Is this document acceptable to the DMLT? What are the remaining open questions? How will we resolve them?
DMTN-148 is almost there suggest 2 weeks review by DMLT.
John Swinbank to setup feedback system with Chris Waters on calibration note (DMTN-148).
This should be baselined (change controlled)
Robert asks when we will start "acting on this" - e..g when could it be used for LATISS on the mountain. On going work from Andres and Merlin - where is the ingest and validate..
KT last stage getting from production system via OODS to summit to be used for ISR on summit. Certified and transferred to where its needed.
Jim - Good to separate operations concerns (how its used on the mountain) from about the code and how we implement. DMTN-111 could have the summit details. Tim - no agreement on every curated calibration had class somewhere, one end - other is the certification
How will DM respond to slips in the overall project schedule?
Calabrese coordinating mail pickup in Tucson.
Services which were used in commissioning/integration are easy to define as “done”.
Would be good to get a statement of thanks from project leadership to DM staff.
Aim to make “blurring” between construction and commissioning a positive opportunity.
Also look for opportunities in the deliveries to ops (but be careful that this is not blurring).
The details of financing, ramps, etc through FY22 & FY23 will have to be addressed on a case-by-case basis, depending on guidance from construction project management and the agencies.
Comments from Victor:
Covid 19 costs are not an appropriate use of current funding (baseline or contingency).
We will therefore adopt a new baseline; the so-called “over target baseline”.
The earlier we do this, the riskier it will be and the less accuracy it will have.
Currently seems like NSF will accept late replanning (October/November).
This is not an opportunity for us to reinstate previously-accepted descopes.
This information can be shared with the rest of the project.
Some concern expressed that operational priorities are different from construction priorities; we should be clear that staff transitioning do so in the project's interest, rather than just because it is financially expedient.
Victor is petitioning the agencies for a minimal status review this year.
The drawback of waiting longer for a rebaselining is that we have to live with bad metrics until it kicks in; might be an issue for reviews.
Not clear what the rebaselining process will be: could imagine an FDR-like process, but it's not clear that will be practical.
Concern raised that DM may be able to reach completion on close to original timescale.
Expectation is a 12 month delay with a cost of $3.5M per month. Do not believe there is a serious risk of this not being approved at the moment. Also do not believe there is a serious risk of being forced to accept technical compromises.
What's the timeline for an IDF decision (if one hasn't been made by the time of this meeting)?
What do we need to do to prepare for the IDF? How does it impact the other tasks that we are working on?
Early FOA discussions seemed to discourage commercial entities from being involved; this wording has been softened/removed in later versions.
Wil hopes to structure the FOA as infrastructure/middleware/execution; expect that commercial cloud vendors might be interested in the infrastructure part, but not the others.
The decision making process is still TBD: expect that a DOE review committee will evaluate responses to the FOA and make a recommendation to the high echelons of the agency.
We should not expect to have a resolution by the end of this calendar year.
Relationship of the IDF to construction and commissioning:
The IDF is intended primarily as a means for the pre-ops project to meet its milestones and prepare for operations, rather than as a service to the construction project.
We expect that, at least until a new USDF site is chosen, construction and commissioning activities will continue at NCSA. Wil O'Mullane has budget which can be used to procure more resources in support of that if necessary.
It may nevertheless be possible to use IDF resources for scale testing of DM services at a level beyond that which can be undertaken at NCSA.
Concerns were expressed that the IDF as envisaged does not provide a clear transition route from the current NCSA infrastructure to future USDF infrastructure. While these are valid, we note that the transition to future USDF will include a transition from NCSA, not just from IDF.
The IDF is sized to cover the activities described in DMTN-135.
POC activities go beyond those simply required to demonstrate the viability of the IDF (e.g. they include alert processing). However, these provide essential inputs for future rounds of decision making.
However, the basic goal for the POC is to replicate regular processing which is currently being carried out at NCSA, but using Gen3 middleware.
The POC is expected to produce meaningful results in early July; decision making on the IDF is expected by July.
IDF & POC implementation:
The platform for batch processing will be Condor on GCE. Alternative solutions (e.g. Airfoil) may be examined, but this is still at an early stage. Solutions which would irrevocably tie us to a particular cloud implementation are obviously unacceptable.
Action item from the previous DMLT vF2F to report on the status of the APDB.
The DMLT acknowledges that this work may have been delayed by the focus on middleware development.
Feedback from NCSA/Michelle is that a deep understanding of the structure of the data is essential, regardless of the database implementation chosen. We note that Andy Salnikovhas already undertaken this analysis, but the DAX, AP and LDF teams are ready to collaborate further as needed.
More test nodes are needed to fully understand scaling of the current system.
As yet, there is no story about the (user-facing) PPDB.
We encourage the DAX team to convert the information on DM-23881 to a technote at their convenience; the timing on this should be up to Fritz Mueller.
Fritz Mueller — engage with the LDF and, as necessary, AP team to best understand the data structures required for the APDB.
Discussion of longer term planning for ingest. What are the key questions? How do we get them answered?
Expect “authorized ingesters” to be able to self-serve, but they will have to follow a (TBD) process.
In principle, “bad” values in the data being provided to ingest should be fed back to the Pipelines developers as bugs. In practice, the processing systems are sufficiently in flux, and many of these are artefacts introduced by the ingest process, so this is not yet regularly happening.
We do not have a written specification for the semantics of database contents (e.g. use of IEEE inf, NULL vs NaN, etc), despite some memories of previous (undocumented) agreements.
Extensive discussion, but much of it seemed to retreat ground that we have visited before.
We discussed the right level of detail for the DPDD, and whether it needs to be radically (or slightly) redrafted. There was no conclusion to this.
We agreed to prioritise the production of a DMTN describing the overall architecture being developed here. This is effectively refreshing the action item on Wil O'Mullane from our previous vF2F and now codified as DM-23658. We agreed that Wil and Colin Slater should be charged with making this happen. The aim here is to propose as comprehensive and concrete a system as possible for future DMLT discussion.
What exactly are “prompt services” (in terms of the product tree, system architecture, etc)?
What are the desires and use cases that have been advanced for an expanded scope “prompt processing” system?
How practical is it for the DM construction team to meet those desires?
If it is practical, what is the timeline and plan for doing so?
This rather wide-ranging discussion provided more background material for further thinking than concrete decisions which can usefully be minuted.
We discussed whether the “commissioning” use cases championed by Robert Lupton can be unified with the alert production use cases. There was no really concrete decision here.
We note the requirement expressed by Robert Lupton for flexibility, and acknowledge that this is often more important that extremely high reliability in a commissioning situation.
We further note the desire to provide uniform interfaces at all our various processing sites as far as is possible.
We agreed that the best way to proceed is for Robert Gruendl to develop a prototype OCPS capable of executing pipelines based around the NCSA test stand. Michelle Butler agreed to provide staffing to make this possible.