space allocation for a parallel matrix

[Yixun Liu]

Matthew Knepley wrote:
> On Tue, Feb 3, 2009 at 9:47 AM, Yixun Liu <enjoywm at cs.wm.edu> wrote:
>   
>> Matthew Knepley wrote:
>>     
>>> On Tue, Feb 3, 2009 at 9:25 AM, Yixun Liu <enjoywm at cs.wm.edu> wrote:
>>>
>>>       
>>>> Hi,
>>>> I am confusing to the allocation for a parallel matrix in PETSC manual,
>>>>
>>>> "If PETSC DECIDE is not used for the arguments m and n, then the user
>>>> must ensure that they are chosen
>>>> to be compatible with the vectors. To do this, one first considers the
>>>> matrix-vector product y = Ax. The m
>>>> that is used in the matrix creation routine MatCreateMPIAIJ() must match
>>>> the local size used in the vector
>>>> creation routine VecCreateMPI() for y. Likewise, the n used must match
>>>> that used as the local size in
>>>> VecCreateMPI() for x."
>>>>
>>>> Since the process holds all the columns, why does it only hold part of
>>>> vector x? Is it convenient for the matrix vector multiplication?
>>>>
>>>>         
>>> I am not sure I understand your question. However, Mat objects do hold
>>> all columns of a row. The local column size 'n' is used to match against
>>> a Vec, not for storage.
>>>
>>>   Matt
>>>
>>>
>>>       
>>>> Thanks.
>>>>
>>>> Yixun
>>>>
>>>>         
>> What I mean is that the process holds all the column, but only part of
>> the vector, how to do the multiplication?  It needs  communicate with
>> other processes. Right?
>>     
>
> Yes. This is explained fairly well in Yousef Saad's book "Iterative Methods
> for Sparse Linear Systems".
>
>    Matt
>
>   
It means that it is better to set n to be equal to the column to avoid
the communication. Right? If it does, I am confusing why we need this
parameter. Just for dealing with very large vector x?
Thanks.!