BS ALERT: "Temporary Tablespaces Like Large Blocks"
The myth is still alive ...
http://dba.ipbhost.com/index.php?showtopic=1512&st=15
"To answer your questions, in general, DSS and OLAP databases (those characterized with lot's of full scans) might see a reduction in consistent gets with larger blocksizes. I have some client's use a db_cache_size=32k so that TEMP gets a large blocksize, and then define smaller buffers to hold tables that experience random small-row fetches."
If the clients are using a default block size of 32kb to do that, Don, I hope it wasn't based on your advice, because the blocks size of a temporary tablespace is irrelevant to the size of i/o it uses, because that is governed by a hidden parameter. The is no logical i/o involved, it is 100% physical.
Better get your excuses lined up for when they find out that this is a Big Oracle Myth, Don. Not only are you giving bad advice to people for free on your forum, you're actually charging your clients for advice that will make their database more complex to create, more difficult to manage, probably more prone to bugs, and with no performance advantages whatsoever.
Mike Ault has excuses ready, by the way ... he's just standing by a friend. http://www.blogger.com/comment.g?blogID=11462313&postID=111369118788103463
Comments welcome. As always.
posted by David Aldridge @ 10:30 AM
8 Comments:
I take it you didn't get the job working for Don, then ;)
I am now learnt yet another thing from Don - a new definition of "anonymous" meaning a person who validates that they answer the correct question by asking for clarification
I don't think I understand the "anonymous" thing at all.
On Don's response in his forum - ranting about anonymous posters - odd, but I can see your name there
Ah, well DKB pointed out to me in an email an unpleasant anonymous posting to Robert Freeman's blog, and he's suffered from such things himself before of course. So I understand why he's sensitive on the subject, but not why he's being sensitive about it to me.
Oh, I was WAITING for this! Saw the post last night, and wondered how long it would take to escalate. I guess you are going to see yet another Burleson article get changed - the one that invites people to the forum to get sage advice from folks like.....Mr. Rude And Obnoxious, David Aldridge! Man, I feel your pain. It is why, even though I am no longer IP blocked to Circus Maximus, I won't post there again. You have been more than gracious, and once again since you offer better advice than THEY could, you get trashed. Something about schoolyards and bullies comes to mind now....if I could ONLY remember where I saw that.....
David,
It's funny as I tried to post a reply to DKB's post about you being rude and whatnot for seeking clarification on the OP's intent and I found that I'm not able to log on to this forum from my work machine.
Will have to do that when I get home.
Jeez, some folks just need to be hung upside down and be spanked senseless.
What's fascinating is the complete DKB's complete misreading of the situation, then denial, then threatening LIBEL to some posters! And then MORE b.s. links about people proving "32kb blocks are best for indexes".
What an ass.
Stu,
The big issue for me is this dichotomy over the concept of "proof by script".
I can find no means of reconciling Don's recently-yet-frequently stated belief that "performance proof by script means nothing" with this belief that a very simple script can "prove that indexes love large block sizes".
Now I happen to believe that you can make performance proofs with a simple script -- I can prove that a full scan of an index-organized table can use multiblock reads, for example. I don't believe that all scripts prove something though, and I don't find the Schumacher script very convincing.
But how does Don reconcile his thoughts on this?
Post a Comment
<< Home