So what does Oracle Exadata mean for HP Neoview?
That HP is committed to selling a lot of data warehouse hardware — and probably data warehouse appliances in particular — seems obvious, for reasons including:
- HP bought a big BI/data warehousing consulting operation in Knightsbridge.
- HP has put considerable effort into its data warehouse appliance Neoview.
- HP CEO Mark Hurd comes from data warehouse appliance vendor Teradata.
- Data warehousing where the big bucks are.
But Oracle Exadata could produce those appliance sales. So where does HP Neoview fit in?
I was told by an investor today that HP’s investor relations department is saying Oracle Exadata is a Netezza competitor, while Neoview is more in the Teradata market. That’s laughable. The differences between the Teradata and Netezza markets are:
- Teradata scales credibly into the multi-petabyte range, while Netezza today is still down under half a petabyte (each of those figures could be cut in half, if one has stringent standards for credibility).
- Teradata can handle just about any workload for just about any enterprise. Netezza gets up into several hundred concurrent user range, but is more solidly proven at smaller user loads.
Generally, Release 1 of a product has trouble with more complex and concurrent workloads, and the same usually goes for Release 2.* While HP Neoview is promoted as having some high-end features, I have no reason to think it’s actually an exception to that rule. Thus, Neoview should be regarded as being in the Netezza market — and that perhaps only at the wannabe level. There’s no reason to believe in Neoview as being higher-end at this time than Netezza, which has been shipping product for several more years, and probably still has >10X the customers Neoview does.
*Actually, that’s a generalization that extends far beyond data warehouse appliances.
Oracle, on the other hand, is the rare “new” entrant that does deserve benefit of the doubt about concurrency. The part of Exadata that’s really new, getting data off of disk in parallel, is pretty simple. Either Oracle is doing it well out of the gate, or they’ll get it sorted out pretty shortly. What happens to the data after that is that it gets sent up to a RAC cluster. Well, if there’s one thing Oracle RAC clusters are good at, it’s handling concurrency. Anybody can have bottlenecks in a Release 1, but by Release 2 of Oracle’s Exadata-based data warehouse appliances, I don’t expect concurrency to be a big problem.
Nor are mixed workloads a problem in Oracle, now or in the future. The only thing that’s been wrong with Oracle data warehouses is that they can’t run queries very fast, except perhaps after staggering amounts of administrative work and hardware investment. So the area where Oracle should be regarded as guilty-until-proven-innocent is in performance of big queries, not database or workload complexity.
Thus, it’s more accurate to call Oracle Exadata a Teradata competitor and HP Neoview a Netezza competitor than the other way around. But any such labeling overlooks an even more basic issue — Neoview isn’t really competing with anything at all. HP has gotten some initial Neoview sales from a variety of friendly customers, some of whom — surely not all — may even have paid something resembling official prices for the product. But the next loss report a Teradata or Netezza or Greenplum salesman files about a Neoview deal will be very close to the first one. From everything I can tell, HP Neoview isn’t a significant market factor at this time.
Comments
16 Responses to “So what does Oracle Exadata mean for HP Neoview?”
Leave a Reply
I think the question that you ask in your post title is not answered in the post itself…
What I would like to know is: what does HP want to do in the datawarehouse area???, is planing to compete in partnership with Oracle and forget about neoview, are exadata and neoview competitorr or play in different leagues???
I would like to know about HP strategy in the datawarehouse game, as I see weird moves in the Hurd’s house
Luis,
The questions you’re thinking of are unanswerable. Different people at HP believe, plan, and predict different things. It’s always that way at huge companies — for any reasonable strategy, you can find somebody somewhere in the company who not only believes the strategy is correct, but also believes the company will actually follow it.
All we can do is predict which of the competing strategies will (more or less obviously in advance) fail and which will come closer to succeeding.
Best,
CAM
Curt, on the Netezza scalability I think you mean half a petabyte, not half a terabyte…
I have to possibly put my ignorance on the line with regards to the following paragraph:
“Nor are mixed workloads a problem in Oracle, now or in the future. The only thing that’s been wrong with Oracle data warehouses is that they can’t run queries very fast, except perhaps after staggering amounts of administrative work and hardware investment. So the area where Oracle should be regarded as guilty-until-proven-innocent is in performance of big queries, not database or workload complexity.”
What is a “big query”? How do you quantify it? If several different “big queries” (or even just one) are run and other small queries are affected does it become a workload complexity problem? Also, what do you mean by “mixed workload”?
This will certainly be an awkward little dance for the two partners.
I have no doubt that the Oracle Exadata server will be fast. The question are..
What will that speed cost me in “all in” H/W+S/W $s?
What are the person-hours needed to manage the Exadata environment?
Good catch, Derek. Fixed. Thanks!
Carl,
Concurrency problems arise when the resources required to run a bunch of simultaneous queries are significantly more than the sum of the resources that would be required to run them one after the other.
Mixed workloads are ones in which the concurrent queries have very different characteristics.
Big queries are ones that, for example, scan huge tables or join medium-to-huge tables with each other, or do large numbers of joins. Smaller ones might do a SELECT on a single table, or just join a couple of non-huge ones, generating a report.
Query volumes of small queries are commonly higher than volumes of big ones. One may populate 100s of dashboards at once. One probably is not going to run 100s of data mining extracts at the same moment.
CAM
CAM,
I’ll buy into your concurrency and mixed workload definitions (although there are some slight ambiguities). Now I’ll get back to the “big query” question and why you think that Oracle cannot run big queries fast. Is the issue one of architecture (e.g., row versus column storage) or of a poor query optimizer? I infer from this and your other posts that you consider it to be more the latter. I ask whether you consider any of the following row-based RDBMSs to be bad/so-so/good with respect to big queries: MS SQL Server, IBM DB2, Sybase ASE, and Sybase ASA? (I’ll leave it you as to which releases/platforms you are familiar with. I know that ASE 15.0.2 is generally faster than 12.5.x but sometimes is slower.)
Thank you, Carl.
Carl,
With the possible exception of DB2, none of those has supported the I/O to query large amounts of data quickly. Exadata proposes to change that in Oracle’s case.
DB2 hasn’t covered itself in glory at the super high end of database sizes either, but I’m not totally clear on why.
CAM
CAM
as a user of teradata system for more than 15 years, I just want to tell you guys to stop promoting things that you don’t know!!! if you don’t have teradata balanced configuation then teradata is not better than oracle (et all)! shared nothing obviously not good in heavy joining (PIs are not located in the same node!need to redistribute rows)…full table scan could be the only a differentiator for teradata in a big config…what is the cost of that?
Jef,
My rule of thumb is that Teradata’s actual users usually love Teradata, but their bosses who sign the checks sometimes hate Teradata. You seem to be an exception to that rule — not that it ever was 100% true anyway :)– which is a valuable contribution. So thanks for speaking up!
CAM
The question for the Neoview team is whether HP EDS and Knightsbridge architects and proposal teams are going to recommend Oracle or Neoview. In a competitive evaluation and bidding situation they are unlikely to propose both. My guess is that Oracle will be recommended for Oracle shops and Neoview recommended elsewhere and that Neoview has lost about 40% of it’s potential market.
[…] evidently got HP’s attention by a recent post in which I questioned its stance on the relative positioning of the Exadata-based HP Oracle data warehouse appliance and […]
[…] Автор: Curt Monash Дата публикации оригинала: 2008-09-26 Источник: Блог Курта Монаша […]
Neoview will be a failure. It fails to perform, runs queries slower than Teradata, and will in due course be scrapped by HP.
a hayes,
What is your experience with Neoview?