Strumenti Utente

Strumenti Sito


Distributed Systems: Paradigms and Models (A.Y. 2014-2015)

Link to the official course lesson page.

Exercise page.

Date Hour Arguments Audio Blackboard
23/09 4-6pm Introduction to the course. Parallel programming scenario: multi core and many core H1H2 PDF
24/09 4-6pm More on scenario: cloud and mobile. Sample applications with possibilities for parallelism: mandelbrot, video stream processing (to be finshed) H1H2 jpg1jpg2PDF
29/09 9-11am More on sample parallel applications: differential equations (kind of), reduce. Stream parallelism (farm and pipeline). Speculative parallelism. Measures: speedup and scalability. H1H2 PDF
30/09 16-18pm Service time, latency, completion time. Efficiency. Amdahl law (with top500 and link). Concurrent activity graph. H1H2 PDF
01/10 16-18pm Concurrency activity graph, coordination, implementation. Non functional concerns (load balancing). H1H2 PDF
06/10 9-11am More on non functional concenrs: fault tolerance, power management. Survey of parallel patterns: pipeline (first part). H1H2 PDF
07/10 16-18pm Pipeline and farm: modelling, rewriting, removing bottlenecks. H1H2 PDF
08/10 16-18pm Howto: measure times in C programs, gather results from experiments, structure experiment sessions, keep data for plotting. H1H2 PDF
13/10 9-11am Setup of concurrent activities (pthreads and processes). Typical communication and synchronization mechanisms (locks, pipes, SysV primitives, sockets). H1H2 PDF
14/10 9-11am Task farm implemented “from scratch” with pthreads and C++11 threads (passive emitter version). Task farm on a cluster of workstations (ssh+sockets). H1H2 PDF Sources
15/10 4-6pm Data parallel patterns: map, reduce, scan/prefix, stencil, mapreduce. H1H2 PDF Sources
20/10 9-11 Class exercises (with laptops, please be sure to have an ssh client installed)
27/10 9-11am Introduction to skeleton programming frameworks. Attention: students attending the course should write a message to the prof. with Subject: PHI SPM ACCOUNT, containing in the body Name, Family Name, Matricola (Enrollement No.) and either MCSN or CS depending whether they registered to the COmputer Science and Networking or to the Computer Science Master degree. They will get back a message with the instructions to access the XEON PHI machine. H1H2 PDF
28/10 16-18am Presentation of the solution for the pipeline exercise (pthread only). Status management in skeletons (“resource” kind of state). Template based implementation of algorithmic skeletons. H1H2 PDF
29/10 16-18am Introduction to FastFlow: stream parallel patterns (pipeline and farm) (M. Torquati) Code: Makefilepipe_square.cppparfor_square.cppfarm_square.cpp H1H2 PDF
10/11 9-11am Macro data flow based implementation of algorithmic skeletons. H1H2 PDF
11/11 4-6pm Class exercises (with laptops, please be sure to have an ssh client installed) (M: Torquati)
12/11 16-18am More on Macro Data FLow implementation of structured parallel programming frameworks: shared task pool, maintinaing locality, accessing the fireable instructions. Parallel building blocks (introduction) H1H2 PDF
17/11 9-11am Parallel building blocks: definition and compositionality. Generalities related to general purpose and domain specific patterns and parallel computing models. Sample template implementation: map and BSP. H1H2 PDF
18/11 4-6pm Discussion of the solutions for the second class work exercise. Introduction to FastFlow parallel for (M. Torquati). H2 PDF
19/11 4-6pm Class work 3: matrix multiplication with parallel for in FastFLow (M. Torquati) PDF
24/11 9-11am More on parallel building blocks: rewriting and optimizations. Introduction to parallel design patterns H1H2 PDF
25/11 4-6pm Macro Data-Flow pattern in FastFlow: ff_mdf. The FastFlow ff_taskf pattern for parallel scheduling of functions/lambdas. Class work 4: matrix multiplication using the ff_mdf pattern. (M. Torquati) FastFLow H1 PDF PDF Sources
26/11 4-6pm Design patterns: design spaces and relative patterns. H1H2 PDF
01/12 9-11am Mechanisms: double/triple biffering, client/server (with multithreaded server), excess parallelism. H1H2 PDF
02/12 9-11am Mechanisms: vectorization (see also intel compiler manual), cache optim (false sharing, perfetching) H1&2 PDF
3/12 4-6pm More parallel programming frameworks: Skandium, SKEPU H1H2 PDF
5/12 2-5pm Discussion on previous class work. Assignment of a new Class Work: dynamic map computation. (M. Torquati) PDF
9/12 4-6pm Presentation of the Final project PDF. Rewriting rules and normal form. H12 PDF
10/12 4-6pm More on rewriting rules. Behavioural skeletons. H12 PDF
15/12 9-11am Bheavioural skeletons. General “SPM Philosophy”. H1H2 PDF
16/12 4-6pm Discussion on previous class work assignment: dynamic map skeleton implementation (see Class Works here). Sketch of possible other implementations. Discussion on exam project (M. Torquati)
magistraleinformaticanetworking/spm/sdpm1415lessons.txt · Ultima modifica: 17/12/2014 alle 09:37 (3 anni fa) da Massimo Torquati