1. High Performance Computing and the FLAME Framework Prof C Greenough, LS Chin and Dr DJ Worth STFC Rutherford Appleton Laboratory Prof M Holcombe and Dr S Coakley Computer Science, Sheffield University

2.  Application can not be run on a conventional computing system – Insufficient memory – Insufficient compute power  High Performance Computing (HPC) generally now means: – Large multi-processor system – Complex communications hardware – Specialised attached processors – GRID/Cloud computing STFC Rutherford Appleton Laboratory 2CLIMACE Meeting - 14 May 2009 Why High Performance Computing?

3.  Parallel system are in constant development  Their hardware architectures are ever changing – simple distributed memory on multiple processors – share memory between multiple processors – hybrid systems –  clusters of share memory multiple processors  clusters of multi-core systems – the processors often have a multi-level cache system STFC Rutherford Appleton Laboratory 3CLIMACE Meeting - 14 May 2009 Issues in High Performance Computing

4.  Most have high speed multi-level communication switches  GRID architectures are now being used for very large simulations – many large high-performance systems – loosely coupled together over the internet  Performance can be improved by optimising to a specific architecture  Can very easily become architecture dependent STFC Rutherford Appleton Laboratory 4CLIMACE Meeting - 14 May 2009 Issues in High Performance Computing

5. STFC Rutherford Appleton Laboratory 5CLIMACE Meeting - 14 May 2009 The FLAME Framework

6.  Based on X-Machines  Agents: – Have memory – Have states – Communicate through messages  Structure of Application: – Embedded in XML and C-code – Application generation driven by state graph – Agent communication managed by library STFC Rutherford Appleton Laboratory 6CLIMACE Meeting - 14 May 2009 Characteristics of FLAME

7.  The Data Load – Size of agents internal memory – The number of size of message boards  The Computational Load – Work performed in any state change – Any I/O performed  FLAME Framework – Programme generator (serial/parallel) – Provides control of states – Provide communications network STFC Rutherford Appleton Laboratory 7CLIMACE Meeting - 14 May 2009 Characteristics of FLAME

8.  Based on : – the distribution of agents – computational load – distribution of message boards – data load  Agents only communicate via MBs  Cross-node message information is made available to agents by message board synchronisation  Communication between nodes are minimised – Halo regions – Message filtering STFC Rutherford Appleton Laboratory 8CLIMACE Meeting - 14 May 2009 Initial Parallel Implementation

9. STFC Rutherford Appleton Laboratory CLIMACE Meeting - 14 May 20099 Geometric Partitioning halos radius P1P1 P2P2 P3P3 P4P4 P7P7 P 10 P 11 P 12 P9P9 P6P6 P5P5 P8P8 Processors P i

10. STFC Rutherford Appleton Laboratory 10CLIMACE Meeting - 14 May 2009 Parallelism in FLAME

11.  Parallelism is hidden in the XML model and the C-code – this is in term of agent locality or groupings  Communications captured in XML – In agent function descriptions – In message descriptions  The States are the computational load – weight not known until run time – could be fine or course grained  Initial distribution based on a static analysis  Final distributions method be based on dynamic behaviour STFC Rutherford Appleton Laboratory 11CLIMACE Meeting - 14 May 2009 Issues with HPC and FLAME

12. STFC Rutherford Appleton Laboratory 12CLIMACE Meeting - 14 May 2009 Parallelism in FLAME Parallel agents grouped on parallel nodes. Messages synchronised Message board library allows both serial and parallel versions to work Implementation details hidden from modellers System automatically manages the simulation

13.  Decoupled from the FLAME framework  Well defined Application Program Interface (API)  Includes functions for creating, deleting, managing and accessing information on the Message Boards  Details such as internal data representations, memory management and communication strategies are hidden  Uses multi-threading for work and communications STFC Rutherford Appleton Laboratory 13CLIMACE Meeting - 14 May 2009 Message Boards

14. STFC Rutherford Appleton Laboratory 14CLIMACE Meeting - 14 May 2009 FLAME & the Message Boards

15.  MB Management – create, delete, add message, clear board  Access to message information (iterators) – plain, filtered, sorted, randomise  MB Synchronisation – moving information between nodes – full data replication – very expensive – filtered information using tagging – overlapped with computation STFC Rutherford Appleton Laboratory 15CLIMACE Meeting - 14 May 2009 Message Board API

16.  Message Board Management – MB_Env_Init - Initialises MB environment – MB_Env_Finalise - Finalises the MB environment – MB_Create - Creates a new Message Board object – MB_AddMessage - Adds a message to a Message Board – MB_Clear - Clears a Message Board – MB_Delete - Deletes a Message Board STFC Rutherford Appleton Laboratory 16CLIMACE Meeting - 14 May 2009 The MB Environment

