Miracle database forum day 1

31 05 2007

We all had a good nights sleep, allbeit a short one and started off with Doug Burns about Dtrace. I had seen some posts by Doug about Dtrace but it did not catch on until today. He told enough about it to start to think about uses for tracing when the standard database tracing is not enough. Dtrace is an open source tool mainly found on Sun Solaris although there are some ports. It is a scripting language that can trace any part of a process running as long as you know the name of the module that you want to look to. And he showed some ways to get to know the names of the modules. Very light footprint so it is a tool you can even use in a production environment. Not something you do lightly, but enough interesting hooks to want to try that sometime.

Carel-Jan Engel had a novel use for his beloved data guard environment. When you specify a delay in the apply of redo log files on a standby database, you can use the standby as a means to recover from logical errors (user dropping a table, you know the kind). As long as the drop of the table has not arrived on the standby, you can temporarily stop applying redo logs, open the standby in read only mode, read the contents of the dropped table through a database link or export and use that to recover the table on the primary. Then you can restart the standby and start applying the redo logs again. With your dropped table in place. There are other ways to do this of course, some cheaper. But it is a nice example of thinking outside of the box.

Alex Gorbachev, a Russian now living in Canada showed us the results of his project in testing a hardware cluster solution for a data center and hammering it with a load of data. The hardware is called LiquidIQ The most interesting about the presentation to me, was that he used Hammerora to load the oracle database on the hardware with lots of data. Another tool to look into when I need to stress test a configure database. Now looking for some project that is able to sponsor that.

Joze Senegacnik from DbProf showed everything about the automatic sizing of work areas (for sorting, hash joins ….) that I wanted to know about a year ago when we were struggling to get a database and an optimizer to use as much hash joins as possible. I know why it was so difficult now but at least now I know it for a future use.

Michael Moller was asked by a customer to measure the benefits of buying the partitioning option for a table that was essential for the customer. Finally a customer that had enough sense to try to determine the effect of an option before believing the consultant that says it will be beneficial. Although I do not agree with his conclusions that partitioning is not to be used, it was refreshing to see the methodical way he approached the subject. A pity for the customer they did not use other partitioning schemes. I’m positive there is more to gain to get there.

Anjo Kolko has tought of lots of meanings for TBD, besides the obvious. His presentation was mainly an essay to start (DBA) people to think about what they are doing with the database. Tools are surprisingly simple: stats, rule, engine, advice. This goes for statspack, ASH, YAPP, even your own tool. Think about it.

Piet de Visser (Databases, Systems, Relationships (KISS)) held a plea for simplicity as opposed to complexity. He noticed there has been some consolidation of late resulting in reducing the number of databases. This lead to all kinds of different data into one database on one server. The process has been difficult due to requirements by vendors, suppliers and application owners. But it succeeded in many cases. And all this leads to simpler maintenance. The ideal situation would be one system; one database. Utopia of course, but it happens now and then. Currently ‘the one database’ is evolving into one database on many systems (a.k.a Grid). This adds a lot of complexity that is opposing the simplicity strived for in consolidation.

Julian Dyke (Flashback logging internals) dives another time into the deep dungeons of Oracle. Trying to get detailed knowledge of how Flashback logging is done. It is based on redo logging of course, but because of its nature it is different, very different. It looks a lot like before images of blocks are saved into the flashback area. Julian found out about a lot of undocumented commands that work on flashback logging and by their nature and names reveal many of the secrets of the flashback area. But a lot of detective work still has to be done.

Kurt van Meerbeek (Dude, where’s my data?) is the author of DUDE, Data Unloading by Data Extraction. Extract data from datafiles of a database that cannot be brought online any more due to loss of hardware, software or even the SYSTEM tablespace. Most people have seen this some time in their professional life; far beyond the ‘Oops’ moment, when you really start to panic. And of course not having a backup that vaguely usable. DUDE only reads data! It cannot change anything. It works on oracle data blocks and deduces the contents of a datafile, even if the system tablespace with the catalog is not available. Oracle (consultants) can do that, but this is very expensive and often not fast enough. DUDE is.

After all this we could do with some refreshment and some talking about other subjects. We could wander around Edinburgh Castle and see the Scottish crown jewels and the Stone of Destiny. A pleasant evening followed with a good meal, excellently topped off with a talk by Dave Ensor, former and now retired colleague of both Graham Wood and Jonathan Lewis. Some Gaelic music after that and anybody that was not tired yet could (and did) continue with some live, more modern music in Whistlebinkies.
Certainly a good day.




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: