Strumenti Utente

Strumenti Sito


magistraleinformaticanetworking:spm:sdpm1314lessons

Questa è una vecchia versione del documento!


Distributed Systems: Paradigms and Models (A.Y. 2013-2014)

Link to the official lesson register page.

Date Hour Arguments Support material
23/09/13 2-4pm Course introduction. Hardware evolution: multicores, GPUs, cloud. Blackboard Audio 1st hourAudio 2nd hour
24/09/13 4-6pm Parallelisation of sample applications: issues and advantages Blackboard Audio 1st hourAudio 2nd hour
25/09/13 2-4pm Kind of parallelism: data, stream and control parallelism. Measures realtive to parallelism: latency, service time, speedup and efficiency. Amdahl law. Blackboard Audio 1st hourAudio 2nd hour
30/09/13 2-4pm Concurrent activity graph. Coordination of concurrent activities. Non functional concerns in parallel computing: performance, fault tolerance, security, power management. Blackboard Audio 1st hourAudio 2nd hour
1/10/13 4-6pm Load balancing: static and dynamic techniques. Fault tolerance: checkpointing, replication of computing resouces. Blackboard Audio 1st hourAudio 2nd hour
2/10/13 2-4pm (lab) Implementation of a simple stream parallel application with POSIX-TCP/IP Blackboard Audio 1st+2nd hour
7/10/13 2-4pm Implementation of a simple stream parallel application with POSIX/Pthreads (lab). Stream parallel patterns: pipeline, farm and loopback. Data parallel patterns: map, reduce Blackboard Audio 1st hourAudio 2nd hour
8/10/13 4-6pm Data parallel patterns: scan, stencil, D&C. Control parallel patterns: conditional, fork/join. High level patterns (intro): BSP, Simulated annealing, Genetic algorithm. Blackboard Audio 1st hourAudio 2nd hour
9/10/13 2-4pm Parallel design patterns (the software engineer perspective) Blackboard Audio 1st hourAudio 2nd hour
14/10/13 2-4pm Sample parallel patterns (D&C and geometric decomposition: discussion of the pattern description in Mattson' book. Introduction to algorithmic skeletons. Blackboard 1st hour Blackboard 2nd hour Audio 1st hourAudio 2nd hour
15/10/13 4-6pm Implementation of algorithmic skeletons: template based and macro data flow implementation techniques. Blackboard Audio 1st hourAudio 2nd hour
16/10/13 2-4pm Optimizations of a MDF interpreter executing skeleton code. Affinity scheduling in MDF implementation of skeletons. Handling state in skeletons. Blackboard Audio 1st hourAudio 2nd hour
21/10/13 2-4pm Handling state in skeletons: owner computes rule variables and “resource” variables. Optimizations explioting the known access pattern of non state parameter in skeleton compositions. Performance models: skeleton, template and architecture models. Exact and approximate modelling. Blackboard Audio 1st hourAudio 2nd hour
22/10/13 4-6pm Compile time, run time and post mortem usage of performance models. Mechanisms for code monitoring. Blackboard Audio 1st hourAudio 2nd hour
23/10/13 2-4pm Analytical and queue theory based modelling. Modelling of a GPU map. 2nd hour: Introduction to SKEPU. Blackboard (2nd hour only) Audio 1st hourAudio 2nd hour
28/10/13 2-4pm SKEPU: using the system on a CPU/GPU hardware. Compiling options. Differences using a map followed by a reduce w.r.t. a mapreduce. Cole's Manifesto principles. Blackboard Audio 1st hourAudio 2nd hour
29/10/13 4-6pm Introduction to FastFlow. How to install. Sample code. Blackboard (1st hour) Audio 1st hourAudio 2nd hour
30/10/13 2-4pm Introduction to FastFlow (2): composition of skeletons, ff_node lifecycle, implementation of a master/worker pattern, map. Audio 1st hourAudio 2nd hour
11/11/13 2-4pm Using SKEPU within FastFlow (sample code). Vectorizing compilers (sample code). Blackboard Audio 1st hourAudio 2nd hour
12/11/13 4-6pm Template implementation techniques: prefetching, name and time servers in COW/NOW parallel applications Blackboard Audio 1st hourAudio 2nd hour
13/11/13 2-4pm Avoiding unnecessary synchronizations (Owner computes rule and false sharing). Client/server model for parallel computations. Blackboard Audio 1st hourAudio 2nd hour
18/11/13 2-4pm Introduction to Skandium (T. De Matteis, 1st hour, with sample code). Portability of parallel applications (introduction of the problem) Skandium SlidesBlackboard Audio 1st hourAudio 2nd hour
19/11/13 2-4pm Portability of parallel applications (second part). Job stealing. Virtual machines. Targeting heterogeneous architectures. Blackboard Audio 1st hourAudio 2nd hour
20/11/13 2-4pm Rewriting rules. Blackboard Audio 1st hourAudio 2nd hour
25/11/13 2-4pm Rewriting rules: data parallelism into stream parallelism. Adaptivity, MAPE loop and introduction to behavioural skeletons. Blackboard Audio 1st hourAudio 2nd hour
26/11/13 4-6pm Behavioural skeletons. Blackboard Audio 1st hourAudio 2nd hour
27/11/13 2-4pm FastFlow: implementation of a map with a custom farm, parallel for, targeting distributed machines (Torquati) Blackboard Audio 1st hourAudio 2nd hour
28/11/13 9-11pm Behavioural skeletons: multiple concern management. Introduction to Intel Xeon PHI. Blackboard1Blackboard2 Audio 1st hourAudio 2nd hour
4/12/2013 2-4pm Guest lesson by P. Dazzi: BSP, Pregel and the need for Parallel Graph Processing slides Code
05/12/13 2-4pm First view on the final project: arguments of the application and skeleton project classes. RISC-pbb: motivations and components, legal compositions, sample skeletons implemented with RISC-pbb (farm and map). Blackboard Audio 1st hourAudio 2nd hour
magistraleinformaticanetworking/spm/sdpm1314lessons.1386320260.txt.gz · Ultima modifica: 06/12/2013 alle 08:57 (10 anni fa) da Marco Danelutto