TechPulse 360

Teradata R&D boss Scott Gnau dismisses competitors

At Teradata’s user conference this week in Washington D.C., I chatted with CTO Scott Gnau about the database company main competitors (Oracle, IBM and a lesser extend HP Neoview and Netezza) as well as its views on cloud computing, virtualisation and Flash-memory.

“Our win-rate [against competitors] is much higher than 50%,” says Gnau.

Let’s try to understand why with part 1 of our interview with Scott Gnau on Teradata competitors.

What are your thoughts on Oracle’s Exadata database machine version 2?

The architecture of Exadata v.2 still has the same bottleneck than the prior version, because it’s still a rack, a monolithic non-parallel database. The database is a single instance on the rack, a single point of failure, and it will [fail]. The only implementation of the [Oracle] rack that I’ve seen are for failover and disaster recovery. So it’s not a massively scalable database.

But Oracle did optimise their I/O channel. So it’s running better certainly in a lot of comparisons of Oracle [v.1] to Oracle [v.2]. But they haven’t fixed the memory locking that exists in the normal frontside Oracle database in the rack.

So when you get to the level where you need to do extreme datamining, it’s constrained. But if you want to do simple reporting, simple list pulls, it works okay!

IBM?

I heard they had something coming last week! The rumour is that it’s coming soon! [probably at next week's Information OnDemand 2009 conference in Las Vegas where Big Blue should unveil a database machine competing with Oracle's Exadata].

And HP’s inability to compete with its Neoview offering?

You can use hammer to put a screw in the wall. But a screwdriver works much better. I think they are trying to retrofit some technologies and have had some trouble. Does go to proof that one person [Mark Hurd, in this case, who was a supporter of Teradata while at NCR] can’t a change a company sometimes!

Finally, Netezza?

They have got this hardware assist but they still got a problem in a host node, kind of the front end, where the actual database isn’t parallel. They have the same bottleneck than Oracle, but they have a lot less robust database [than Oracle] (feature, functionality, integration). So they have similar architecture constrain. But if you want to go run a couple reports that are not complicated, well defined, it will work fine. But if you want to do more – and that’s where the money is – you’re going to find yourself limited.