Intelligent Enterprise

Oracle seems to be making big announcements in the quietest of news periods lately. A big Oracle Fusion 11g Middleware announcement was made earlier this summer on July 1, just two days before the Independence Day three-day weekend. Yesterday it was the Oracle Database 11g Release 2 (R2), announced in the doldrums of August just days before Labor Day weekend.

There didn't seem to be many of us reporters dialed into the conference call during the Q&A session; there were several long pauses while the operator waited for next questions. And only a handful of questions were ultimately asked. Granted, this is an R2 announcement, with mostly refinements rather than new features, but it's still an important release.

It's Oracle's practice to follow a major new database release (R1) with a second release about 18 to 24 months later. Somewhere between 15% and 20% of customers typically upgrade to the R1 while the majority wait for the "polished" R2, according to Andy Mendelsohn, senior vice president of Database Server Technologies.

11g R2 includes upgrades to availability, storage management and administrative productivity, but I was most interested in the new news related to data warehousing. On that front a new in-memory parallel query capability in R2 is said to support faster querying whenever tables and databases can be stored entirely within cache.

"If your tables are relatively small -- let's say under a terabyte of data -– you can create clusters with enough memory capacity across all the servers in your cluster to cache the entire database," Mendelsohn explained. "If you can do that, you can run massively parallel queries against the data all in memory."

By doing this, you'll get "quite a significant performance boost" by eliminating all the data I/O that would otherwise be required, Mendelsohn says. Sounds promising, but just how big a mart or warehouse would be suitable for this technique (the lone question I put to Mendelsohn during the Q&A session)?

The 11g R2 in-memory feature is intelligent, Mendelsohn explained; it will first examine the query and if just the tables or even just the partitions of the tables you are querying will fit in memory, the database will cache the data across all available servers in order to run an in-memory parallel query. Scalability depends on your hardware, but modern servers offer quite a bit of memory.

"Even commodity servers will have 100 or 200 gigabytes of main memory," he elaborated. "So if you have a ten-node cluster, for example, you can have a terabyte or two of memory in those clusters. It's fairly practical, and a lot of customers have databases are small enough that they can fit in memory."

It remains to be seen how easy it will be for this in-memory parallel query capability to be put in the hands of business users. Seems to me you would want some administrative control over the capability from within the BI toolset. Nonetheless, it's another sign that the market is responding to practitioner demands for faster querying.

E-MAIL | FOLLOW US