Oracle Buys Sun

April 21, 2009

I am surprised that anyone is surprised by this. I was more surprised that IBM would try to buy Sun, I’ve been waiting for Oracle to buy Sun for a long time.

With all due respect to MySQL, it isn’t about MySQL at all. I wonder if Larry even knows he bought MySQL. It is all about Java. Oracle has a huge and lucrative enterprise application stack on Java, as does IBM. Owning Java gives Oracle total control if its own stack. and some control over IBM’s. That’s got to be worth the $7billion alone.

As for the future of MySQL. Yes technically MySQL 5.x could do what a low end Oracle installation can do. But it hardly matters, no Oracle customer is going to risk porting a big enterprise app to MySQL to save a few bucks on license fees. Oracle may be expensive but it works so the customers will stay locked in. And remember that Oracle already gives away low end versions of its database now. It may affect new customers at the bottom but then again, a customer that would have chosen MySQL anyway is now an Oracle customer.

A look at Innobase or BerkeleyDB likely points to MySQL’s immediate future, ie. much the same as it was but with a big RED label pasted on it. Long term there are some challenges for MySQL though. Oracle isn’t going to let MySQL encroach onto Oracle Database’s rich territory. They would probably be happier if the 16 way support in MySQL 5.4 didn’t see the light of day but that cat is out of the bag already. I can’t see much more of that coming from Oracle. Which leads to something more serious, the splintering of MySQL. This was already happening with Sun but it will get worse with Oracle, especially if Oracle puts some brakes on its MySQL. Already there is the Oracle/Sun version, Percona, Monty, etc. It may not be bad, Linux flourishes under many versions, but it will be different from the focus that MySQL AB had.

The future of Solaris and Sparc are more uncertain. I doubt Oracle really wants to be in the hardware business. We can look at the HP Exadata product for a precedent. Expect the hardware busness to be sold to HP and HP to kill the Sparc as it did its own PA-RISC. Solaris has been poorly supported on Intel for years, HP likely has the skills to fix that and carry Solaris customers forward until it can convert them to HP/Linux or HP-UX. ( I can’t imagine a Solaris SA being happy with the prospect of being stuck with HP-UX though ).Whether Oracle sells or keeps the hardware, look for pre-configured Oracle/Solaris/hardware packages, especially for the SMB market

There is likely strong nostalgia within Oracle to keep Solaris, it was the development platform for a long long time and it is still the Cadillac environment for Oracle. Nothing really beats Oracle RAC on a big multi-cpu Solaris box. Sadly, the market just doesn’t support such luxury anymore so it probably has to go. We and Oracle will have to settle for RAC on Linux, especially if Oracle buys RedHat next.


Data Warehouses Redux

April 7, 2009

About a year ago I blogged about my dissatisfaction with Kimball style star schema based DWs. Well, lately I’ve been re-reading Kimball’s books and website and I’m now eating my words.

I guess I can I say this time I get it. This time I see how well conformed dimensions can create a usable DW. I still don’t think it is as elegant as the Inmon style but it can work.

More importantly, I see that the Kimball style DW is the only thing that will get built today. No one is going to get the sponsorship to build a centralized Inmom DW and then the data marts to hang off it. You have to start with the data marts and build the DW up gradually. A Kimball style DW with well conformed dimensions is your only hope. Franky. there are more and more people who try or must build DWs out of “federated” OLTP systems. A Kimball DW is a step up from that.

There is also a performance or latency issue. More and more we want “real-time” DWs or at least reduced latency between the DW and its source. The Inmom style inherantly has more latency becauses it a two step architecture of DW -> DM. If your star schema DMs are also the DW, its one step and less latency.