Database, heal thyself
IBM DB2 8.2 works with admins to anticipate problems
- By John Breeden II
- Apr 28, 2005
It's pretty neat to think of a computer or a server healing itself, and that is exactly what IBM is going for across its product line. IBM's DB2 Universal Database 8.2 incorporates some of these self-awareness features, which puts it ahead of most other databases in this category.
A secret shared by good database administrators is that one of their main missions is to streamline database performance. This normally entails a complex process of examining all queries going through a database to nail down trends.
They then can keep the database's searching and execution plans for the most requested information on file, so the CPU does not have to waste time forming a new plan each time for a common elements search. In this manner a good administrator might lose a lot of sleep, but the clients will end up saving a lot of time using the database.Self-awareness
DB2 does this automatically as part of its self-awareness programming. The program constantly scans all the queries made by users and looks for common elements. When a new query to the database is made using a common element (we used an expense-tracking model for our tests), the system simply follows a known execution plan instead of having to recalculate a new one.
Using a fairly large database, though small for many government agencies, we ran some preliminary searches for reports relating to entertainment expenses. Those searches took between 20 and 25 seconds to complete. But once the system realized that this was a common search, future searches for the same or similar topics took between 15 and 20 seconds, which amounts to significant timesaving.
When you consider that some government databases can reach terabytes, the timesaving for queries could be minutes or longer per search, which quickly adds up.
Still, that process is more like self-tuning than self-healing. But DB2 can also heal itself, at least from common problems.
Administrators can probably rattle off common database problems, or their indicators, in their sleep. You have the deadly High Availability Disaster Recovery log delay, which always seems to spring up in a crisis. And you also have deadlock rates and lock escalations, which are accurate indications that something is terribly wrong with the database server. DB2 can help heal these problems.
But before you fire your admin, know that this new technology does not work without the administrator's assistance.
What the admin does is create rules in the form of script files that trigger when conditions such as a rapidly escalating lock rate are met. Without the administrator, the database will know it is sick but will have no idea what to do about it.
One of the best things an administrator can do is make use of the client redirector feature. What you do is configure the High Availability Disaster Recovery logs to copy themselves to a separate server. The backup server can then apply the logs. You can set the backup server to synchronize with the main server in several different ways, depending on how closely you need the two to match.
When the main server goes down, clients are redirected to the backup server and likely won't even know a failure has occurred. We've seen this technology before, but in the past it required use of a third server to monitor the health of the primary and the backup. With DB2, the redirect information sits with the client, so the need for the third server is eliminated.Role reversals
You can hand-trigger a rollover through the easy-to-use interface, or set up a script to do it for you when poor health is detected. Clever administrators can even have the broken primary server reboot and become the backup server once it comes back online. Simply switching the roles of the two servers, which can be done in just a few clicks, or automatically using scripts, quickly accomplishes the task. Keep in mind, though, the HADR management features cost extra'$6,000 if you use the DB2 Express or Workgroup editions. It comes standard in the Enterprise edition.
The latest release of IBM's Universal Database shows what the future of computers might be like, with administrators and their servers working together to fix problems.
The interface right now is still a bit unforgiving, with the administrator forced to create a triggered script for every possible failure event. But the auto-sensing work performance modes show that it is possible to do some advance tasks without user intervention. So as computers grow smarter, humans can concentrate on other pursuits.
DB2 would be a worthwhile database product for any office, especially those that can't afford to lose their data, even for a couple of hours.