17.  Message Selection & Reading - Iterators – MB_Iterator_Create - Creates an iterator – MB_Iterator_CreateSorted - Create a sorted iterator – MB_Iterator_CreateFiltered - Create a filtered iterator – MB_Iterator_Delete - Deletes an iterator – MB_Iterator_Rewind - Rewinds an iterator – MB_Iterator_Randomise - Randomises an Iterator – MB_Iterator_GetMessage - Returns next message STFC Rutherford Appleton Laboratory 17CLIMACE Meeting - 14 May 2009 The Message Board API (2)

18.  Message Synchronisation: Synchronisation of boards involves the propagation of message data out across the processing nodes as required by the agents on each node – MB_SyncStart - Synchronises a message board – MB_SyncTest - Tests for synchronisation completion – MB_SyncComplete - Completes the synchronisation STFC Rutherford Appleton Laboratory 18CLIMACE Meeting - 14 May 2009 The Message Board API (3)

19.  MB Sychronisation: – The simplest form is full replication of message data - very expensive in communication and memory – The MB uses message tagging to reduce the volume of data being transferred and stored – Tagging uses message FILTERs to select message information to be transferred – FILTERs are specified in the Model File XMML STFC Rutherford Appleton Laboratory 19CLIMACE Meeting - 14 May 2009 The Message Board API (4)

20. The Message Board API (5)  Selection based on filters  Filters defined in XMML  Filters can be used: – in creating iterators to reduce local message list – during synchronisation to minimise cross-node communications STFC Rutherford Appleton Laboratory 20CLIMACE Meeting - 14 May 2009

21.  Iterators objects used for traversing Message Board content. They provide users access to messages while isolating them from the internal data representation of Boards.  Creating an Iterator generates a list of the available messages within the Board against a specific criteria. This is a snapshot of the content of a local Board. STFC Rutherford Appleton Laboratory 21CLIMACE Meeting - 14 May 2009 MB Iterators (1)

22. STFC Rutherford Appleton Laboratory 22CLIMACE Meeting - 14 May 2009 MB Iterators (2)

23.  FLAME has been successfully ported to the to various HPC systems: – SCARF – 360x2.2 GHz AMD Opteron cores, 1.3TB total memory – HAPU – 128x2.4 GHz Opteron cores, 2GB memory / core – NW-Grid – 384x2.4 GHz Opteron cores, 2 or 4 GB memory/core – HPCx – 2560x1.5GHz Power5 cores, 2GB memory / core – Legion (Blue Gene/P) – 1026xPowerPC 850 MHz; 4096 cores – Leviathan (UNIBI) – 3xIntel Xeon E5355 (Quad Core), 24 cores STFC Rutherford Appleton Laboratory 23CLIMACE Meeting - 14 May 2009 Porting to Parallel Platforms

24. Test Models  Circles Model – Very simple agents – all have position data – x,y,fx,fy,radius in memory – Repulsion from neighbours – 1message type – Domain decomposition  C@S Model – Mix of agents: Malls, Firms, People – A mixture of state complexities – All have position data – Agents have range of influence – 9 message types – Domain decomposition STFC Rutherford Appleton Laboratory 24CLIMACE Meeting - 14 May 2009

25. STFC Rutherford Appleton Laboratory 25CLIMACE Meeting - 14 May 2009 Circles Model

26. STFC Rutherford Appleton Laboratory 26CLIMACE Meeting - 14 May 2009 C@S Model

27. STFC Rutherford Appleton Laboratory 27CLIMACE Meeting - 14 May 2009 Bielefeld Model

28.  Work only just started  Goal to move agents between compute nodes: – reduce overall elapsed time – increase parallel efficiency  There is an interaction between computational efficiency and overall elapsed time  The requirements of communications and load may conflict! STFC Rutherford Appleton Laboratory 28CLIMACE Meeting - 14 May 2009 Dynamic Load Balancing

29. Balance - Load vs. Communication  Distribution 1 – P1: 13 agents – P2: 3 agents – P2 P1: 1 channel  Distribution 2 – P1: 9 agents – P2: 7 agents – P1 P2: 6 channels STFC Rutherford Appleton Laboratory 29CLIMACE Meeting - 14 May 2009 Distribution A Distribution B P1P2 Frequent Occasional

30. Moving Wrong Agents Moving wrong agents could increase elapsed time STFC Rutherford Appleton Laboratory 30CLIMACE Meeting - 14 May 2009

31.  Size of agent population  Granularity of agents – is there are large computational load – How often do they communicate  Inherent parallelism (locality) in model – Are the agents in groups – Do they have short range communication  Size of initial data  Size of outputs STFC Rutherford Appleton Laboratory 31CLIMACE Meeting - 14 May 2009 HPC Issues in CLIMACE

32.  Effect initial static distributions  Effect dynamic agent migration algorithms  Sophisticated communication strategies – To reduce the number of communications – To reduce synchronisations – To reduce communication volumes – Pre-tagging information to allow pre-fetching  Overlapping of computation with communications  Efficient use of multi-code nodes on large systems  Efficient use of attached processors STFC Rutherford Appleton Laboratory 32CLIMACE Meeting - 14 May 2009 HCP Challenges for ABM