Gianni presented the multi-bunch framework developed for the simulation of multi-bunch PyHEADTAIL-PyECLOUD simulations.
- Gianni presents the outline and context showing some use cases requiring the new framework. A particular challenge is the resolution of the very different time scales of proton and electron motion to capture the coupled dynamics.
- Gianni went through the parallelization strategy used for multi bunch simulations. Parallelization is made on the segments on the ring. Assuming e-cloud resets itself between individual turns, the simulation over several turns can be executes in a kind of helix manner. A final optimization is done by increasing the granularity of "bunching" by reducing the size of the slots that are sent around to the different processors. This can give another factor 5 of speed up.
- PyPARIS is used as an additional layer to manage the parallelization and communication between the various Python objects. Data passed around are beam slots along with a small dictionary describing the fundamental properties of each slot.
- The modifications for PyECLOUD were significant. Gianni went through the necessary changes.
- Time discretization is more complex and changes depending on whether or not interaction with the beam takes place.
- Scaling was studied as well. Hyper threading works well.
There are minutes attached to this event.
Show them.