If I was enthusiastic about SIOUG 2007, I can’t say the same for this year event, SIOUG 2008 (21-24th September 2008).

It doesn’t bother me so much the place of the event (yet again, standard-unimaginative-dull Bernardine near Portorož…if you like tons and tons of concrete along the sea coast, you’ll probably enjoy both, Bernardin and Portorož.:-), as the noticeable lack of prominent foreign speakers. Yeah, I know – SIOUG 2007 will be hard to beat, especially when SIOUG is competing with Oracle Open Word, which will take place in the same week of September (grin).

With one or two presentations that I would like to attend on Monday and Tuesday I could not justify wasting two full days there – I sincerely tried to put together a full day agenda for the Monday and Tuesday, without a success. The content that would sparkle my interest just isn’t there. The last day of the conference is more promising. If I’ll attend SIOUG 2008 at all, I’ll book the last day of the conference – to listen to Jože, Riyaj and Jurij.

During the yesterday session at seminar, someone from the class “surprised” Mr. Craig Shallahamer with a question about a detail from the slide (on page 128, picture 2); “…how does Oracle recover “In-Memory Undo” blocks in case system crashes at T3″.
It’s one of those stressful moments for the intructor, when a good instructor if it’s not sure about the correct answer will admit that he/she doesn’t know it (or can’t recollect the info) and that he/she will investigate the subject further, rather than speculate on the subject. Of course, Mr. Shallahamer is a superb instructor with enough mileage that he gracefully admitted that he doesn’t know the answer (imho, I would say that he was at the moment of thinking about the answer focusing too much on the slide – after all, it was late afternoon and we all had our heads full of information). (Btw. if you’re not familiar with Oracle IMU, Mr. Shallahamer released a white paper recently on the subject – recommended!)
Thinking about unanswered question on my way back home, I came to the conclusion that Oracle most likely protects the in memory undo (IMU) with the same mechanism as undo stored in segments, with the redo logging. So, I believe the redo log keeps all the information needed to “redo” Undo-Information in memory, that can then serve for undoing changes after the crash. Simple answer for a “tricky” question. [Note added: My assumption was wrong and if you read an answer from Mr. Shallahamer that I posted in my comment, you’ll see why]. Well, but is my conclusion correct? I couldn’t find any useful Metalink note about IMU. Am I lost?
Here we come to the subject of this blog, why not ask Google Patent search engine that is specialized for the patents search and presentation to find the document related to patent# 6981004 (Oracle IMU). This is such a great tool if you’re interested in all those nifty details – Oracle can choose what information to include in Metalink documents and what to leave out. Guess what, they really don’t have a choice in a document that is a baseline for the patent :-). They must cover all the details to prevent others to copy-cat + “patch” the solution. On the con side, some of the patents (and yes, IMU is one such patent) are documented with so much details that documents are extremely heavy to digest. You have been warned, it’s your choice, after all!
(A side note: I’m still digesting the IMU patent pdf, so I can’t confirm (nor deny) my conclusion/speculation yet, it well might be that Oracle uses some other nifty trick to protect IMU in case of instance crash – perhaps some other patented algorithm…)