The new in-memory option for Oracle Database 12c allows queries to be processed 100 times faster - or even more - while also delivering a useful doubling of transaction processing speed.
This is achieved through a combination of keeping data in memory and maintaining two separate structures: one organised by rows for speedy transaction processing, the other arranged by columns for fast query processing.
Both are simultaneously active and transactionally consistent.
Since the columnar data is only used for queries, there is no need for logging and so "there's very little overhead," Ellison said.
Each CPU core scans the columns in its local memory using vector instructions and so billions of rows can be processed per second.. "It's at least times faster," he said, and even complex queries are accelerated as tables can be joined around ten times faster.
The new option means "you're getting results at the speed of thought," he said.
The reason transaction processing operations are accelerated is that there is no longer any need to maintain indexes that exist only to improve query performance. Where perhaps ten or 20 analytical indexes would have to be updated every time a new row was inserted, now it is just the few actually required to make the database work.
A further advantage is that the database administrator no longer has to anticipate the types of queries users will make and therefore the indexes required, as data can be retrieved based on any field with equal performance.
"We can throw all those [analytical indexes] away," Ellison said, so "OLTP runs dramatically faster" and administrators do not need to worry about what to index: "everything runs fast" so tuning is unnecessary.
"Flip a switch and all your existing applications run much faster," he added. "It's a unique proposition' - an easy way to get much higher performance with no loss or reliability or security.
A live demonstration showed a search of a three billion record table running at five million rows per second with 'traditional' Oracle Database 12c. With the new new in-memory technology, performance soared to 7151 million rows per second, a 1367x increase (well in excess of the overall claim of 100x improvement).
This development could eliminate the need for a data warehouse in some situations, as the massive performance improvement could make it feasible to run analytics on production databases.
If there is not enough memory for all the data, 12c will use a combination of memory, flash and disk, automatically migrating data according to access patterns.
That segued into Mr Ellison's second announcement of the afternoon, the M6-32 Big Data Machine - a server with 32TB of memory, 32 SPARC M6 processors (each with 12 cores and 96 threads), 3TBps bandwidth for the shared memory and 1TBps of I/O bandwidth.
"It's the fastest machine in the world for databases stored in memory," he claimed, saying it offers twice the performance of IBM's P795 at less than one-third the cost.
The M6-32 is also offered in SuperCluster form with Exabyte storage giving tenfold database I/O acceleration.
Another on-stage demonstration showed the M6-32 working through a 218 billion row table at 341072M rows per second.
Ellison's final announcement was that Oracle Database Backup, Logging, Recovery Appliance. "That's the actual product name," he said. "Who was the genius that named the product? You're looking at him."
The point is that traditional backup appliances are designed to back up files, not databases, he said. They provide restoration from the most recent backup so there is potential for data loss, they slow processing, and are not scalable so multiple appliances are required, resulting in greater managemet effort.
It is very fast and handles huge loads according to Ellison. "It backs up all of your Oracle databases" and catalogues them so "you have a database of databases."
The incremental updates put little load on the network, so it is practical to install an appliance in another data centre, and "it could be in the cloud," he said.
And Oracle is indeed offering the appliance as a service running in the Oracle cloud. Ellison suggested using the service as a "belt and suspenders" second backup in addition to a locally installed appliance, as an alternative to two appliances installed in separate data centres.
The writer attended Oracle Open World as the guest of Oracle.