Why did rebuilding index fix our cache buffers chains latch issues

I know there are some Oracle performance tuning experts that post on this group and so I thought I would post this problem to see if you have any ideas.

Our datawarehouse suddenly started getting all kinds of cache buffers chains latch waits. We went through the standard scripts from Metalink and books we have that say to look at v$latch_children and x$bh. We found a hot block on an index. We rebuilt the index. All of the latch issues went away.

But why did they? I've read about cases with increasing key values (like from a sequence) where you will get a hot index block. But these cases involve deletes on the lower values of the key. This is that kind of index - a primary key which comes from a sequence. But, as far as I can tell, records are never deleted from this table. So, the index shouldn't be unbalanced in terms of having a bunch of leaf blocks that are mostly empty. And, rebuilding the index really shouldn't change much about its internal structure.

The only remaining clue that I have is that the index shrank to about 10% of its size before the rebuild. Also, its last ddl time was two months ago, so maybe it was rebuilt it some wierd way in March and that caused it to have a hot block? Then our rebuilt made it normal again? I'm really puzzled.

I may not have enough information to go on here, but I thought I would put it out there.

Thanks,
Bobby Received on Wed May 24 2006 - 11:01:32 CDT