Miracle database forum day 2

3 06 2007

It is apparently more difficult to get up in time as a conference continues. There were some people really late, but everybody got into Edinburgh castle again to listen to another full day of back-to-back presentations. Which is one of the reasons why it is so difficult to get out of bed. Anyway here is a short summary of what was presented.

Mogens Norgaard (You probably Need RAC) has kicked some sensitive knees (or other parts of the anatomy) with his (in)famous paper about RAC within the oracle corporation. He is still convinced most companies are better off without RAC and is even in the process of getting one of his customers to stop using RAC. The customer has previously been convinced they needed RAC and accepted that it took at least 8 to 9 months to get the RAC production environment anywhere near stability. But after about two years of working with it (and experiencing a max-out on CPU power with 4X4 Opteron processors, where a maximum load of 20% was expected), other (less complex) configurations are considered. If you want to get it stable sooner, you will need a production alike (almost exact) testing environment.

Daniel Fink (No Mogens) did get us thinking. He told us that we just fix what we know, we don’t know any other way. But you can’t resolve technical issues by staying technical. It’s not about the technical issue, it’s about response time as perceived by users. As a consequence of this, almost all performance tools ‘suck’. They only show the issue’s the developer of that tool knows about. Nothing else. Therefore you miss an important part of the big picture. Daniel pressed us there one part that surely should not be missed. That is that we should be looking for SQL developers (or SQL generators) that are using bad SQL. Try to educate both!

Christophe Dobroschke. AMD is the main sponsor of the event and Cristophe had the difficult task of interesting all those oracle adepts in the AMD processor all should be run on. And after sponsoring this event, they will have moved a place up on the list of preferred suppliers for many a representative.

Jonathan Lewis (Understanding statspack data). It is always with mixed feelings I see presentations by Jonathan. On the one hand it is always useful to follow his ideas and he is a good presenter. But on the other hand it becomes so clearly obvious there is so much you don’t know about Oracle that he does. Jonathan took us through several examples of statspack data on systems ‘with a problem’. He showed us the lines of attack he did follow and urged us to read statspack reports of correctly performing systems. If you practice it’s much easier to spot the things that are different. A presentation that was both discouraging and invigorating.

James Morle (Brewing Benchmark) is the founder of Scaleabilities and author of Scaling Oracle8i. James therefore sure has some right to talk about scaling. He took us through an Open Source project he has started called SimOra and showed us how he is doing it currently. The software resulting from the project is able to generate test data based on a trace file (mainly a 10046 trace) of activity that has taken place on the database. That way a user can use an application to get some data in (or out of) the database and representative test data can be generated from that. The result of the generation is a tcl based script that can edited and expanded which provides a flexible and accurate way to load test a database.

Frits Hoogland (Tuning Multi-tier J2EE applications). As has been discussed by several of the other speakers before him, Frits focused his presentation of many of the things that transpire outside of the database. More specific, he was looking at the stack ‘above’ the oracle database; the web server (mostly Apache) and the application server (Oracle iAS, but Tomcat/JBoss as well). He showed several places Apache could be configured wrong, but concluded that the best place to focus on, would be the application server. He has been looking for tools to make it possible to measure the performance of the total stack, from webserver through the application server to the database and back again. He did not find any, but as the main bottleneck is with the application server (the database server we have all tuned excellently of course) he did find a freeware tool called Glass box. Very useful to have that in your toolkit.

Bill ‘Shrek’ Tater (It’s not paranoia if they’re really out to get you). Bill is from the The Austin (Texas) Independent School District and initially you would not expect someone working there to be concerned with security. Until you think about it. Every grade given is administrated in some database, so every student is interested in getting his/her hands on unlimited access to the database. Bill had some sane advice about using all the possibilities oracle provides out of the box, with schema’s having create table rights, other schema’s that only have access to the data in the schema’s and still other schema’s that have no rights at all but get them granted through roles within applications. Seems simple, but everyone is so busy preventing access on any level that the simple things tend to be forgotten. Unfortunately, Bill had lost his presentation due to a faulty computer, so we have something to look forward to for his next appearance.

Graham Wood (You probably don’t need SQL*Trace). As one of the architects of ASH (Active Session History) and AWR (Automatic Workload Repository), which are both new in 10g, James was showing us many of the virtues of both. And his conclusion is that you won’t ever start an SQL*trace again if you use these tools. Especially ASH is a lightweight, non-intrusive tool that provides performance insights after the fact with virtually no (CPU) cost. Measurements show that there is less than 1% overhead on a database when both are activated. Of course, ASH and AWR are separately licensed features but if Graham had his way they would become a free available part of the next major release. Oh well, we can always hope.

Tuomas Pystynen (RMAN incremental backups) showed the capability to recover a database is one of the major virtues of a good DBA (you won’t need DUDE then). So a presentation about rman is to be part of a good conference. And Tuomas did indeed present excellently. Giving the overview that is needed, explaining some definitions and get down to the gory details that every DBA should know. Reading through the manuals you get into so much detail, you easily forget about the big picture. The big picture is clear again!

And that was a full day of conference. Learned a lot, I must say.

We had some drinks in the oldest pub in Scotland (The Sheep Heid Inn, but I have seen more ‘oldest’ Pubs in Scotland though this one looked genuine) and went to bed early. For some that meant early in the evening, most decided to make that early the next morning.




5 responses

5 06 2007

Thanks Erik,
That is a proper Report!

Being lazy, I’ll link straight throught to this.

5 06 2007

No problem, Pied. As long as you spell my name correctly next time. And Yes, I’m a nitpicker on that one 🙂

30 11 2008

Hi. I regularly scan this forum. This is the oldest together unequivocal to ask a query.
How numberless in this forum are references Nautical port behind, disingenuous users?
Can I bank all the communication that there is?

30 11 2008


I don’t quite understand your comment. But if you are asking if you may copy the contents of this and other blog post, you may. Provided it is for your own use and not to publish it further.

4 12 2008

I am here at a forum newcomer. Until I read and deal with the forum.
Let’s learn!

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s

%d bloggers like this: