Tuesday, April 29, 2008

How not to do it...

Two years ago at Oracle Open World, I delivered a "worst practices" talk - how not to do stuff.  I used the word "probably" a lot, for example "you probably don't need to use bind variables" (there were slides stuck in this slide deck throughout that said 'hey, these are all the opposite of reality - just in case you are reading this - they are not true!').

In the same genre - we have "Top Ways How NOT To Scale Your Data Warehouse".  This comes from the Structured Data blog penned by Greg Rahn - that article as well as the others make for some really good reads.

While I'm pumping that blog - another one to definitely look at is Richard Foote's blog.  He's been undertaking the destruction of many a myth regarding indexing (like 'indexes like large blocks', 'rebuild when height hits N', 'Separate tables from indexes for performance' and the like).  Very easy to read, very enjoyable to read.

If you use multiple computers like I do - you might be interested in http://www.instapaper.com/ - find something you like, but don't have time to read right now - save it for later.  Very nice.



Anonymous Anonymous said....

instapaper... and "they" know what you want to read?

like that service which prints from a
blackberry via bluetooth by sending your document to "them", "they" convert it for you, send it back as .pdf to get it printed...

or google cal/apps/mail

I would never store my docs/mail/links/whatsoever on someone elses service.

make yourself a webdav drive, make it with ssl enabled on your own web/mail/...server, use something like scrapbook to save what you want to read

just my 2cent of privacy

Tue Apr 29, 03:33:00 PM EDT  

Blogger Ontario Emperor said....

For things that I don't need to keep private I use StumbleUpon.

Tue Apr 29, 03:51:00 PM EDT  

Blogger Thomas Kyte said....


they know that xabad32fa@yahoo.com reads stuff.

they do not know what "I" read necessarily.

sort of like the grocery store knows what "I" eat.

and best buy knows what electronics "I" own.

and amazon.com knows what I read (they actually do, because they can actually identify me)

I use online services heavily.

webdav - not even remotely practical in the environments I travel in. Not even very appealing - even if I used one computer, I would not drag links into a folder (been there, done that), I'd rather just hit a button and be done with it.

Tue Apr 29, 05:32:00 PM EDT  

Anonymous Scott Mattes said....

the instapaper site is way, way sparse on what they do.

since i don't like signing up for something just to get a tour that should have been offered to anyone (it seems like a gimmick to artificially inflate user numbers to me)...

could you describe how instapaper is different from del.icio.us or google notebook?

thank you.

Wed Apr 30, 10:41:00 AM EDT  

Blogger Thomas Kyte said....


... the instapaper site is way, way sparse on what they do. ...

that is why I like them. It does exactly one thing

a) it remembers a url for me at the press of a button

Later, I go to instapaper and they have two lists

1) items I've asked to save but haven't looked at

2) items I've saved and looked at

and that is it, period, nothing else. No images, no flash, no ads, just a list of pages. Period.

Very unix like :)

Wed Apr 30, 10:47:00 AM EDT  

Anonymous Anonymous said....

You never said what you choose as topic for coming OpenWorld presentation from the ideas people posted in your previous blog entry?

Eager to know what you are goin to prepare on/select?

Wed Apr 30, 03:15:00 PM EDT  

Anonymous Anonymous said....

@Scott -

You may benefit also from seeing a psychotherapist to learn how to deal with your paranoia. You sound like my crazy neighbor who always says, "I swear they're out there and just because you can't not see them doesn't necessarily mean that maybe they won't not be coming for you."

Breath in, breath out. "They" won't come for you, I promise.

Fri May 02, 12:40:00 PM EDT  

OpenID richardfoote said....

Hi Tom

Thanks for the nice comments, much appreciated. Based on many a reaction to what I've written, many of these myths have a bit of life left in them yet.

Hopefully, slowly but surely, the message is getting through.



Fri May 02, 09:02:00 PM EDT  

Blogger Thomas said....

It's enough to make me cry.

The problem with Oracle "myths" is that one time they were "best practices".

Right now we are in the middle of an exercise to move all indexes to their own tablespace to improve performance and eliminate read head contention.

We are getting rid of our uniform extent tablespaces and adding storage parameters to each table and index so that it will fit in one extent.

We are getting rid of parallelism (Oracle support suggestion). Too many qref latch waits. I suspect that the PARALLEL_EXECUTION_MESSAGE_SIZE is too small but it's at the Oracle default /* sarcasm on */ so it must be right /* sarcasm off */. Parallel_max_servers is 189 but it the database only uses 74 under max load.

We don't use compression.

We are getting rid of partitioning because it takes the ETL too long to load the table. They're loading 100,000,000 row tables one row at a time.

We're setting up new LUNs for the exclusive use of our database but the LUNs are striped across a Network Storage Device.

We generate statistics on all objects daily at 100%. If they're not new then they might be the reason that everything's so slow.

I appreciate the sarcasm in Mr. Rahn's blog but I'm afraid that he overestimates his audience.

When I go to management and say, "Don't do that," they pull out some old best practices whitepaper and ask me to find a best practices whitepaper that says "Don't do that."

I try to keep up by reading blogs, mail lists, books and even (gasp) the Oracle documentation. But management wants five bullet points and a one sentence summary.

A short, simple (management simple) whitepaper of "New Oracle Best Practices" that explicitely refutes the Oracle myths with Oracle's imprimature on it might help.

Talking to Oracle support about the difference between making the symptoms and fixing the underlying problem might also help.

Tue Dec 02, 12:39:00 PM EST  


<< Home