Larry Ellison kicked off Oracle OpenWorld with his annual keynote, and we covered it with a live blog earlier tonight.
In-Memory Option for Databases:
The first announcement was the most important: In Memory Option for databases. While initially we thought this was simply pinning tables in memory, something thats been possible for years, it turns out that isnt the case. According to Ellison, this new method not only runs things in memory, but also stores the data as both a row and a column. Transactional statements run faster on row-based data, whereas analytics run faster on column-based data. While its a little confusing how that actually increases performance, Oracle has some magic sauce that provides the following performance benefits:
- 100x times faster queries
- 3-4x faster inserts
- Fully compatible with all existing apps; no changes needed
To implement this new feature in Oracle 12c, the DBA needs to simply change a parameter, run a command, and drop the indexes.
Everything else is supposed to be completely transparent and according to Ellison, "Just throw a switch and turn on in-memory database. EVERYTHING runs faster without a SINGLE change to the app." They showed an on-stage demo where they sorted through three billion rows of Wikipedia searches. The first demo, a standard database with indexes, processed the three billion rows in roughly 1.5 seconds. Removing the indexes resulted in a query that ran for several minutes. Turning the in-memory option on processed over seven billion rows a second.
Oracle also announced the new M6-32 server, the "Big Memory Machine." This server contains 32 M6 SPARC processors, each with 12 cores, 96 threads, and 3TB/sec bandwidth. Compared to IBMs P795, this machine has twice the memory capacity, 50% more cores and twice the bandwidth at roughly 1/3 of the cost. However, Oracle licensing costs were not discussed. This was the machine that Oracle was running their Wikipedia test on, and its available now, although pricing was not mentioned and is not available on the website.
This has to be the device with the worst name.
Ellison: I came up with the name, thats why I make the big bucks. iPhone? Pfft, thats a boring name."Larry Ellison quipped that it was a terrible name but that he named it and thats why he makes the big bucks. He also joked that the iPhone is a boring name in comparison which drew a laugh from the crowd. The appliance itself is designed to backup thousands of databases. Whereas other appliances backup at the file level, the Oracle Database Backup, Logging, Recovery Appliance (ODBLRA?) backs up at the transaction layer. While Ellison didnt dig into the deep technical details, he explained that transaction logs are sent to the appliance and can be done in the same datacenter or across a WAN to a remote one. Two appliances can also be setup to replicate between themselves.
Datacenter of the Future
Ellison ended the keynote by discussing the datacenter of the future. He noted that everyone loves buying the cheap two socket Intel servers, installing virtualized Linux on them, and connecting them via wired Ethernet. He stated that theyre cheap, but are not good for every workload, emphasizing that Oracles engineered systems are still very important. We found this to be an ironic statement, considering the Exadata platform currently runs on Intel Linux servers, albeit with Oracles Flash Cache technology and special software.