSDIS Panel and invited DAAC scientists attended a meeting held on the Quissett Campus of the Woods Hole Oceanographic Institution September 27-29. The primary purpose of this meeting was to discuss how we could aid Bruce Barkstrom in his attempt to independently cost EOSDIS and to discuss the recommendations recently made by the National Research Council (NRC).
In particular, the NRC made two recommendations that applied directly to EOSDIS:
NRC EOSDIS Recommendation #1
Components of the EOSDIS now under development for flight control, data downlink, and initial processing should be retained but streamlined.
NRC EOSDIS Recommendation #2
The NRC Report
Most of the first day was taken up with an open and free-ranging dialog with John Dutton (Penn State). He started the discussion with a walk through of the NRC report, in particular, Appendix F. In his presentation Dutton stated that it was clear to the NRC subcommittee on EOSDIS that the management challenges of EOSDIS will be much more difficult to solve than the technological ones. He compared the cost of running EOSDIS to a number of other enterprises (national laboratories, universities, airlines) and made the point that some cost more on a per employee basis than others. But when he examined EOSDIS the cost was head and shoulders above the others. Why? Dutton reiterated a plea from the NRC report: we need the right model for EOSDIS -- re-engineering the system won't solve the problem. In addition, scientists must take more responsibility for EOSDIS.
In discussion afterwards, Dutton admitted that many of the recommendations were based on intuition, only. When he was asked if he or the panel were aware of the Independent Architecture Studies, Dutton told us that they became aware of them after the NRC had finished its study. In addition, it seemed to the EOSDIS Panel that the numbered subparts of the two main recommendations of Appendix F were actually recommended studies that should be done right away. Finally, Dutton stressed the need for a reliable cost estimate of the recommended federated data system to be made available as soon as possible.
Independent EOSDIS Cost Model
Bruce Barkstrom presented a questionnaire designed to explore the depth of consensus within the Panel on four primary ingredients of his cost model: hardware, people, distribution, and development. The results were analyzed by Barkstrom and presented to the panel on the last day. The amount of concurrence was surprising, as were the bimodal distributions. It was decided that a similar exercise be done with the entire IWG, perhaps at the next IWG meeting.
The xDR Improvement Process
Skip Reber presented the results from, and requested further input into, the xDR (EOSDIS Core Sytem [ECS] development track) improvement process. Of the many excellent ideas obtained (mostly via e-mail), only the suggestion that the reviews be broken into parallel sessions met with any strong opposition. It was commented that the flow of information about the review should also be improved. Currently the information density supplied to the board members is either overwhelming (six shelf feet of documentation) or "underwhelming" (viewgraphs of low information content). It was suggested that the review board could prepare a list of questions to focus the discussion quickly on items of interest; perhaps about half of the discussion could be handled this way. Other recommended improvements (fewer people, less theatrical presentations, alternative presentations, less-formal Review Item Discrepancy Documents [RIDs], early demonstrations of system capabilities) were met with general approval, although the number of people attending may be very difficult to reduce.
Data Assimilation Office (DAO) Status Report
Richard Rood presented a brief overview of the current status of the DAO. His primary objective was to encourage more panel members to become involved in the DAO operations. Menas Kafatos volunteered to become more involved in the activities of the DAO.
Agenda Modification
At the beginning of the second day it was decided by the group that breaking the meeting into two splinter groups (one for discussing the NRC report recommendations and one for helping Barkstrom better design his EOSDIS cost model) would be counter to the desires of the group. Consequently, the agenda was reorganized to allow for full participation in all discussions.
Headquarters/Goddard Code 170 Reorganization and Relationship to EOSDIS Panel
Vanessa Griffin presented the status of the latest reorganization of NASA Headquarters Office of Mission To Planet Earth (MTPE) and the new relationships between EOSDIS and the Mission to Planet Earth Office (Code 170) at GSFC. Essentially Code 170 has been put in charge of all implementation issues, science remains at Headquarters (except for responsibility for funding of the EOS instrument science teams). Dixon Butler will move to the Science Division. He will head up the EOSDIS NRC response team for the next 12 months. The primary programmatic interface for the Data Panel has moved to GSFC Code 170. Only the top positions in the revised organization have been filled at this time.
