General overview of some of the highlights in PyHEADTAIL development and applications during 2015:
- Space charge study efforts
- Start of implementation for fast ion instabilities
Presentation of Stefan's master thesis work. Generate an abstraction layer to make GPU easily accessible in PyHEADTAIL, implement GPU into some modules, benchmark and demonstrate speed-up.
- The abstraction layer is ready. Module that now have GPU capability are the transverse and longitudinal trackers, wake fields and monitors (bunch and slice monitors)
- Switching between CPU and GPU is handled by a context manager.
- All routines are tracked in hash tables.
General comments and follow-up items:
- We will reserve the slot Friday 11-12 for the future meeting throughout the year. We will target a bi-weekly cycle. Meetings should be short and concise not necessarily having a presentation at every meeting but rather to serve as platform for discussions and share and distribute information.
- We may want to add interfacing with PySixtrack as something to be checked.
- Might have to check new synchrotron class in combination with GPU context manager
- We should try employing the Bologna cluster:
- Copy installation from Adrian and use it as it is
- Get a GPU ipython notebook operational
- Can we launch a study?
- Gianni: investigating MPI parallelisation on the slices
- Fast ions may not profit as much from parallelisation - much more turns are required with less time spent per turn
- GR: What type of problems will be used multi-turn makes for? KL: The PSB is one of the main candidates where multi-turn wakes are likely to be important. Crab cavity studies for HL-LHC may also need to consider multi-turn wakes.
- A. Passarelli: There is an issue for very short bunches when trying to generate matched distributions; the matching algorithm fails -> check if this can be fixed
- Upcoming meetings:
- Show basics of RF matching
- Show application of pizza slice analysis
There are minutes attached to this event.
Show them.