of the technology. Importantly, it will be up to the vendors, not customers, to do that optimization work as part of 12c certification.
4. You can't explore data in memory unless you put it there. Ellison spoke of exploratory analysis scenarios and the possibility of getting rid of predefined multi-dimensional cubes, but how is this possible without everything in memory?
"In this first release, there will be no dynamic loading of data into memory [based on queries], so you have to predefine that," said Shetler. "If you put five columns in memory and you query demands access to six, you're hobbling response times because you're going to flash or disk for that last column. In our tests, customers tended to put whole tables or whole partitions into memory, and if they have enough DRAM, they don't even think about economizing."
[Want more on the in-memory options and choices? Read In-Memory Databases: Do You Need The Speed?]
In a future release, Oracle is planning to add automated advisor tools that look at data-access patterns over time and suggest which tables to place into memory, Shetler said. Microsoft introduced similar tools with its Microsoft SQL Server OLTP In-Memory option, and customers say they greatly simplify decisions around which data to store in memory.
5. Oracle is not proposing to get rid of data warehouses. At one point during his presentation, Ellison said the In-Memory option would "make everything a lot simpler ... and you can use a transactional system as a data warehouse." This sounds like an echo of SAP's "radical simplification" claims, but it seemed unexpected coming from Oracle, which makes tons of money on data warehouse deployments. What did Ellison really mean?
"We're not talking about getting rid of data warehouses, because they're not just for analytics, they're a system of record," said Shetler, noting that Ellison "probably meant to say or should have said" data marts. "Marts tend to be separate copies of the production database used for analytics and reporting. With the In-Memory Option, you run mart applications against the production database."
6. The in-memory option does not require Oracle Engineered Systems. Despite all the touts of Exadata and M6 Sparc during Tuesday's presentation, the in-memory option will run on any hardware certified to run Oracle 12c, Shetler said. "Scale-out [deployment] and fault-tolerance are dependent upon the RAC clustering technology, but that's available on any [certified] hardware."
Oracle took its in-memory option through a three-month, 60-company beta program, according to Shetler. In contrast, Microsoft went through its customary private and public community technology previews with In-Memory OLTP. Each preview was months long, and in the end, Microsoft saw tens of thousands of downloads and bug reports from customers. By comparison, the Oracle release seems a little rushed. Rest assured features like deployment advisors and dynamic loading based on query patterns -- techniques already used by competitors and elsewhere by Oracle itself -- will appear in a follow-up release of the in-memory option.
Make no mistake: Oracle Database In-Memory will bring important benefits to customers. And Oracle is certainly not alone in highlighting the high points while soft pedaling certain details of its in-memory promise. SAP's promised "radical transformation" of IT infrastructure, for example, can't happen "without disruption" as it claims -- particularly in analytical deployments.
With Oracle Database In-Memory, analytical improvements are a sure thing and modest transactional improvements are possible, but you or your software vendors will have to make some changes to make the most of the technology.
IBM, Microsoft, Oracle, and SAP are fighting to become your in-memory technology provider. Do you really need the speed? Get the digital In-Memory Databases issue of InformationWeek today.