It's hard to believe that Oracle Real Application Clusters (RAC) has been out for almost 10 years. But why open...
an article about Oracle Exadata X2-8 by reminiscing about Oracle RAC? Well, it's important to understand that Exadata is an evolutionary step in moving database workloads to commoditized hardware that began with Oracle RAC. Ironically, Exadata X2-8 is a step back toward large-scale computing, as we will see.
It has been an interesting ride for those of us who have lived with and around RAC, including its previous incarnation, Parallel Server. Along with the maturity of Intel and Linux, Oracle RAC has changed the landscape of database workloads by shifting IT spend from hardware to software. Consequently, dollars have shifted from proprietary Unix providers to Oracle, and operational management has evolved from disparate administrative roles to database administration.
Exadata has enabled Oracle to overcome challenges associated with RAC. The addition of InfiniBand between compute servers and Flash Cache has opened bottlenecks for data from disk to memory to processing. New compression and indexing techniques have reduced the amount of data pulled from disk to begin with. Finally, there has been the addition of software that rides on storage server CPUs to offload database processing from compute nodes.
That's all good, but if the story ended there we'd still have what some consider a performance tax associated with spreading workloads across multiple nodes. Until X2-8, Exadata relied on dual-CPU server nodes. That required three, four or more compute nodes for large database workloads. The evolution of Intel gear has enabled Oracle to build Exadata around two 8-CPU nodes, thereby inching back toward SMP-style computing. The trade-off for having two larger compute nodes is that X2-8 is only available as a full rack. Consequently, smaller organizations may opt for Exadata X2-2, which is based on two-socket nodes and available in quarter- and half-rack configurations.
When is Oracle Exadata X2-8 a good option?
When and where to consider Exadata: That is the question. If you are running Oracle RAC and outgrowing existing hardware, then Exadata is a natural platform to consider. Oracle Exadata X2-8 should be considered an evolutionary improvement to existing workloads that are optimized for clustered databases. Furthermore, I/O-intensive applications that require high availability are a candidate for Exadata since RAC provides an active-active environment with Flash Cache. Finally, Exadata is the genesis of Oracle's goal to provide integrated systems, so the company will likely continue to invest heavily in it for the next 10 years. It’s a safe bet that Exadata is here to stay.
On the cautionary side, it's not a foregone conclusion that Exadata provides the best price-performance for database workloads, even though it is based on low-cost Lintel gear. The software costs associated with Exadata are substantial, and Oracle RAC remains more administratively difficult to manage than single-instance deployments. While Oracle has invested substantially in Unix technology through its acquisition and subsequent marketing of Sun Microsystems hardware, Hewlett-Packard Co. and IBM have slashed the price of Unix servers. IBM has even built "Oracle mode" into DB2 that eases the migration path between database vendors.
This has leveled the playing field within the Unix-versus-Lintel value equation, especially considering that Oracle Standard Edition can be licensed on as many as 32 cores of IBM’s Power 7 processor. So x86 solutions such as Exadata X2-8 should be compared both technically and financially with Unix solutions that would have been considered far more expensive through the mid- to late 2000s.
Eric Guyer consults with Fortune 1000 companies to provide vendor-independent assessment and optimization of Oracle software and associated hardware assets. Forsythe is a multi-vendor IT infrastructure integrator that provides consulting services, financial solutions and products. Guyer also writes the Oracle Optimization blog.