At Oracle OpenWorld, the highlight of the show typically comes from the Sunday afternoon keynote from Oracle founder Larry Ellison. This year, however, he skipped the keynote in favor of watching his sailboat race around the San Francisco Bay, so Oracle president Mark Hurd was on hand Sunday evening to do the new product introductions. That night, he introduced the new Oracle In-Memory Database.
Hurd characterized the new database as a simple way to speed up existing applications. “You flip a switch, and with no rewriting of your applications you get the performance benefit of in-memory,” he said.
And while Oracle is pitching its new in-memory option as an entirely new product, many at the show were asking specifically how it stacked up to SAP’s HANA in-memory datastore. Hurd addressed these questions during a press conference.
“I don’t like it when Exadata or the In-Memory option gets compared to SAP’s HANA, because I don’t even think they’re comparable. HANA has to be programmed. What we told you about doesn’t have to do that. You’re not rewriting anything. All the magic is below the database layer,” he said.
The SAP bus tour
SAP, for its part, is pushing HANA as an analytics platform. In early September, SAP’s Byron Banks, vice president of database and technology, hosted an unveiling of a new bus his company purchased. This bus is acting as a roving expo booth, spreading the word of HANA across the country.
Banks has been following this bus city to city, where he shows off HANA demos. One of those demos comes from the NFL, showing a football stadium. Each seat’s status can be checked through a HANA-backed application, and by using facial recognition, sentiment can be gleaned from fans.
Banks and his team also showed off a similar HANA-backed app that showed the maintenance status for airplane parts via a large real-time flight map.
But Hurd maintained that, despite the introduction of Oracle In-Memory Database, the company has been selling in-memory solutions for some time. Indeed, it has been selling TimesTen for years now. “When you look at In-Memory, I don’t want anyone walking away thinking In-Memory was first announced today. We’ve been in-memory long before HANA was a product,” he said.
Amit Sinha, senior vice president of database and technology innovation at SAP, said that Oracle’s solution lacks some of HANA’s capabilities.
“SAP HANA has one columnar store for both transactions and analytics,” he said. “Oracle’s approach keeps two redundant copies of data, one for transactions in a row, and a duplicate one for analytics in a column. They have fundamentally not erased the traditional divide between OLTP and OLAP—simply masked it. This implies greater data footprint, complexity and more DBA-intensive work (e.g., DBAs need to designate which rows to change to in-memory).
“While Oracle was in catch-up mode, SAP HANA matured from a database to a platform to perform application functions close to the database, e.g., predictive libraries, business functions and app-server capabilities are all performed in memory.” He added that HANA has been on the market for more than two years now.
But Oracle and SAP aren’t the only ones playing in the in-memory market. ScaleOut Software yesterday announced the availability of hServer version 2, which brings real-time analytics capabilities to Map/Reduce in Hadoop.
And JavaOne, the sister show that took place alongside Oracle OpenWorld, also saw the release of another new in-memory datastore: Hazelcast.
Miko Matsumura, vice president of marketing and developer relations for Hazelcast, said that this in-memory data grid offers a lot more than just a place to quickly store data during execution. And that’s where Hazelcast distinguishes itself; it’s not just an in-memory data grid; it also includes a message queue, metadata for each operation in the queue, and other handlers typically associated with components further up the stack.
Matsumura added that Hazelcast is also about bringing processing to the data. Instead of pulling data from the database, modifying it and replacing it, Hazelcast allows developers to quickly access data and modify it without the need to write it back to disk or waste time reading procedures from that same disk.