Thursday, October 13, 2011

10/11/2011

One of the best ways to learn map reduce is to use an emulator on a
local machine and write a map reduce program. Control of map and
reduce tasks and distribution is the responsibility of the OS. Dynamic
indexing can be done through computation of a "main" index and when
searching merge the results with the smaller in memory newer index.
Efficient PageRank can be computed for web sized matrices by splitting
the link matrix into smaller blocks and dividing the work. Sweeping
iteration of PageRank can be done over enough iterations, if you
update each value in place the page rank vector will be the same as
making individual copies for each step. -Thomas Hayden