Big Blue Pits PureData Appliance Against Ellison’s Exadata

October 15, 2012 Timothy Prickett Morgan

IBM is settling into a spring-fall cadence for system announcements, with some wiggling here and there as machines come out early or a little late. We got the first Power7+ machines on October 3, and the new “Project Sparta” PureData appliances for “big data” and “cloud” came out last week on October 9. I get the big data part of the PureData announcements, but after analyzing the announcements and getting briefed by Big Blue, I still don’t get the cloud part. And I don’t get why the names on all of this stuff need to be so complicated.

The Project Sparta machines are actually one new database appliance tuned for online transaction processing workloads using the DB2 PureScale parallel database extensions that are analogous to Oracle‘s Real Application Cluster (RAC) parallel extensions for its 10g and 11g databases. The other two machines are a Netezza 1000 data warehouse and a Smart Analytics Systems 7700 data warehouse that are brought into the PureSystems family and given some of the same management tools but still based on their existing and respective BladeCenter HS22 and Power 740 rack system architectures.

The PureData System for Transactions T1500 is the new box, and is in fact based on the Flex x240 Xeon E5-2600 server nodes and the Flex System modular chassis that IBM launched last April after its development as “Project Troy” across several different IBM divisions.

I think that it is good that IBM is taking the fight directly to Oracle in the clustered systems arena with the PureScale database extensions, and I think it is also probably wise that IBM push X86-based server nodes against similar X86 technology that Oracle uses in its Exadata database appliances, which co-founder Larry Ellison insists are appropriate for both OLTP and data warehousing workloads. IBM has three different machines where Ellison has one. And until now, the PureScale parallel database appliance was only available on clusters of Power-AIX systems, but with the Pure Data T1500, IBM is moving over to its Flex x240 server nodes running Red Hat Enterprise Linux 6.

This brings me to a few different points I want to make. First, Oracle’s product naming makes more sense then the mish-mash of names that IBM has used over the past few years for systems and appliances. Exadata is a parallel database engine, Exalogic is a parallel WebLogic middleware machine, and Exalytics is an in-memory analytics box. Anyone can understand that. The names that IBM has chosen for its three new PureData machines are too complex. The OLTP appliance is called the PureData Systems for Transactions T15, and I think you could just simplify that right away to PureData T1500 and let people figure out the T is for Transactions. I have no idea what “Pure” is in any of these product names, but in Larry’s World, Exa is short for X86 and exascale, meant to imply the underlying technology as well as the scalability goal of the systems. (We are in the petascale era in supercomputing now, and hoping to get to the exascale era by the end of the decade, with systems capable of executing an exaflops of number-crunching and handle an exabyte of data, or a million times what was breakthrough technology back in the late 1990s.)

I think IBM used to know how to name stuff. Think about how perfectly reasonable System/360 was as a name–it covers the gamut of workloads and system sizes. Or Application System/400–this box is for people who only want to understand their applications and could care less about systems software or heaven forbid hardware. IBM should stop worrying so much about Smarter Planet and maybe spend a little more time on Smarter Somers, that being the town where most of the marketeering gets done at Big Blue. It could be that all of the good names are taken and trademarked. What do I know?

The other thing I would like to see, and I have said this plenty of times, is for IBM to make a PureScale-style cluster out of Power 720 servers with lots of flash running IBM i and using the DB2 Multisystem feature of the operating system and DB2 for i database. The price/performance of such a machine could be compelling and offer serious resiliency and scalability benefits to small and midrange shops running IBM i. IBM could even do it with Flex p260 nodes and Storwize V7000 storage if that helps sell the idea.

Anyway, in announcement letter 212-404, you’ll see that the PureData T1500 database appliance comes in quarter, half, and full rack configurations will differing numbers of compute nodes and flash and disk storage capacity to meet various levels of OLTP throughput needs. The full rack has 24 server nodes, each with two Xeon E5-2670 processors, which have eight cores and which run at 2.6 GHz. Each node has two 10 Gigabit Ethernet adapters to cluster the servers and two 8Gb/sec Fibre Channel links to reach out to Storwize V7000 disk arrays, which are equipped with a mix of flash and disk drives for storage and EasyTier hierarchical storage management to keep the hottest data on flash and the coldest data on disk. Across those 24 nodes, the systems have an aggregate of 6.2TB of main memory, and the four V7000 arrays (each with one expansion unit) have 19.2 TB of SSD storage and 128 TB of disk capacity that yields 74.4 TB of usable space for the DB2 10.1 database that straddles the nodes.

Speaking of software, the PureData T1500 has all that you need preloaded on it, unlike the Exadata machines, which comes with Oracle Linux but not with Oracle 11g, RAC, or Exadata storage server licenses for the storage arrays in the Exadata rack. IBM’s software stack includes the Enterprise Edition of DB2 10.1, plus InfoSphere Optim Query Workload Tuner 3, PureQuery Runtime 3, Optim Performance Manager 5, Optim Configuration Manager 2, and Tivoli Storage Manager Client. You can also load up InfoSphere Data Architect 8 and Data Studio 3 development tools if you want to pay extra.

The quarter rack of the PureData T1500 costs just under $500,000, including the hardware (servers, storage, and networking) and all that software, which should put a full rack at under $2 million before discounts. At list price, a full rack of Exadata without the software costs $1.1 million, but once you add the software onto it, list price to Larry is more like $4.5 million. The question, of course, is who has the faster rack?

The PureData T1500 will be available on October 26. And so will the Netezza 1000 data warehousing machine, which is designed to do very complex queries in seconds and which has been rebranded as the PureData System for Analytics N1001. The Smart Analytics System 7700, a data warehouse machine that is supposed to be able to process thousands of queries at the same time running DB2, InfoSphere Warehouse, and Cognos 8, is now known as the PureData System for Operational Analytics A1791.