HP and Oracle Launch Database Machine, and So Can IBM with i
September 29, 2008 Timothy Prickett Morgan
Sometimes, I think of myself simply as a journalist and analyst, reporting on what is happening or what will happen soon in the IT racket. Sometimes, I feel a bit like the Greek chorus in a comedy or tragedy in the ancient world. And so it was last week as I watched Hewlett-Packard and Oracle announce a co-developed data warehouse appliance called the Database Machine at Oracle’s OpenWorld event in San Francisco.
The HP-Oracle machine is based on HP’s ProLiant DL360 G5 servers–five of them in a rack, to be specific, each with two sockets and using four-core Xeon processors from Intel–that are running Oracle’s Enterprise Linux clone of Red Hat‘s Enterprise Linux 5 and the Oracle 11g database with the Real Application Clusters (RAC) extensions for clustering the ProLiants together to support a single database image. The machinery is configured with four InfiniBand switches and 14 of HP’s Exadata storage servers, which are really just ProLiant DL185 G5 servers with Xeon processors equipped with 8 GB of memory and a dozen SAS or SATA drives that have funky software for accelerating database table scans (called Smart Scan), among other features to make data warehousing better that are inside the 11g database.
Smart Scan does table scans in databases and only sends the data from the rows and columns in the database that are being tickled by a particular from the Exadata storage servers out over the InfiniBand network to the ProLiant database servers, which actually crunch the data and do the query. Oracle’s 11g database has bitmap indexes, partitioning, multidimensional cubes, data mining, and online analytical processing features built in. The Database Machine has an aggregate I/O bandwidth of 14 GB/sec, and with compressed data, it can hit 50 GB/sec. With SAS drives, the Database Machine can store up to 14 TB of uncompressed data and with physically larger SATA drives, it can hold up to 46 TB. (Bandwidth is only 10.5 GB/sec uncompressed for the SATA drives.)
HP is the only hardware partner for the box; Sun Microsystems, a big Oracle ally back in the dot-com boom, was left out in the cold, and so was Oracle RAC enthusiast Dell. Pricing was not announced, and Oracle has hinted that it will create Database Machines on other chip architectures and operating systems.
The Database Machine is exactly and precisely the kind of database appliance that many of us have been after IBM to create based on the OS/400, i5/OS, and now i platform for many years. Three years ago, I was saying IBM could and should build a cluster of small OS/400 servers into a database engine and use the price/performance advantage that such clustering, enabled by the DB2 Multisystem capability that has been part of OS/400 V3R6, which is now 13 years old. (See Forget Oracle 10g. Let’s Talk About i5/OS V5g for more on that.)
Of course, it is never too late to get a good idea into the market. A decade ago, IBM’s AS/400 division–when Big Blue still had one–was all hot to trot about data warehousing among its more than quarter million AS/400 shops, which generally used Excel spreadsheets to make sense of their businesses. That was a good time to get a clustered database server for fast SQL queries to market. But, 2008 is not too late, either. A cluster of five Power 520s running DB2 Multisystem and a bunch of storage servers based on the BladeCenter chassis with JS12 blades linked to storage blades could be rigged up to look very much like Oracle’s Database Machine. And, it could turn out to be competitive on pricing and familiar to i shops that do not want to use 11g or Microsoft‘s SQL Server, which is the king of OLAP serving in the SMB world.
It’s something to think about, IBM. Particularly since Oracle did not work all that hard to get the Database Machine to market, if you look closely. It is a mix of existing hardware and software with a name and a price tag. Well, maybe not a price tag.