Jeff Jones, director of strategy for IBMs DB2 Universal Database, says IBM is wielding affordability and automation as it prepares for battle in the new world of second-generation databases. As the company prepares to unleash Stinger, code name for DB2 UDB 8.2, he took time out to chat with eWEEK.com Associate Editor Lisa Vaas about the update, where it fits into grid computing, how it differs from Oracle Corp.s offering and how great push-button databases are.
Whats the thinking behind freezing the price of the update at DB2 8.1 levels?
We believe value is the driving force right now, and how best to make that point than to not raise prices and to invest in significant amounts of automation?
What about the specter of Microsoft Corp.s SQL Server, which has traditionally been seen as having a hard-to-beat total cost of ownership?
[DB2 is] closing in on SQL Server. Low-end implementations of DB2 can now give it a run for its money.
Plus, were all kept honest by open source. Free software and cheap software really keep us commercial guys awake and watching what were doing. It forces some rigor on us all.
One of the autonomous features of Stinger were hearing about is the Learning Optimizer feature, which will help the database learn how to speed up command execution and optimize queries. What can you tell me about it?
What Stinger can do … is to continuously look at what it expects to see and compare that with what it does find and analyze where hot spots are and how a database is changing over time and use that analysis to continuously adjust its choice of access plans.
It represents real-time, constant analysis going on in the background by DB2.
That sounds like a potential performance anchor.
Lest we end up eating all performance resources, we use querying sampling to scan the database. We arent actually reading every record in the database over and over.
In LeO, query sampling allows us to work without bogging everything down. Thats a way to get a general idea of what the answer to a complicated question might be. You can get a 6-minute query to run in 20 seconds, and you can get pretty decent accuracy. Its a typical starting strategy in business intelligence or data warehousing environments.
As you refine questions and want absolute accurate answers, you can take query sampling off and let [the straight] query run. But for ballpark answers to questions, its very powerful.
The reason why were so excited about LeO is that theres only one human interface: to turn it on or to turn it off. Thats all thats required to get this automatic performance improvement, without any human investment at all.
IBMs Jones: Stinger Targets Enterprises with Automation, Price Freeze – Page 2
People are referring to Stinger and to Oracles Database 10g as representing the second generation of database technology. What does that mean, and whats motivating this evolution?
One [force driving this evolution] is maximizing the value and minimizing the cost of DBA [database administrator] resources. Which means, with the database technology, more automation, so DBAs can do more in the time they have.
Another force is the increasing requirement for information integration. … The whole service-oriented architecture move … is a third, inescapable force. The notion is that you must be able to provide service that can be understood by other applications, easily contacted and easily worked with in an application way, to invoke wherever they sit over the Web.
At the end of this thinking is that with the notion of a Web server and a database server and an application server, customers are growing more and more impatient with having to think of them separately. Theyd rather spend less time thinking of database servers and more time thinking of business logic and better ways to automate business processes.
Stinger incorporates features to serve clustered environments. What does it bring to the table that Oracle 10g doesnt?
Clustering for HA [high availability] is what Oracles RAC [Real Application Clusters] is about. … Weve taken a giant leap forward in Stinger with automation applied to that HA process.
You can have a cluster of two servers, each with a backup. Thats a total of four servers. Two share the database work; two can be failed over if needed. When I hear Oracle talk about clustering, its generally talking about HA clustering.
Scalability is the ability for database systems to manage increasingly larger amounts of data and increasing amounts of users banging away requesting things.
Because of the architecture with DB2, which is almost antithetical to Oracles architecture, we have customers building warehouses on 150 servers lashed together. One IBM partner in the life sciences got involved with this with Linux. They were building a 1,250-server cluster a couple years ago.
The point is, at least in the high two digits and frequently in the low three digits of servers lashed together is what we see in our accounts.
This second half of clustering, scalability, is the perfect jumping-off point to get to the idea of grid. Grid is the notion of taking both reasons to cluster and doing it in such a seamless fashion that you simply have an enormous data resource to plug into when you need. You dont need to know database servers, their names or where they are.
Grid virtualizes the idea of tapping into an enormous infrastructure to meet requirements for speed, scale and database processing.
[Stinger entails] no earth-shattering drama; weve supported clustering [since the 90s].
[Stinger incorporates] the way weve applied more automation to the HA part of the clustering story for DB2.