incubator-hama-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Samuel Guo" <guosi...@gmail.com>
Subject Re: 2D blocking algorithm
Date Tue, 21 Oct 2008 02:03:10 GMT
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 <edwardyoon@apache.org>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 <guosijie@gmail.com> 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 <edwardyoon@apache.org>
> >
> >> Can anyone review this ??
> >>
> >> http://blog.udanax.org/2008/10/parallel-matrix-multiply-on-hadoop.html
> >>
> >> --
> >> Best regards, Edward J. Yoon
> >> edwardyoon@apache.org
> >> http://blog.udanax.org
> >>
> >
>
>
>
> --
> Best regards, Edward J. Yoon
> edwardyoon@apache.org
> http://blog.udanax.org
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message