Neoview comes out to play

Hewlett-Packard sort of announced its data warehouse appliance (if that is the right term—see later) last autumn. The company refers to this as a soft launch in the sense that they didn’t talk much about it except to a few beta clients. Well, now it has had its hard launch (which is officially version 2.0) and the company is going to be talking about it much more widely.

The first thing to question is whether this is, or is not, an appliance? Technically the answer is yes. However, the company views the term appliance, as currently perceived within the data warehousing market, as pertaining to a particular environment where you throw the appliance at difficult analytical problems and resolve them through brute force. Neoview can do that. However, H-P sees this market as a niche and doesn’t want to get restricted by it. The company believes that it is way ahead of the appliance vendors in terms of its support for enterprise data warehousing (EDW), mixed query workloads and so forth, so the company sees its potential market as consisting of two areas: one in which it competes with the likes of Netezza and the other in which it competes with Teradata and IBM. This is why it is not entirely happy with being labelled as an appliance vendor even though, technically, it is.

Neoview is actually based on the Tandem Non-Stop database. However, while it uses a Non-Stop OS kernel and a Non-Stop SQL database it is not the same version of the database that most of the investment banks (and, for that matter, the New York Stock Exchange) still use for transaction processing. During the 90s Tandem developed a data warehousing version of its database that stripped out all of the transaction processing stuff that you don’t need for warehousing and included a new top-down optimiser based on the Cascades project (as also used by Microsoft). It also had enterprise data warehousing features built into it such as materialised views, prioritisation, scheduling and so forth. In all, the development produced 1½ million lines of new code that never got sold or marketed. It is on top of this that Neoview has been implemented, which means that you can’t reasonably compare Neoview with data warehouse appliances from vendors such as DATAllegro and Netezza, to begin with because it has indexing (using Btrees via a mechanism known as MDAM), which is both a good and a bad thing depending on your perspective and what you want to do; and because it has far more features (at present at least—this will no doubt change over time) to support enterprise data warehousing and mixed query workloads. On the other hand that doesn’t mean to say that H-P won’t compete for the same business that the appliances do.

Another reason why Neoview can not be directly compared with the various appliance vendors in the market is that its architecture is different. Netezza, DATAllegro, Greenplum and so on all have basically the same architecture: an SMP system at the front-end and a dual interconnect that links a series of MPP (massively parallel processing) nodes that attach directly to their disks. Neoview, on the other hand, has the Interconnect at the front (with which BI and other relevant tools communicate), linking a series of H-P Integrity servers that connect to fibre channel disks via a switch fabric (using H-P ServerNet technology).

So, how well will H-P do? That depends on various factors. Perhaps the most worrying is that H-P has never been very good at selling software (the company would, of course, disagree with this statement). I have always felt that the company saw software as a way of shifting tin rather than a worthwhile exercise in its own right. To a certain extent I am mollified by the fact that the company is recruiting a specialised workforce to sell Neoview. This will be fine, provided that staff members do not get sullied by any tin-based thinking from other parts of the company. Mark Hurd’s influence should help.

Secondly, let’s consider the Netezza market and assume for the moment that Neoview has the performance one would expect from an appliance (even if it isn’t called that). It has the big advantage that it is H-P as opposed to a company that has just started the process of floating on the stock market. On the other hand it is priced to take business away from (in particular) Teradata (at something over $200,000 per terabyte for the faster of the two models, or $80,000 per node) and therefore is likely to be at a price disadvantage when competing with Netezza. This probably cancels out: so we can expect win some, lose some but it will certainly make life tough for the other appliance vendors out there that haven’t got the momentum of Netezza.

Thirdly, it is worth noting H-P’s avowed reason for entering this market: that BI is becoming more operational and mission critical. I completely agree. As a result, it feels that Neoview’s 24×7, user scalability and mixed workload capabilities are major differentiators for the company. Sorry guys: I disagree. I don’t disagree that you are not good at these things but I do disagree that you have a major advantage over your EDW competitors here or that you will retain such an advantage for long, and in the case of 24×7 the big issue is not unplanned outages anymore (which have pretty much been cracked) but planned outages (which is another article).

Anyway, regardless of my views (or yours) on the three previous paragraphs, the fact is that, in the EDW market, Neoview is going to heat things up considerably, as it will mean significant additional competition for Teradata and IBM (and not forgetting Oracle) going forward. However, H-P does not expect to win many customers in 2007 and it is really only looking for marquee customers during the course of this year that it can go to market with as references in 2008.

This sounds very sensible and reasonable. However, Netezza will be introducing more EDW capabilities over that time frame (as will other appliance vendors) and there are also other companies waiting in wings that have not come to market yet, but which will change the dynamics of the EDW market. The danger, therefore, is that Neoview is slow out of the blocks. It needs to gain rapid momentum if it is to make an impression in an increasingly crowded market. While I can understand the company’s soft launch last year and its (planned) gradual progress this year, this may not be fast enough.