IBM responds to Oracle's Exadata 2 database appliance with DB2 pureScale clustering feature.
I remember once I was on my high school cross country team. Only a limited number of runners were going to be chosen to travel across state to compete against another school. At lunch, the coach read the names of those who were going to compete in the meet. My name was not among them, and I was shocked. I knew I was not among the top runners, but I still was proud of what I could do and felt I was a serious team contributor. I didn't appreciate being left off the list.
The list of database stars among the IBM team has just been read, and, sadly, DB2 for i is not on the list. It takes money to field a team, and the audience cheering for DB2 for i apparently doesn't have enough clout to get the uniforms, shoes, and transportation that it will take to get DB2 for i into the competition.
There is a serious contest brewing between Oracle and IBM now that Oracle fancies itself a company with Sun's hardware, software, and operating system assets. Although the European Union has stymied the $7.4 billion purchase of Sun by Oracle because it's worried about the future of the open source MySQL in the hands of Larry Ellison, Oracle continues to move forward as if Sun were already part of the Oracle family. The evidence of that is in the recent company announcement of its supposed "DB2 killer" as Ellison likes to call its new Exadata V2 appliance.
The idea of clustering servers in the midrange arena started to get some play back in the 90s, but it was seen largely as a high availability solution and never really caught on except at a few large financial institutions. Since then, there has been an explosion of data in business. The general acceptance of RFID with its recordable little tags for the purpose of controlling inventory, to companies and consumers having to monitor energy use, to the government-backed adoption of electronic health records, to delivering consolidated information to telecommunications companies' mobile devices, the need to seamlessly scale databases without compromising performance has moved to the top of the list for some IT shops when selecting a new system.
Oracle has recognized the need to offer a scalable database architecture while also being well aware of the budget limitations on IT during these recessionary times. Its answer is Exadata 2, a combination of Sun hardware sporting flash memory and conventional disk drives running Intel quad-core Xeon 5500 processors and Oracle's 11g database on Linux with Oracle's Real Application Clusters (RAC) technology. The combined package of database and storage servers linked with InfiniBand switches apparently can deliver a super 100 TB data warehouse on which one can perform online transaction processing at the rate of 1 million random I/O operations per second. A single complete rack, which sells for $1.15 million, can be scaled up to an eight-unit cluster.
IBM fired back last week by announcing it would release a clustering feature for DB2 on Power Systems that it is calling DB2 pureScale and will be available in December on Power 550 Express and Power 595 systems running AIX 6.1 and DB2 V9.7. It shot back at Oracle that pureScale would be an "Exadata killer." The pureScale feature will support application transparency and won't require sophisticated database tuning, according to IBM. This means that IBM customers won't have do anything to their current applications running on a pureScale enhanced system, which reduces the risk and cost of growing their systems. Essentially all they have to do is add additional servers. The company also plans to offer the service "on-demand" for occasional use at a fraction of the server and package purchase price.
IBM says that tests of pureScale on clustered servers indicate excellent efficiency. In a system performance test on more than 100 Power servers, DB2 pureScale brought down system efficiency by only 20 percent (or achieved a total system productivity score of more than 80 percent), the company said. When 64 servers were clustered, productivity was reduced by only 10 percent, according to the company (methodology and specific measurement data weren't provided in the initial company release).
One of the features of Oracle's Exadata 2 with Real Application Clusters is the way it quickly and constantly checks all the database nodes before committing to a transaction. The IBM solution is slightly different in that DB2 pureScale has a mirrored master database locking and caching server node that communicates with all the other nodes to grant access to a database field for updates. Server node clusters are linked with InfiniBand switches featuring very fast Remote Direct Memory Access (RDMA), which greatly speeds up data transfer.
Sadly, DB2 pureScale probably won't be supported on DB2 for i, at least not in the foreseeable future. The bus is going to leave for the competition against Oracle, and IBM's choice for the first round of this contest is AIX. Second and third on the list of candidates are likely Linux and Windows. It's alright, though—really—since IBM i has its own clustering technology. DB2 Multisystem and DB2 SMP are available to use if and when anyone running IBM i wants to consider—or reconsider—clustering. If you're building a large data warehouse, particularly one where speed is paramount because you also want to do online transaction processing, which OS are you going to choose—IBM i or AIX? Whatever--but let's not leave DB2 for i out in the cold on this if there is a legitimate business reason to support pureScale on i (and why wouldn't there be?).
Since cost is a big factor in creating large, scalable databases, which can get expensive, users should be interested in learning that a new runner will be on that bus headed to the meet against Oracle and other non-IBM vendors—the iSN1000 storage appliance under development by Avnet Technology Solutions and Scale Computing running on the System x. SMBs will be eager to cheer for this runner. The iSN1000 is a 1 (usable) TB storage node that can be clustered with two others for a 3 TB StarterSAN for as little as $18,300. While more of a storage appliance solution than a complete database machine, the iSN1000 will use Scale's Intelligent Clustered Storage (ICS) technology that employs IBM's General Parallel File System (GPFS) as its core file system. GPFS is a sophisticated symmetrical parallel file system. ICS taps into a number of its features and adds additional ones of its own including the ability to run CIFS, NFS, and iSCSI protocols simultaneously. It creates a unified SAN/NAS environment. The iSN1000 with ICS technology is a clustered storage system that can scale per TB to more than 2.2 PBs on a single file system. While the iSN1000 is a 1TB node, plans call for future release of 2, 4, and even 32TB nodes. The unit is symmetrical, and all drives are data aware as information is mirrored and striped across the entire cluster.
Avnet will make the iSN1000 available through its IBM value added reseller partners in the U.S. and Canada. This competitively priced, scalable clustered storage solution running on the System x should offer mid-market customers a leg up on ways to deal with the exponential growth in data.