NASA's Plans for Responding to the NRC Recommendations
Vanessa Griffin also gave a presentation on NASA's response to the NRC recommendations relative to EOSDIS. The question that occupies their thoughts is not whether to follow the NRC recommendations, but rather how to follow and when. NASA has organized four "streams" of activity to respond to the NRC recommendations. The first is to determine the "objects" of competition. The basis of these objects is the standard data product list. The second stream is to decide what values, "business rules," and standards must be kept common. A common values workshop was subsequently held at GSFC November 6-8 to help define these; it was by invitation only. The third stream is to model the enterprise and its functions. The constitution of the United States is a model for a federated system; we need something similar for the proposed federated data system. The last stream deals with drafting the solicitation and determining which mechanism ought to be used, i. e., Announcement of Opportunity (AO), Cooperative Agreement Notice (CAN) or NASA Research Announcement (NRA). NASA Headquarters asked the EOSDIS Panel to strengthen its role within the IWG and to help engage more users in the NRC response process. ESDIS's Thoughts on NRC Recommendations
H.K. Ramapriyan presented the ESDIS project's views on the NRC recommendations in lieu of Dale Harris. He told us that the Project stands ready to work with the community to do the best job possible, within the constraints of the NRC report, for a revised mission with reallocation of funding to a broader community. He informed us of the existence of a steering committee (Kennel, Price, King, Harriss), a core study team (Butler, 3 project members, 3 science members, 2 Headquarters members, 1 Hughes Applied Information Systems [HAIS] liaison), and the formation of active analysis support teams. All of these groups are to support the initiation of a study that will clearly articulate new mission goals for EOSDIS; develop a process that permits adequate science community review; develop a schedule that takes launch pressures, etc., into account; and develops a process where costs, schedules, requirements, and expectations are aligned.
DAAC Managers' Response to NRC Recommendations
Michael Goodman (MSFC) presented the DAAC managers' response to the NRC report recommendations. Their response was a model of what EOSDIS data processing responsibilities might look like in a federated system. They propose a three-category system wherein data products will be managed according to the category they fall into. Category one would contain data that have high standards for continuity, quality, rigor, consistency, and promptness. These data products would be managed by the government (Level 0 archiving, Level 1 processing and archive). Category two would contain data with known "downstream dependencies" so that, if they were not produced on time, other data products would suffer. These data products would be managed by a federation tightly controlled by the government (much in the same way the DAACs are managed today). The final category of data products would be those that focus on innovation, have broad participation by the user community, and show a high degree of flexibility in product design. These products would be managed by a self-governing board under the aforementioned federation.
This model was brought up again later when the group was discussing bootstrapping the current system of DAACs into a federated system of data producers and became known as the DAAC managers' model.
Status Report from the Ad Hoc Working Group on Consumers (AHWGC)
Bill Emery (Colorado) presented some results from the first 11 responses to the call for input from the Interdisciplinary Science (IDS) teams. Some clear modalities have shown up in the data. In the statistics compiled so far two things have stood out clearly: (i) users want the data in monthly "chunks," and (ii) they prefer to receive the data as the result of a standing subscription order. Emery is hoping that the remaining 18 IDS teams will get their responses in soon.
Introduction of the SSIG
Bruce Barkstrom gave a brief introduction to a group recently formed. The Subsetting Special Interest Group (SSIG) has a mission to develop descriptions of experiences and needs, to use this information to develop EOSDIS requirements, and make this information available on the World-Wide Web. Their first workshop was subsequently held at Langley November 8-9, 1995 (see report on page 18).
Governance Model for a Federated Data System
The remainder of the EOSDIS Panel meeting was given over to a discussion of how a federated data system should be organized, empowered, and started. Although the details of this discussion will be given in a following report, a summary is provided here to provide a flavor of the discussion. The first and most important consideration was given to the framework of such a federation of data centers. A number of examples were drawn, but two were used more often than the others: (i) analogy to the constitution of the federal republic of the United States of America, and (ii) examples drawn from the development and growth of the Internet.
The suggested political model for a federation of data centers had as its components: the entities involved in, the purposes of, the activities of, the rules governing, and the means for activating and revising the federation. Some parts of this structure were filled out more completely than others, but the discussion led to an enlivened exchange of ideas. First and foremost the entities involved in the constitution of the federated data centers were discussed. After a relatively brief discussion, it was decided that the federation should include users, producer/users, and producers of EOS data.
The purpose of a federation of data centers should be to stimulate creativity, provide understanding of the data, preserve the data, increase stakeholder's participation, and improve the flexibility of the system overall. Activities of the federation should include producing the data, making the data available (publishing, distributing), providing infrastructure, developing mechanisms to accomplish these things, storing the data, and supporting the users.
Certainly, the problems surrounding the governance of a federated system of data centers are larger than the technological problems associated with it. Therefore, a system of rules will be required in order to govern the interactions between the various parts. For no other reason than that the U.S. Federal Government has three parts, the rules section was divided into three parts: developing plans, implementing plans, and adjudicating disputes. Developing plans should include deciding which items should be controlled (standards, interfaces, etc.); establishing requirements for participation in the federation; establishing standards, interfaces, and protocols; and providing the principles for resource allocation. The part that implements these plans should document the standards, define the interfaces, allocate resources, and engage in rapid prototyping. The part that adjudicates disputes received a lot of discussion, but in the end no clear mixture of mechanisms could be agreed upon, and this part was identified as requiring further study.
