Original URL: https://www.theregister.co.uk/2010/04/07/ibm_smart_purescale_systems/

IBM widens data analytics fleet

Smartie boxes multiply

By Timothy Prickett Morgan

Posted in The Channel, 7th April 2010 22:09 GMT

IBM has expanding its lineup of appliances for data analytics.

Big Blue's Systems and Technology Group, which designs and makes servers, and its Software Group, which codes systems software for those machines as well as for those from Big Blue's competitors, are broadening the cross-group Smart Analytic Systems for real-time data analytics with the addition of mainframe and x64 variants of the platform.

Today, the company is also packaging up its PureScale database clustering software with its WebSphere middleware and new Power7-based systems to create what it is calling the PureScale Application System.

The Smart Analytics Systems for data warehousing and analytics and the PureScale appliances for online transaction processing aim to make it easier for customers to deploy and Big Blue to sell clustered systems, which offer much more scalability than big SMP iron at a more competitive price for a unit of work done. They are also, not coincidentally, designed to blunt the attack Oracle launched last summer on big iron with its Exadata V2 clusters. The Exadata machines match Sun x64 servers and flash and disk storage with Oracle database and clustering technology to create a parallel system that Oracle says is equally good for doing data warehousing and OLTP workloads.

IBM, by contrast, still thinks that these very different workloads require two different systems and hence the two different products from Big Blue.

The Smart Analytics System setups announced last July were based on clusters of the company's Power 550 midrange machines, which use its dual-core Power6+ processors, DS3500 midrange disk arrays with Fibre Channel links to the server nodes and plain old Gigabit Ethernet switches from Juniper. The servers ran IBM's AIX 6.1 operating system, General Parallel File System, DB2 V9.5 database, and InfoSphere Warehouse 9.5.1 data warehouse. Cognos 8 business intelligence and analytics software runs on some of the nodes - IBM offered versions with 7, 14, and 27 nodes with 25, 50, and 100 TB of user space.

The plan was to give the Smartie machines predictive analytics at some point in the future thanks to the acquisition of SPSS, which IBM accomplished on the same day the Smartie system was announced with a bag stuffed full of $1.2bn in cash. These initial Power 550-based versions of the Smartie systems shipped at the end of September last year, and thus far, SPSS software is not part of the stack.

Today, IBM is fleshing out the Smartie lineup with two more setups, one based on its System z mainframes and the other based on its System x x64 servers. The mainframe version is called the Smart Analytics System 9600, and it is based on a System z10 Business Class (BC) midrange mainframe with two logical partitions in a base configuration. The Smartie 9600 is available in six different configurations, ranging from small to extra extra large, using the shirt sizes that IBM is increasingly using to describe its Smartie and PureScale machines.

That base Smartie 9600 has z/VM for managing partitions and puts IBM's z/OS operating system in one partition and Linux on another. The z/OS partition is equipped with the DB2 for z/OS database in its Value Unit Edition (which means less expensive than it might otherwise be outside of this bundle) for databases scaling from 1 to 100 TB in size. IBM is also tossing in a license to its DB2 Utilities Suite for managing and tuning the database.

The Linux partition runs Cognos 8 analytics and InfoSphere Warehouse; these are the special mainframe Linux editions of those programs. This partition is capable of doing the analytic work for between 5 and 10,000 users, depending on how many processors in the System z10 box you dedicate to it. The bundle includes some base DS8000 storage arrays as well, but does not include the Linux license, which you need to buy separately from Red Hat or Novell.

Bernie Spang, director of product strategy for database software and systems at IBM, said the company is not announcing prices yet for the Smartie 9600, but did say this setup would be shipping at the end of this quarter. Spang also added that if customers have spare capacity on their existing z10 mainframes, they could plunk the Smartie 9600 stack on them and just pay the software and services part of the bill. The Smartie 9600 price, whatever it turns out to be, includes installation and configuration of the software.

The Smartie 5600 is the System x variant of the business analytics and optimization machinery, and it is based on a cluster of System x3650 M2 machines. Just in case you are wondering, these two-socket machines were announced in early 2009 sporting Intel's quad-core "Nehalem-EP" Xeon 5500 processors. These boxes can be equipped with the new six-core "Westmere-EP" Xeon 5600s, which came out in mid-March; IBM did not see fit to use its much flashier (and more expensive) System x rack servers based on the new eight-core "Nehalem-EX" Xeon 7500 processors from Intel and its own eX5 chipset.

You might be wondering why IBM went with System x rack servers instead of BladeCenter blade servers with the Smartie 5600. Our guess? Blades are more expensive, and the whole point is to get a bundled system together to do analytics and racks are an easier sell (both to customers and through partners) than blades.

