With everyone talking about big data, the need for a big sever to handle the bytes may be necessary.
Oracle Corp.’s solution is the “Big Data Machine,” the newest server running its new Sparc M6 processor.
Officially called the M6-32, the server was announced Sunday at the opening of the annual five-day OracleWorld conference in San Francisco.
At the same time CEO Larry Ellison also announced a upcoming in-memory processing option for Oracle 12C Database that will run at “ungodly speeds” — running queries at 100 times faster and updating data two to three times faster than a standard database.
Designed to take advantage of that in-memory procession option, the M6 CPU has 12 cores (twice as many as the M5), 96 threads for parallel processing and up to 32 Terabytes of memory. Thanks to its silicon-based switching network, throughput is 3 TBps.
It’s “the fastest machine in the world for databases stored in-memory,” CEO Larry Ellison told the opening day crowd.
He said the M6-32 has twice the memory, more cores and double the bandwidth than IBM’s biggest server, the P795, which runs up to 256 Power 4 GHz CPUs. But the M6-32 costs US$3 million vs. the P795′s US$9.6 million, he said.
It can be bought in supercluster form attached via Infiniband to Oracle Exadata database and storage systems.
Oracle says the M6-32 can search 341 billion database rows a second.
For those with an M-series chassis, the M6-32 is available as a plug-in board that plugs.
It is available immediately.
An in-memory database processes data faster than one where data is stored on disk. A number of companies sell in-memory databases, including SAP HANA, MemSQL, EXASolution. IBM’s DB2 has an in-memory option, and the upcoming Microsoft SQL Sever 2014 will have it as well. Oracle also sells the Exalytics In-Memory Machine, a hardware-software bundle that uses the TimesTen In-Memory database Oracle purchased in 2005.
Now Oracle is adding the option of processing data in-memory to its flagship 12C Database. Pricing of the option wasn’t announced. But Ellison did say that it’s as easy to use as checking off an option on a menu.
Not only will the in-memory processing option work on queries stored in rows, Ellison said, it will also work on transactional data stored in columns because with the in-memory option turned on data will be stored simultaneously in rows and columns. There’s no need to create and update indexes needed for transaction processing. In fact, he said, after the in-memory option is turned on database analysts essentially delete the indexes.
Usually there’s a performance hit on analyzing columnar data, but Ellison said that doesn’t happen using 12C’s in-memory option because any changes to the database are only stored in the rows, not the columns.
Thanks to parallel processing, in multi-core systems each CPU core can scan billions of rows of column per second, he said.
In a conference demonstration using an M6 server, a query of a 3 billion rom table scanned at a speed of 7 billion rows a second.
To speed things, an entire database doesn’t have to be in memory. Oracle 12C will automatically import often-used (“hot”) data.
“The database will be easier to tune (than a standard disk-based DB) , it will run faster, it will be just as reliable, just as secure as it is today,” Ellison said.
Ellison also announced another backup solution from the company, the Oracle Database Backup Logging Recovery Appliance. Traditional backup appliances are really aimed at backing up files, he said, not databases. As a result, during the time it takes to do a backup changes to the live database aren’t included. The DBLRA includes log changes so databases can be restored from any time.
The DBLRA can backup any DB and handle thousands of database because it catalogues everything it saves, preventing databases from getting lost. The appliance can be placed in a remote data centre or in the cloud. In fact, the company is offering it as a service on the Oracle Public Cloud.
No price or availability was announced
More than 60,000 attendees are expected from over 145 countries for this year’s conference, Oracle [Nasdaq: ORCL] said. The JavaOne, MySQL Connect, and Oracle PartnerNetwork Exchange conferences are also being held at the same time.
Oracle president Mark Hurd will address the conference Monday.
Sponsor: IBM Canada Ltd
Increasing Profitability with Analytics in Midsize Companies
Complete with several mini-case studies, this Nucleus Research Note dispels the misperceptions surrounding Analytics, revealing how small and midsize companies have cost-effectively implemented and deployed Analytics solutions and gained benefits.