Friday, June 04, 2010

Concepts Guide: 9/27 - Process Architecture

"Figure 9-1 can represent multiple concurrent users running an application on the same computer as Oracle. This particular configuration usually runs on a mainframe or minicomputer."

Wow, this section of the documentation must have been recycled for a number of years. =)

Good pictures, descriptions of various processes.

In general, I like the "See also" sections, but I wish the link would go directly to the relevant section of the reference, instead of the top-most TOC page.

This section confused me:
"When a checkpoint occurs, Oracle must update the headers of all datafiles to record the details of the checkpoint. This is done by the CKPT process. The CKPT process does not write blocks to disk; DBWn always performs that work.

The statistic DBWR checkpoints displayed by the System_Statistics monitor in Enterprise Manager indicates the number of checkpoint requests completed."

If The CKPT process is responsible for updating the datafile headers and DBWR is responsible for something else (writing blocks to disk), why is the statistic called DBWR checkpoints? That is quite misleading, and perhaps leads to the confusion that spawned the warning about the DBWR in the first place. =)

Both PMON and SMON "check regularly". What is "regularly"?

While there are a lot of good ideas imbedded in Oracle, it is surprising that some of the still have such an antiquated and/or obfuscated interfaced. For example, the job scheduling system. The job queue processes are quite cool, but using them is a pain in the arse. The EMGC GUI is not too shabby, but what really sucks is the API; what about a simple API for those of us who do command-line work? VPD and Streams are the same way (have not yet seen any GUI for VPD). At least Shared Server is a little easier to grasp and implement, but it is still very easy to shoot yourself in the foot.

In terms of performance in the context of Shared Server, would not immediate results from FIRST_ROWS_N operations be queued as well? So it would be possible that queued results would actually return slower than when using a dedicated server?

Overall I found this chapter disappointingly light on details, or examples for that matter. I would love to see the program flow, end-to-end, of requesting, establishing, executing and concluding a transaction. Likewise, the last few sections (under "The Program Interface") don't really say much at all - it is most useful as a dictionary or appendix, nothing really that describes what things are or how they work, or the role they play in the larger picture. I mean, they do a little, but not a whole lot.

1 comment:

Joel Garry said...

It's in the 7.3, (...), 10.2 and 11.1 docs at least, but not the 11.2 docs. I guess the mainframe is finally dead. Don't tell IBM.

In the 7.3 docs it says: "The statistic DBWR checkpoints displayed by the
System_Statistics monitor in Server Manager indicates the number of
checkpoint messages completed, regardless of whether the CKPT
process is enabled or not."

So perhaps you'd be less confused if you disabled the CKPT process and used Server Manager :-)

I agree, we should be able to drill down and see in more detail how things work, without having to do the potentially misleading poking at a black box. But then we wouldn't be DBA 2.0, happily skipping along with pretty pictures.

word: inksiv