The Smartie 5600 machines support Linux and are equipped with DS3400 midrange disk arrays for storage. On the software side, the Smartie 5600 includes InfoSphere Warehouse Enterprise Edition and Cognos 8 data analytics.

We know what you are thinking: Didn't IBM just slap Cognos 8 onto its existing InfoSphere Balanced Warehouse D5100 systems and put a new brand on it? Well, yes. And, with the Smartie 5600, IBM is allowing the option of adding solid state disks from Fusion-io to boost database performance. IBM says that sprinkling Fusion-io SSDs into the Smartie 5600 can boost the performance (in terms of I/O throughput) by a factor of four. The Smartie 5600 system has has three times the storage capacity and two times the query performance of the InfoSphere Balanced Warehouse D5100. Presumably including the effect of flash.

Sun and Oracle are big on the flash with the Exadata V2 machines - and rightly so, given the performance boost and reduced heat and space flash has over traditional disk storage.

The Smartie 5600 will be available on April 13. Pricing information was not available at press time.

IBM added that it will eventually debut a Smartie 7600 system based on Power7 servers, and given the machines Big Blue used in the original Smartie boxes from last summer, it stands to reason that the four-socket Power 750 machine announced in early February.

Extended PureScale

On the OLTP front, the PureScale cluster extensions for its DB2 database that IBM that debuted last fall, also based on the Power 550 midrange Power6 servers (up to 16 cores) as well as on the big Power 595 boxes (up to 64 cores). The PureScale extensions to DB2 were put into the field to rain on the Oracle and Sun Exadata V2 parade, and it is a fair guess that IBM did not intend to launch this feature until Power7 systems were out the door this year along with DB2 V9.8.

But Oracle forced IBM's hand, and complaints about the scalability of Oracle's Real Application Clustering for its 11g database on OLTP workloads is giving IBM an opportunity to fight back. Customers also complain that using Oracle RAC requires a lot of tuning. But IBM says PureScale is the holy grail of sorts for OLTP databases in that it can make a database spread across multiple server nodes look and act like it is on one giant SMP server with a shared, single memory space. As El Reg explained last fall, the secret sauce in the PureScale setup is the head node in the cluster that acts like a master database record lock as it locks and unlocks memory and storage on the distributed database.

IBM is exploiting the Remote Direct Memory Access (RDMA) features of InfiniBand to tightly link the main memories of the servers in the PureScale clusters. IBM is linking the Power System servers through their 12X remote I/O buses, 12X being a variant of double-data rate InfiniBand that Big Blue has used for years to link remote I/O drawers holding disks and controllers back to servers.

Spang says that in benchmark tests using real workloads, DB2 plus PureScale has been able to attain 95 per cent efficiency in 64 node configurations and 85 per cent efficiency in 128 node setups. Meaning, on a 64 node machine, you get about 61 times the performance of a single node. That's pretty good linear scalability.

The DB2 PureScale feature was available in December, and IBM has come to the conclusion (after some complaining from customers and partners) that what it needed to do was put together a complete system exploiting PureScale, not just preview a database feature and hope that is enough to counter the Oracle Exadata V2 marketing onslaught.

And so, IBM has whipped up a little something called the PureScale Application System, an homage, no doubt, to the king of the application servers, the AS/400 minicomputer. The PureScale Application System, which needs a shorter name, is being previewed now for shipments in June. This preconfigured setup will consist of two of the Power 770 machines, which come with sixteen Power7 cores each and have four of them activated in each server. The machines run AIX 6.1 and DB2 V9.8, since PureScale is basically what turns the current DB2 V9.7 into DB2 V9.8.

IBM went with the Power 770 box rather than the Power 750 because it allows customers to scale database nodes horizontally through SMP nodes in a single system if large memory helps with their particular database workloads. The Power 770, as El Reg previously explained, has two processors per system card, with two system cards per chassis; up to four chassis can be lashed together in a system that has up to 64 Power7 cores and up to 2 TB of main memory when fat memory cards are available this fall for the system; the machine tops out at 512 GB of main memory now.

The PureScale Application System also includes IBM's WebSphere Application Server, and it's ready to go to support Java applications and their transaction workloads. Once again, IBM did not announce formal pricing for these boxes, but Spang said the intent was to have the entry configuration cost "a few hundred thousands of dollars," not millions. IBM has also tweaked its Optim Performance Manager Extended Edition tool to optimize the performance of DB2 plus WebSphere or DB2 plus SAP middleware and applications running in conjunction with PureScale database clusters.

The uptake on the Smartie boxes and PureScale DB2 cluster extensions has been slow, but you expect it to be with a new way of bringing systems to market. Spang said that IBM has sold dozens of Smartie analytic machines and has a number of PureScale deployments in production. In both cases, Spang said interest is running very high. ®