EDIT: Big whoops, and apologies to Philip. I didn’t check the date, and what I linked to was last year’s article. That said, it read as if it could have been this year’s, which tells us something about the pace of Netezza’s information disclosure. Resulting errors of mine are left in place.
Netezza perennially annoys me by the secrecy with which it surrounds its information disclosure, especially at the annual user conference (just concluded). Essentially, except for what has also been separately disclosed, the whole thing is under NDA beyond the generality “We told you that we intend to improve our product by making more use of the FPGA.” Blech. That said, Philip Howard* has a long and — no surprise there! — upbeat article. So I’ll link to that, saving me some worries about what I myself am or am not allowed to say. E.g., I wouldn’t dare suggest — as Philip does — that Netezza’s zone maps (essentially, one-dimensional partitioning) could be enhanced going forward. And while I think Netezza has made strong efforts to tell the marketing stories Philip describes as being “hidden under a bushel,” I agree that — largely because of its self-defeating mania for secrecy — Netezza hasn’t done nearly as good a job of getting those messages accepted as it could have.
*Just to be clear — notwithstanding how much I tweak him for his exuberance, Philip seems to be a great guy, both in his publications and in person.
In general, much of what Philip wrote I would agree with. That said, let me hasten to point out some exceptions, including:
- I don’t understand his feelings of surprise, given that I saw him at last year’s conference as well.
- I don’t understand his point about “streaming.” Yes, Netezza streams data off of disk rather than doing a lot of random seeks. But DATAllegro does the same thing, without recourse to FPGAs. That doesn’t really have much to do with complex event processing (CEP) at all.
- He’s over-impressed with Netezza’s watts/terabyte figures. A Netezza 10400 is rated at 50 TBs of user data. It consumes a little over 30 watts for each of its 432 active SPUs, for a little over 260 watts/TB (or more if there’s some other power cost I’m overlooking). By way of contrast, Stuart Frost tells me that in a high-performance DATAllegro system, power consumption is a little over 120 watts/TB of user data.
But notwithstanding those quibbles, I recommend Philip’s article highly.
And as I’ve said before — if Netezza could get over it’s secrecy fetish about technology, and DATAllegro could get over its secrecy fetish about customers, both companies (and therefore the whole industry) would probably be a lot better off.
Keep getting great research about database management systems, enterprise applications, and related technologies. Get a FREE subscription by RSS/Atom or e-mail!