Oh... one more... In Hbase, each cell can be sub-matrix.
On Tue, Oct 21, 2008 at 11:09 AM, Edward J. Yoon <[EMAIL PROTECTED]> wrote: >>> Is this block layout efficient in every matrix operation? >>> How to choose the block size to make the parallel computation efficient? > > Good point. We should think/discuss about it. > > However, We have a capacity of 2d blocking at this time. :) > > On Tue, Oct 21, 2008 at 11:03 AM, Samuel Guo <[EMAIL PROTECTED]> wrote: >> hmm, your explanation is clear. >> >> but if we store the matrix to file, the block layout is fixed. >> Is this block layout efficient in every matrix operation? >> How to choose the block size to make the parallel computation efficient? >> >> On Tue, Oct 21, 2008 at 9:36 AM, Edward J. Yoon <[EMAIL PROTECTED]>wrote: >> >>> For example, >>> We have a 4 * 4 Matrix a : >>> >>> [a, b, c, d] >>> [e, f, g, h] >>> [i, j, k, l] >>> [m, n, o, p]] >>> >>> Then, we can store an 2 * 2 block matrices to file. >>> >>> block(0, 0) is a 2 * 2 sub-matrix as below. >>> >>> [a, b] >>> [e, f] >>> >>> Key : block(0, 0), Value : submatrix [a, b], [e, f] >>> Key : block(0, 1), Value : submatrix [c, d], [g, h] >>> Key : block(1, 0), Value : submatrix [i, j], [m, n] >>> Key : block(1, 1), Value : submatrix [k, l], [o, p] >>> >>> ..... >>> >>> Also, If we can access sub-matrix from Hbase table, blocking can be >>> pre-computed. >>> >>> On Tue, Oct 21, 2008 at 10:18 AM, Samuel Guo <[EMAIL PROTECTED]> wrote: >>> > should the matrices that do the multiplication have the same blockID >>> layout? >>> > >>> > And How to divide a matrix into BlockIDs to make the parallel computation >>> > most efficiently? >>> > >>> > 2008/10/21 Edward J. Yoon <[EMAIL PROTECTED]> >>> > >>> >> Can anyone review this ?? >>> >> >>> >> http://blog.udanax.org/2008/10/parallel-matrix-multiply-on-hadoop.html >>> >> >>> >> -- >>> >> Best regards, Edward J. Yoon >>> >> [EMAIL PROTECTED] >>> >> http://blog.udanax.org >>> >> >>> > >>> >>> >>> >>> -- >>> Best regards, Edward J. Yoon >>> [EMAIL PROTECTED] >>> http://blog.udanax.org >>> >> > > > > -- > Best regards, Edward J. Yoon > [EMAIL PROTECTED] > http://blog.udanax.org > -- Best regards, Edward J. Yoon [EMAIL PROTECTED] http://blog.udanax.org
