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

Reply via email to