Oracle has high expectations for its newest release (Oracle Database 11g R2.) “We expect 45-50% adoption of R2 by next year,” said Mark Townsend, Vice President of Product Management, at the database analyst day during Oracle Open World recently. Such a rate would be unprecedented, but Oracle has good reasons for its optimism. Many new features target extending cost-effective use of the systems (server, storage and software) in place, and the financial drumbeat was clear and consistent. Many of these benefits are due to Oracle’s increasing ability to leverage organizations’ architectural tiers: smarter use of interconnected servers, storage, and memory are driving performance improvements at many levels. Should Oracle’s acquisition of Sun win through, one can expect to see an acceleration of this trend.
By now, Oracle RAC is well established (Gartner reported 15,000 companies using it as of the beginning of 2009) as a way to leverage middle tier commodity server hardware. “When RAC arrived with Oracle 9i, it was seen as a high availability (HA) play,” Townsend said.
With 10g we saw people unlocking the value of smaller servers, but it was still typically one app per cluster. Now we see big customers stand up a grid to consolidate mission-critical and departmental apps onto one cluster.”
Oracle reports 40-400 databases in the wild on a single cluster. And whereas today many customers still buy the biggest box they can, SMP boxes have a “tax” associated with them. Logical partitioning of big boxes is a typical practice; 11gR2 introduces policy-managed dynamic cluster partitioning via Server Pools, which can be assigned to take similar workloads. Oracle can dynamically adjust the boundaries of the pools, and the promise for the future is to enable adjusting them based on performance, not only on failure. Prior to R2 it was possible to do so within a database; now it is possible to do so across them. It may not seem obvious to critics (and they are legion) of Oracle’s pricing – especially on maintenance – but these are ways to save money. Although of course in many ways they shift spending from other vendors, not Oracle itself, an even more intriguing play.
RAC OneNode offers another opportunity: an additional way to tackle consolidation with a single instance of RAC running on one node in a cluster. Oracle calls its strategy a “Maximum Availability Architecture” – when a customer adds redundancy, the database can use it for added performance. Active Data Guard lets customers move backup and reporting to redundant machines. It’s also useful in maintenance scenarios, reducing the hardware requirements: temp start a second instance, do maintenance and then roll back (this was referred to as an “Oracle Motion type of event.”
Increasingly, this is an “online everything” story. Apply patches across the cluster, do full db/OS upgrades across cluster, online reorg – the opportunities are legion. One of the most intriguing features is online application upgrades. Planned downtime is always a hard problem – organizations often need to stand up an entire parallel environment for such changes. Instead, it is now possible to make code changes in a new “Edition.” Redefinition views project to a new version only seen by its users; triggers put updates into both places to ensure updates are replicated. “All Oracle’s applications are instrumented to use this,” we were told. An 8-node RAC cluster is the typical configuration, but some 16s are showing up, Oracle says.
Leveraging storage is the next key opportunity. “11gR2 now is a general purpose cluster file system,” Townsend asserted. Storage growth is not about to slow down, and Oracle is targeting the issue by targeting storage virtualization as a means to lower costs. Automatic Storage Management (ASM) helps break down the divide between sysadmins and DBAs; it implements the “stripe and mirror everything” practices in wide use. I have not looked at the relative cost numbers yet, but Veritas replacement is clearly in Oracle’s sights here. The claim is that over 50% of Oracle’s customers on 10g and 11 adopted ASM, but it didn’t have binaries and external files. Now it will, extending its value. Customers tend to map system files to types of storage based on performance requirements – but that means year by year, they still buy both (cost) levels of storage. “In any system only about 5% of the data is active,” Townsend says, “so we’re enabling Oracle to move the less used data over time to less expensive storage. With 11g we can use advanced compression and get 4x compression – giving customers half of their disk back.” We’ll need to see what the performance hit is with real customers, but this is clearly another promising value proposition for Oracle to sell from.