Query for row by rows-before using seq number

Question

Hi,
I have a table of CatalogAuthor records
long : id
varchar : author
long : seqnum
I have made the table so that initially, if there are 10000 unique authors, then the rows will have sequence numbers from 1 to 10000. As authors are dynamically added, they are given the sequence number of an existing row in the table, i.e. the nearest row whose author comes before the new author. As time goes there are more and more duplicates, and at a certain point, the table has all its sequence numbers reassigned to be monotonic increasing by author name order.
We want to use this table to build a tree of about 10 nodes per parent, culminating in author names. So to expand the tree at the first level, we could divide 10000 by a suitable power of 10 to get the groups of sequence numbers. However, with duplicate seqnum, there may be 14365 rows so ideally, to overcome the problem of clumping, I would do some kind of query to identify the 1436 record in sorted order by sequence number, the 2872 record by seq number and so on.
What query would that be? With appropriate indices, would it still be slow? Would I have to do 10 of these queries to expand one node with 10 children? Is there a better design to accomplish the same goal of exploding a balanced tree by author name sort order, culminating in a leaf of one particular author?
Andy

Answer 1

Hi, Andy,
user9990110 wrote:
Hi,
I have a table of CatalogAuthor records
long : id
varchar : author
long : seqnumYou can't be serious! The LONG data type is nothing but trouble, and completely unnecessary. You must mean NUMBER instead of LONG.
I have made the table so that initially, if there are 10000 unique authors, then the rows will have sequence numbers from 1 to 10000. As authors are dynamically added, they are given the sequence number of an existing row in the table, i.e. the nearest row whose author comes before the new author. As time goes there are more and more duplicates, and at a certain point, the table has all its sequence numbers reassigned to be monotonic increasing by author name order.
We want to use this table to build a tree of about 10 nodes per parent, culminating in author names. So to expand the tree at the first level, we could divide 10000 by a suitable power of 10 to get the groups of sequence numbers. However, with duplicate seqnum, there may be 14365 rows so ideally, to overcome the problem of clumping, I would do some kind of query to identify the 1436 record in sorted order by sequence number, the 2872 record by seq number and so on.
What query would that be? With appropriate indices, would it still be slow? Would I have to do 10 of these queries to expand one node with 10 children? Is there a better design to accomplish the same goal of exploding a balanced tree by author name sort order, culminating in a leaf of one particular author?I'm just curious, why do you need this? What will this tree structure do for you that an index on author won't do?
I don't know of any way to automatically keep a tree balanced every time DML is done.
You could periodically re-balance the tree; that shouldn't be too hard. Let's say you want each parent to have up to r children. Use the analytic ROW_NUMBER function to assign unique numbers 0, 1, 2, 3, ... to each row, in order by author.
The row numbered 0 will have no parent
For all all rows numbered n (n > 0), the parent will be the row numbered FLOOR (n/r).
Whenever you re-configure the tree, use a MERGE statatment. An index on author will probably make generating the ROW_NUMBERs faster.
If you need help, post CREATE TABLE and INSERT statements to show the table as it exists with the tree out of balance, and also show how you want it to look after balancing. Use a small value of r (maybe 2 or 3) to keep the sample set small.
Edited by: Frank Kulash on Jan 12, 2011 12:41 PM