The final major section of this constitution, the activation and revision part, was discussed, and it was agreed that the means for these two critical activities would have to be worked out early if there was to be any hope of bootstrapping this federated system into action. Along those lines, three models were presented.
The first model was the DAAC managers' model, which has already been reported, above (see section "DAAC Managers' Response to NRC Recommendations"). The second model presented was a refinement of the Independent Architecture Study presented to the NASA and ECS contractor in September 1994. In both cases, Menas Kafatos made a presentation of a federated system that featured both tight and loose coupling between the data centers. Those data centers that should be tightly coupled are those that produce standards for Level zero and one (L0 and L1) data products. The remaining data centers should be loosely coupled, allowing them to produce higher level data products that would be stored at, and accessed from, a data warehouse. The data warehouse would then act as a data cache layer between the users and producers, and would optimize data production because higher level products would only be produced when requested and would have a finite lifetime in the data warehouse. This model was known as the George Mason University model.
A third model, presented by David Glover, was an outgrowth of conversations with Mark Abbott (who could not be at the meeting). This model concentrated on the maturity of the algorithms producing the standard data products. It was suggested that those data products that were considered mature, i.e. were in current use by and trusted by Earth scientists, be produced as currently planned (at the DAACs). But those standard data products that were to be produced by developing algorithms, be produced by the winners of an open competition. This model was considered to be a variation on the George Mason model, where the mature algorithm data products would be produced by a tightly coupled system of centers (the DAACs), and the developing algorithm data products would be produced by successful proposers to an open competition. This model was known as the Abbott et al. model.
Of these three models of a federated system implementation, it was concluded that the models were not incompatible but rather focused on different aspects of what the federation should or could be.
Motions Voted
At the close of the final day of the meeting motions were made and voted upon by the following people: Bryan Bailey (EDC), Bruce Barkstrom (LaRC), Bill Emery (CU), Bob Evans (RSMAS), Nahum Gershon (MITRE), David Glover (WHOI), Michael Goodman (MSFC), Bob Haskins (JPL), Michelle Holm (NSIDC), Menas Kafatos (GMU), Anne Kahle (ASTER), Hendrik Meij (SEDAC), Skip Reber (GSFC), Robb Turner (ORNL), Carl Wales (ASF), Warren Wiscombe (GSFC), Victor Zlotnicki (JPL). It was agreed at the beginning that all (except Butler, Ramapriyan, and Vanessa Griffin) who had been present for the discussions, should be allowed to vote.
Bruce Barkstrom advanced the concept that the EOSDIS Panel needed to have three working groups -- producers, consumers, and systems oversight. This led to a discussion of the nature and role of the EOSDIS Panel. The conclusion of the discussion could be characterized as an aspiration for the Panel to serve as the proto-group for the governance of the federation. In Barkstrom's words, "the EOSDIS Panel has to accept real responsibility and not just advise and criticize."
MOTION 1: Bruce Barkstrom moved that a group be formed to write a new charter for the EOSDIS Panel with three subgroups along the lines he presented as a step toward adopting a new approach to governance. This amounts to "self-chartering" the Data Panel into the role of a proto-governing body. VOTE: This motion passed with only one nay vote.
MOTION 2: The question was then put to the group as to whether there was concurrence on Recommendation 2, Item (a) "Responsibility for product generation and publication and for user services should be transferred to a federation...," (see above). The recommendation was read aloud prior to voting. VOTE: The vote was 6 for, 8 against, with 2 abstentions.
MOTION 3: The question was then put to the group as to whether there was concurrence on Recommendation 2, Item (b) "examine the systems implications...," (see above). The recommendation was read aloud prior to voting. VOTE: The vote was unanimous in favor of such concurrence.
MOTION 4: Menas Kafatos moved that full implementation of the federated approach should not be before the launch of AM-1 or TRMM; some sort of partial implementation should be sought. Given these two conditions, the group would favor a federated approach as recommended. VOTE: The vote was unanimous in favor.
MOTION 5: It was then moved that the definition of a federated system should be expanded to include the users, regardless of whether they receive government money to use the data. VOTE: The vote was unanimous in favor.
MOTION 6: It was then moved to task the Ad Hoc Working Group on Production (AHWGP) to categorize algorithms into categories "mature" and "developing" of the Abbott et al. model and report back to the EOSDIS Panel. The AHWGC was tasked with the responsibility of reviewing the AHWGP's categorization so that the two activities would produce the intersection between algorithm readiness and user demand. In the discussion it was stressed that the essence of the first category was intended to be the existence of a ready demand for the product shortly after launch. VOTE: This motion passed unanimously.
MOTION 7: It was moved that the DAAC managers' model be recommended to the Response Task Force (RTF), the group NASA has established led by Dixon Butler, as a basis for studying the federation concept along with the EOSDIS Panel's view of the federation, and include the users and producers, their concerns, and the flow of information within a federation. VOTE: This motion passed unanimously.