Carpet Design Walkthrough

Introduction

The AEI numrel group is hosting a "Carpet Design Walkthrough" on 13-15.Dec.2004. We plan to start at 3:30pm each day (European time), and run for roughly 2 hours each day. The meetings will be held at the numrel group offices at Benzstrasse 8/9, Potsdam Babelsberg. Anyone interested is welcome to attend in person, to listen in by telephone, or to watch/listen/participate by Access Grid video conferencing.

General discussion about this meeting (including topics to be covered) should be on the developers@lists.carpetcode.org mailing list (archives here). Most questions about organization should be directed to Jonathan Thornburg <jthorn@aei.mpg-zebra.de> (remove "-animal"); questions about the Access Grid should be directed to Steve White <swhite@aei.mpg-zebra.de> (again, remove "-animal").

Our current plan is to start things off with a more general "Cactus/Carpet Design Overview" on monday, covering design topics that should be useful to everyone using Carpet. Then on tuesday && wednesday we will move to more technical material (likely to be of interest only to people who are programming either inside the Carpet code itself, or infrastructure thorns that must closely interoperate with Carpet).

Agenda

Here's the current draft agenda:

• Overview of the Carpet (= Berger-Oliger) mesh refinement & time evolution algorithm
• Basic Carpet Terminology
  • terminology for patches, components, grids, refinement levels, ...
  • how grid points are numbered
• Overview of relevant parts of Cactus architecture
  • the schedule tree
  • the MoL thorn for time evolution
  • what's done by the flesh / the driver (Carpet, PUGH, etc) / MoL / other infrastructure thorns
  • how multipatch works (or is supposed to work) in Carpet
• More on Cactus Architecture
  • flesh/driver interface
  • the GH structure
  • how I/O and checkpoint/recover work
  • other important flesh data structures
• More on Carpet Design
  • interaction with the flesh
  • interaction with MoL for time evolution
  • interaction with CoordBase (set up grid size/shape/etc) and SymBase (symmetries and symmetry interpolation)
  • strategy for managing the GH
  • modes & mode switching
  • how grid {functions,arrays} are distributed across processors
• A (Very) High-Level Tour of the Carpet Implementation
  • partitioning into CarpetLib/Carpet/CarpetRegrid
  • main global data structures which infrastructure thorns might want (eg current component, level, patch, ...)
  • APIs for mode switching and setting the current component/level/patch/etc
  • how Carpet manages interprocessor communication (and how this could/should be changed to improve performance scaling)

Course Material

Part I: There is an example schedule.ccl file (pdf version) and an example parameter file meudon-18.par (pdf version).

Part II: Explanation of the schedule (latex source).

Go back to the Carpet home page.