Re: how to place a table in memory

David,

This is not from my original research, this comes from course materials (someone else's that might reply on this post, perhaps). The premise was that a full table scan would not grab the entire extent if a block in that extent is already in memory. The larger the extent size, the higher the chance that a block from that extent might be in memory, thus preventing the entire extent from being fetched from a single io request (it would take at least one more io request to grab the remaining blocks).
I wouldn't call it disrupted, it would be that you see some behavior of less than the full multiblock read count's worth of blocks read from a single extent that had no free blocks.

back when I was testing dbfmbrc efficiency, I was using a stand-alone system and obtained nearly 100%. I could see repeating that exercise having certain blocks in v$bh and observing the behavior ... but not tonight.

8.1.7 to 10.1 migration tomorrow morning. big one.

-bdbafh Received on Fri Feb 11 2005 - 22:05:27 CST