| | 8 | I was only partly in this week, finishing my vacation. I met with Roy & Jim to discuss IPP -> PSPS issues, and have met chatted with Mark & Serge about the progress of a full-scale test suite (moving along). I spent a bit of my time catching up on issues and status. It is clear we've been continuing to struggle with getting data storage balanced across the cluster, particularly with getting data to ATRC. Chris & I looked into options, and we found a temporary work around to boost the throughput by targeting only the raw chip images (avoiding the per-item overheads). I also used some recent stats additions to pantasks to find some of the inefficiencies in our pantasks scripts, including the shuffle. These are mostly now fixed (the shuffle targeting should be a bit updated as well). I also helped Chris with the dark model noise problems -- there is still work to be done, but it looks like the dark model tries to hard to follow 2nd order temperature variations which may not be significant. Chris is working on a scheme to include the higher order terms only if justified (on a per pixel basis). |
