• 140 likes • 229 Views
Scheduling of Experiments. FACET User Meeting . Christine Clarke, 10 th October 2012. 2012 Schedule -- FACET User Run 1. Shift Schedules. https://slacportal.slac.stanford.edu/sites/ard_public/facet/user/Pages/Schedule.aspx Weekly pattern: Wednesday was an Access Day (PAMM)
E N D
Scheduling of Experiments • FACET User Meeting Christine Clarke, 10th October 2012
Shift Schedules • https://slacportal.slac.stanford.edu/sites/ard_public/facet/user/Pages/Schedule.aspx • Weekly pattern: • Wednesday was an Access Day (PAMM) • When LCLS access (every other Wednesday), FACET beam needs to turn off • Thursday was Machine Development • MD is a great buffer between PAMM and User shifts • 5 remaining days for experiments • Later in run, Friday morning became MD time too • MD time was extremely beneficial to delivering good beam
Part 1A Schedule Experiment “groups” • Experiments were grouped together so they could cover 24 hours a day for five days straight • Group A: E-203 (Smith-Purcell), E-206 (THz), T-501 (CERN BBA) • Group B: E-201 (DWF), E-205 (Euclid/energy chirp), E-207 (2-channel dielectric), E-204 (Metallic Structures) • Group C: E-200 (PWFA) • Group D: E-202 (Ultrafast EM Switching) – one shift/week due to sample change-over constraints https://slacportal.slac.stanford.edu/sites/ard_public/facet/user/Pages/Schedule.aspx
Issues… • Group A: • Grouping allowed 24 hour coverage but incurred many config changes • This group lost a week due to power outage • With single shifts, some experiments in this group were unfairly penalised by lack of owl shift accelerator physicist • Group B: • Smaller groups dependent on larger group (E-201) for commissioning hardware • E-205 and E-207 didn’t get beam time • How do we transfer knowledge between Group B groups? • A real strain on a few key players in E-201 • Group C: • No unique issues? • Group D: • Frequent rescheduling of the single shift based on condition of machine and plans of other groups.
Most common complaint: • No time to think! • Access day too rushed • Too little time between weeks on shift • Can’t engineer good solutions • Can’t solve what is going wrong
Beam time allocation 2012 • Aim was to give PWFA ~50% of beam-time, group B ~30% and group C+D ~20% • Similar to beam request (time requested PWFA > time requested group B > time requested group C+D) • Perceived to be in line with FACET’s goals • Given relative sizes of the groups, this was what the personnel could support
Proposal 1.1 – Moving PAMM • As proposed in User Brown Bag in June… • PAMM one day a week (User’s suggested Tuesday) • But there aren’t enough Operators on a Tuesday to regularly support the search and subsequent recovery • Large experiments given a block 3 or 5 days • Preceded by ~2 days PAMM recovery, POMM and MD time • Note set-up time can be longer if next experiment requires it
Proposal 1.2 – Moving PAMM take 2 • There are enough Operators to support a Thursday PAMM • Friday has enough Operators but a PAMM just before a weekend brings risk if there is a bad recovery
Proposal 2 – Double-up • PAMM once a fortnight (at the same time as LCLS) • Longer PAMM • More MD time and more set-up time likely to lead to better delivery • Set-up time could be longer still if next experiment requires it
Combinations, Variations and Further Ideas • Two-day PAMMs with a one week schedule • Is less beam time what we want? • One-day PAMM once a fortnight • Beam quality will improve and there will be more beam time for Users • Interleave MD and set-up with User Time • Scheduling configuration changes during the User Run 1 proved difficult but it is necessary • Schedule beam tuning • Problems arise over time and it may be beneficial to hand over to the Sector0-20 team on a regular and scheduled basis for improved delivery
What’s up with those 8 hour shifts anyway? • 8 hour shifts: • Owl: midnight-8am • Day: 8am-4pm • Swing: 4pm-midnight • Some experiments want less beam time in a single block (E-202) ~4 hour shift. And some operated on 12 hour shift rota (E-201). • SSRL has 11pm-7am, 7am-3pm, 3pm-11pm shifts. • We should not be hung-up on the 8 hour rota if this doesn’t work for us
20 minute discussion • Talking Points: • Group A, B, C and D: how did that work? • How long should PAMMs be? • Can we have fewer PAMMs or do we need to have more PAMMs? • How long should blocks of experiments be? • How much time should we have for MD? • How do we schedule the set-up time? • Should experiments be interleaved more or less? • Should we schedule beam tuning or leave it to User request? • How do we best schedule the configuration changes? Can we get away without any over owl shifts? • Length and pattern of “shifts” • Better to avoid operator shift change-over? • Can we stagger User shifts with accelerator physicist shifts so we don’t have a User shift that never sees an accelerator physicist?