The Oracle/Sun announcement of Exadata Version 2
Sometimes i just look at the comments, article and opinion pieces out there in the internet and asking myself, what happened to logic, what happened to systemic thinking, what happened to see the obvious.
But let’s start at the beginning. Oracle and Sun (still two totally separate companies, there are two brand logos on the systems ;) ) announced yesterday a new version of the Exadata system.
The announcement states in Marketinginese:
The result is a quantum leap in performance over Oracle's previous generation Exadata machine, including a 20x increase in random I/O and a 5x I/O bandwidth speedup with FlashFire Technology, as well as a 2x speedup in compute and network performance capacity. In short, the Exadata Version 2 is the world's fastest machine for both data warehousing and online transaction processing.
It’s a Nehalem based system running on Oracle Enterprise Linux. This system uses 40 GBit/s Infiniband instead of the 20 GBit/s variant. It uses Nehalem instead of of the Core2 based Xeons. Provides 72 GB instead of 24 GB main memory. I would summarize it as some evolutionary changes to the system while adding flash memory to the equation as larger step of development.
A product isn't just hardware and software
Yes, my heart is bleeding, too. Yes, i would have liked to see a SPARC based system with Solaris as well. But we have to see this realistic. You have a running and established product (the Exadata V1 system) running on HP components. You plan to buy a large hardware vendor. What do you do? Yes, you stop to use the components of a company that is possibly a competitor in the near future and use those components from a company you will own in the future. I should tell you something about making products: Don’t change too many things at once when you already have a product. Each change has to be evaluated, evaluation takes time. So at first yield the quick-wins. And in this case it was swapping the hardware. Selling a product is more than just having a running system, it’s about training the support organisations, it’s about the knowledge of to implement it, this takes time. Step by step has it’s advantages. And by the way: There are infinite numbers above 2 in the set of natural numbers ;)
It's an OR, not a XOR
Interestingly there are some news outlets interpreting the fact as contrary to the ad at the WSJ. You have two ways to interpret this announcement: Oracle changed their mind about SPARC and just wants to do x86 and the SPARC thing was just a joke. Sorry, what a large heap of bullshit. The other way to see this is much more positive and it’s somewhat like an additional commitment: “Oh, Oracle uses the x86 gear of Sun”. But i have to assume, that isn’t such good headline line like the of of BlueToTheBones of SPARC missing in action. And sorry Oracle … it’s a RAC beneath it … and it has it’s own share of problems on non-disjunct data enabling you to shard the data (Albeit it’s not that problematic with Infiniband than in the times when people used 100 MBit/s or 1GbE for the RAC interconnect). At the end the idea Workload Computing still holds true. It’s not a XOR, it’s an OR of both technologies.
Reading those news outlets these days is more and more just an outright and major annoyance. At the end just two further comments:
- By the way, the HP based Exadata appliance used x86 and not Itanium, OEL instead of HPUX. You get it ? ;)
- 2. There will be other announcements. One announcement was already announced ... kind of ;)