Talk:CSC 456 Fall 2013/4a bc

dynamic scheduling: http://www.ics.uci.edu/~cs237/reading/parallel.pdf
static load-balancing: http://www.vsrdjournals.com/CSIT/Issue/2013_05_May/Web/1_Jagdeep_Singh_1670_Research_Article_VSRDIJCSIT_May_2013.docx
dynamic load-balancing: http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.98.2736&rep=rep1&type=pdf
static and dynamic LB: http://www.advanceresearchlibrary.com/temp/downloads/jct/may2013/v2.pdf
LB performance: http://masters.donntu.edu.ua/2010/fknt/babkin/library/article11.pdf
LB Performance: http://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=3&ved=0CFgQFjAC&url=http%3A%2F%2Fwww.cs.ucr.edu%2F~bhuyan%2FCS213%2Fload_balancing.ps&ei=VDBUUtj4HYr29gSLh4GADA&usg=AFQjCNFo08VxZ0irGr6e-ejmr1TXDDL7hQ&bvm=bv.53537100,d.eWU&cad=rja

Possible example topics: human-slice project data: http://lspwww.epfl.ch/publications/gigaserver/piiiaapa.pdf
mapreduce applications: http://en.wikipedia.org/wiki/MapReduce
weather modelling: http://cdac.in/HTML/pdf/ECMWF.pdf
weather modelling: http://research.ijcaonline.org/ccsn2012/number4/ccsn1040.pdf
weather modelling: http://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=05645456

weather modeling: http://cisl.ucar.edu/dir/CAS2K11/Presentations/panetta/jairo.panetta.pdf
weather modeling: http://wwwpub.zih.tu-dresden.de/~mlieber/publications/para10web.pdf
weather modeling: http://wwwpub.zih.tu-dresden.de/~mlieber/publications/para10paper.pdf

Comments on first draft

Good organization; look forward to the text. I also suggest this paper

http://paper.ijcsns.org/07_book/201006/20100619.pdf A guide to dynamic load balancing in distributed computer systems AM Alakeel - International Journal of Computer Science and …, 2010 - paper.ijcsns.org

Comments on second draft

Generally well written; would like to see you extend it to describe situations in which each strategy works best. If you can find empirical results to support those guidelines, so much the better.

I think you need a better delineation of static vs. dynamic. Since Central Manager assigns each new task to the processor with the least work, it sounds like it is dividing the work at run time.

The load-balancing pseudocode needs to be accompanied by a prose explanation.

Comments on third draft

-"Work load" --> "Workload"

-In the first section, by "increased performance" do you mean "improved performance"?

Giving a description of the various strategies isn't really sufficient. I'd like to see you tell which strategy works best in various circumstances, preferably backed up by some numbers.

Is Central Manager a static or dynamic strategy? If it assigns work to the processor with the lowest current load, that certainly sounds like a dynamic strategy. How is it differentiated from Central Queue?

In Central Manager, you say, "different overhead than usual." What is "usual"? Perhaps you mean to say that the overheads are distributed differently than with the other strategies encountered so far.

When you say "fewer messages to be sent in order to facilitate load balancing," can you quantify that?

-In "Local Queue", can you quantify the large amount of interprocessor communication, and compare it with other strategies (Central Manager, Central Queue)?

In Central Queue, can you give a diagram of the cluster arrangement? It sounds like strictly hierarchical workload managers.

Please give some real-world applications, filling out the sections for which there are headings.

-Still, more than pseudocode is needed for the example. A description is also needed.

Comments on final version

I'm still a bit unclear on the difference between static and dynamic. What is the difference between Central Queue and Central Manager, for example?