Introduction
This document summarizes the work and resources that would be needed to make an effective EXO-200 analysis environment on the SRCF computing facility, under the assumption that this needs to be done without the user doing the analysis needing a SLAC account.
There are 5 major requirements, each of which is addressed below, namely access to:
- Raw data -- obtained at WIPP but immediately copied to and archived at SLAC
- Processed data -- obtained by running several stages of reconstruction and analysis algorithms on the raw data and also stored at SLAC. The algorithms are often refined and so the processed data is frequently regenerated from the raw data by re-running the algorithms on the SLAC compute farm.
- The source code used for reconstruction and analysis, both to understand what it currently does and to work on improvements (which is often an integral part of a student's research).
- Compute cycles to run analysis on the processed data, typically obtained from the SLAC compute cluster
- Access to collaboration documents and wiki where analysis techniques are discussed, as well as EXO websites where data quality is summarized
In the following discussion we have assumed that the user is able to get an SRCF account and compute resources through EXO's affiliation with Stanford (Giorgio Gratta).
Access to documentation and web applications
This can be provided by setting up a "collaboration enclave account" (== crowd account) for the user. Under the assumption that setting up such an account is allowable (and I already checked with Eric Shupert and he assures me it is) then everything we need is already available.
Work required: negligible
Access to source code
The EXO 200 source code is stored in a subversion code repository at SLAC. Up to now we have used "svn+ssh" access to the repository since this is the simplest mechanism when all users already have SLAC unix accounts. For other experiments we have used direct subversion access, which only requires that users have an entry in a list of "virtual" accounts, very similar to the "collaboration enclave account" described above. Assuming that this access is acceptable then setting up similar access for EXO account should be straightforward.
Work required: <1 day (not including any required cyber security review)
Access to data
The EXO-200 data is currently stored on SLAC servers with private IP addresses. There are several possible ways of making the data available at SRCF
- Copy the entire data set. This has the disadvantage the considerable disk space would be required at SRCF (currently ~150TB). This would also require some mechanism for synchronizing the data when it is updated at SLAC (a frequent occurrence).
- Give (a subset) of the SRCF computers direct access to the EXO-200 data at SLAC. This would require using the (existing but currently unused?) fiber between building 50 and SRCF, and configuring some kind of trusted access between the computer centers
- Use the same xrootd access that we currently use to make the EXO data accessible at NERSC. This would almost certainly work, but may require more work than option 2, and may not give such good performance. We also need to understand the data access limitations on the SRCF batch farm and ensure that it can access the SLAC proxy server (or perhaps run a second data access node at SRCF).
In terms of estimating work required we will assume option 3. The actual manpower required is likely to depend to some extent on the answers to some of the questions above, which are currently unknown.
Resources required: 1-2 weeks, depending on what issues are discovered. Would likely require assistance for SRCF/SLAC networking groups, and xrootd team (Andy Hanushevsky+Wilko Kroeger). Access to a server (or servers) to run xrootd proxy at SLAC and/or SRCF would be required.
Access to batch resources
Batch resources can hopefully be made available via the SRCF. For these resources to be effective we need to reproduce the environment available on the SLAC batch farm. This includes:
- Access to data (see above)
- Create EXO group account for installing software
- Set up automatic builds of EXO software using Jenkins continuous integration system.
- Scratch space for storage of results
Resources required: Server at SRCF to run Jenkins, plus disk space for builds (~200GB). Work required to set up server, and verify that EXO software runs as expected ~1 week. Scratch space and batch cycles as needed for the analysis in question.
Conclusion
Setting up such a facility does seem possible, and would require fairly modest resources, but nonetheless these resources would need to be identified (presumably with some budget code to cover the work). In addition to ensure the required work happens in a timely way some effort would need to be put into spearheading and coordinating the work.