Shortly before the conference, Oracle started to publicly refer to a new version called Oracle9i. Originally rumored to be a renamed version of 8.1.7, its projected features define it as a separate major release.

What’s different in Oracle9i? Product manager Mark Townsend presented a high-level view of changes coming out in 8i. The product is due in the first half of 2001. Since Oracle’s fiscal year ends at the end of May, I’d expect a production shipping announcement by May 31, 2001.

Some of the advertised features:

• Oracle Data Guard, which extends physical standby databases to logical standbys. Redo log entries will be converted to SQL statements and applied in the standby, which means you can have a continuously open standby database that has different indexing structures than the source database.
• Support for online table management. Alter a table that’s being queried by putting the database into a Rapid Quiesce mode.
• Precision repair. Recover just the blocks that are corrupted. Recover past corruptions in the redo log (!).
• Row level change history. Reconstruct data as it existed at a prior point in time. Get a complete history of row-level changes. It seems to me that this would require either a large SGA or the database running in Archivelog mode, but I'm not sure since he just skimmed this.
• Cache Fusion, an extension of OPS that avoids disk pings via a shared cache architecture.
• Query execution time estimates, enabled via a history feature in the CBO that reports on CPU and memory usage. I’ve tried this before, and it’s not simple. It’s also impacted by environmental conditions you have no control over. The same query executed at two separate times may run in vastly different times, and two queries with different costs may run in exactly the same amount of real time. This is the feature that I believe will be the toughest to implement reliably.
• Single-authentication sign-on for three-tier environments. More about this later on in this set of notes.
• Full select trail auditing – what did each user query, and what were their bind variable values?
• Support for ANSI CASE statements and join syntax.
• Scrolling cursors! Go to prior records, first record, last record.
• Pipelines result sets, which enable you to pass cursors to/from procedures.
• The end of rollback segments. Install an Undo Tablespace and forget about rollback segments via the Oracle Managed Files option.
• Resumable statements – long-running jobs that encounter errors will suspend and can be restarted later.
• Database block size can be set at the tablespace level, lending support to transportable tablespaces.

Of course, this list may change significantly over time. I did not see any 9i demos during the week that ran on anything other than Powerpoint.

iFS is still a head-scratcher. In its current implementation, all of the data is stored in BLOB datatypes inside the database. I strongly prefer a BFILE implementation, with the files stored external to the database (that is, use a relational database to store what it's designed for, relationships) but that’s not currently available. In the BLOB implementation, users can consume lots and lots of space in your database, and nothing prevents them from storing multiple copies of MP3 files in your system. How do you perform capacity planning? If the front end is not simplified, how do you prevent people from using the versioning options incorrectly? How do you save space with this option, since people are likely to keep a copy of the file outside the database too (and you have to do this anyway when you check out a versioned document?)? I came away from the iFS hands-on session with more questions than answers.

The tenor of the keynotes doesn’t change much from year to year. Bashing Microsoft (or HP, or IBM, or any other competitor) doesn’t help any users constructively resolve any technical problems they’re encountering. And jokes about politicians are out of place. Sun's Scott McNealy, when not slamming politicians or one vendor or another, spoke about his plan to have the servers Sun makes auctioned off at the end of each day, hopefully guaranteeing Sun a good profit and the customer the best deal. While that seems interesting, and a potentially good model for a maker of commodity machines, I’m not sure it works well for a vendor of large-scale systems. Such vendors should be partners with you in the development of your technical infrastructure, and an auctioneer is not a partner. What would Deming say? What impact would this practice have on ISO-9000 and TQM-focused companies? Similarly, Oracle's pricing approach in recent days has reflected much more of a vendor-customer relationship than a partner-partner relationship. This approach seems to open up those customers to competitors who focus on partnering.

You can get lost in the exhibit hall. It stretches for about two blocks, and it’s jammed. There are huge vendors with elaborate displays, and smaller shops that dot the landscape. It is a village unto itself. Given the crowd size and the hall size, getting a chance to really look through a product’s features was difficult.

This year, the up-and-coming booths (those that are bigger than they were last OOW) belong to the ASPs and the storage area network (SAN) companies. As data increases in volume, and the systems increase in complexity, companies are increasingly looking for outside help to take over the management of data and storage. In that regard, Oracle’s targeting of Oracle9i at ASPs and B2B service providers appears to be a good move from the marketing perspective.

Meanwhile, the sensory deprivation sensation within the exhibit hall hit new highs. One vendor featured loud tap dancing – you have to feel sorry for the vendors who paid to be in the exhibit hall and ended up next door to such a racket. I guess the Lenny Kravitz concert during the conference party really wasn’t that loud by comparison.

The single sign-on (SSO) feature of 9i’s Oracle Portal product set (A quick aside: OAS, which used to be OWS, was renamed iAS and this week has been rechristened OiAS. They’re running out of vowels.) was featured during a presentation today. Oracle Portal relies on a Login Server (LS) for authentication – the LS authenticates a user and then returns a SSO cookie to the user. A subsequent authorization may give the user an application security cookie.

All of this is worrisome, since cookies are not the best security method. I can write my own cookie files, or steal yours, or spoof your PC and get a cookie. But then the discussion took a more unpleasant turn.

Oracle Portal supports external applications (they used hotmail as the example, which is ironic when you consider who owns that service now). You can store the passwords for those external applications via Oracle Portal – they are kept in a password store. So,

• what happens when you change your hotmail password every 30 days? The store doesn’t get updated
• what happens if I get into your password store?
• what happens if I get into your login server?
• what happens if I get past the cookie issue and get all of your passwords for all of your external applications?

I had a very interesting conversation with a similarly concerned woman who is involved in performing security audits for highly secure sites. Chance meetings with professionals who share your concerns over technical directions are one of the best benefits of attending these gatherings.

On the final (extra) day of the conference, Jeremiah Wilton of Amazon delivered a keynote that featured a demo of a standby database failover. Three other folks from Amazon.com were featured on the panels that followed, and the panels were well attended.

ECO is not happening next spring, so the next major conference is IOUG in Orlando in April/May. Next year’s OOW conference will be December 2-6, again in San Francisco.

10i anyone?