MySQL Forums
Forum List  »  Partitioning

Re: partitioning by highly skewed data
Posted by: Mikael Ronström
Date: June 16, 2005 11:57AM

Hi,


Hiroshi Takenaka wrote:
> Will it be possible to partition one specific
> partiton further?
>

Subpartitioning is possible for RANGE and LIST partitions, the
subpartitioning is then HASH partitioning.

> For example, we have a 1TB table. We want to
> partition it,
> and spread over 10 * 140GB disks using one
> column.
> But, unfortunately, 20% of records have NULL on
> that column.

Actually I was planning to only allow non-NULL fields in the partition function but after
checking things up it seems that if one field is NULL then the partition function will return 0
so it should actually be possible to support NULLable fields in the partition function. I'll add
that feature immediately.

And yes it is possible to subpartition. However all partitions must be subpartitioned so one
needs some care in defining the subpartitions such that they spread evenly among the disks.


>
> It would be nice, if we can partiton those records
> further
> by using another column or some random value.
> I think it should be easy if user can define and
> use
> a custom partitioning function.
>

The user can define his own function for both the partitioning function and the
subpartitioning function but the function cannot be random since then it would not be
possible to find the data again unless one scanned all partitions.

Thx for the interesting question and please ask more, it does clarify the requirements from
real applications out there.

Rgrds Mikael



> Regards
> Hiroshi

Mikael Ronstrom
Senior Software Architect, MySQL AB
My blog: http://mikaelronstrom.blogspot.com

Options: ReplyQuote


Subject
Views
Written By
Posted
Re: partitioning by highly skewed data
2528
June 16, 2005 11:57AM


Sorry, you can't reply to this topic. It has been closed.

Content reproduced on this site is the property of the respective copyright holders. It is not reviewed in advance by Oracle and does not necessarily represent the opinion of Oracle or any other party.