Oracle's Table Stakes in Hardware Game

With great fanfare, this week Oracle announced version 2 of its Exadata Database Machine, claiming it to be the world’s fastest for both data warehousing and online transaction processing (OLTP). Using terms like “quantum leap in performance,” “20x increase in random I/O,” “5x I/O bandwidth speedup’” the announcement webcast was big on superlatives but vague on apples-to-apples competitive comparison details. Soon, we hope, will appear the quantitative benchmark substantiation versus competitive offerings.  

Oracle’s initial Exadata hardware platform, labeled "HP Oracle Database Machine", employed HP ProLiant servers as compute and storage server nodes, running Oracle’s own Linux distribution. The new “Sun Oracle Database Machine” employs the same software, but switches to Sun Nehalem-based servers. The jump from Xeon 5400 in the HP edition to Xeon 5500 Nehalem in the v2 Sun version understandably yields substantial performance benefits, with Oracle claiming 80% faster processors and 200% faster memory, as well as faster disk throughput and networking.

In a significant enhancement, the Sun servers incorporate “FlashFire” flash storage cards (addressed as an extended memory not as disk to avoid disk controller limitations). With up to 56 flash cards per rack, each containing 96 GB, the over 5 TB of flash cache per rack, coupled with the nearly 1 TB DRAM per rack, allows for a very large in-memory database. Furthermore, Oracle claims that advanced Exadata compression allows 10-15X, even up to 50X, data compression to permit even larger databases to fully reside in memory.

Further details on the Sun Oracle Database Machine will be available in the next IDEAS TechTrends Monthly newsletter (free with registration).

Thanks to large in-memory databases, Oracle predicts that the Sun Oracle Database Machine will offer outstanding OLTP capabilities in addition to superior Data Warehousing performance. Naturally, we await benchmark substantiation. Oracle indicates it will demonstrate record-breaking 1TB TPC-H, which sounds entirely feasible given the huge “memory” of combined flash and DRAM. What’s less clear is whether Oracle plans to substantiate OLTP claims via TPC-C or TPC-E. 

Oracle emphasized that this announcement demonstrated the benefits of its pending acquisition of Sun. Admittedly, Sun has been a leader in adoption of flash memory, but it is not clear that Sun Nehalem servers offer substantive benefits over HP Nehalem servers, had version 2 been based on HP ProLiant instead of Sun boxes. Nonetheless, it does indicate that Oracle will continue to offer and support Sun x86 hardware, at least for the near term.

Note that Oracle has also issued statements that it plans to continue investment in SPARC and intends to set new TPC-C records.To be clear, Sun Oracle Database Machine is x86 and not SPARC. And, despite the advantages of flash memory, the Exadata Database Machine version 2 is unlikely to be the vehicle to challenge IBM’s 6 million TPC-C mark. Our speculation is that the promised TPC-C result to be unveiled at Oracle World will be measured on a cluster of Victoria Falls servers, likely a collection of Sun T5440

So why is Oracle so aggressive in claiming leadership performance with Sun hardware? It is understood that a major acquisition often endures lengthy regulatory scrutiny. Competitors typically seize upon the uncertainty to preach FUD and poach customers. To preserve the Sun base, Oracle needed to clarify its commitment to Sun hardware, and probably should have done so long ago. Admittedly, Sun and Oracle remain separate until the merger is approved and consummated. Still, the Sun Oracle Database Machine helps to confirm Oracle’s willingness to retain Sun’s hardware business. Reported benchmark results on Sun-based servers, be it TPC-H or TPC-C, will further demonstrate Oracle’s interest in promoting Sun platforms. We look forward to such benchmark results confirming the benefits of an integrated Oracle/Sun.

Comments

[rockmelon]

There goes that theory.

In fact Oracle/Sun did publish a TPC-C benchmark using Flash and delivered better performance and price/performance than the next highest result (IBM).

[K Mann]

No, Oracle/Sun will not be publishing TPC-C or TPC-E.

If they did, we'd all see how the cost-per-transaction-per-minute is between 4x-12x that of a well-designed DRAMHDD system.

If they did, we'd all see how it takes >20 years for these systems to pay for themselves when the smaller-than-expected energy savings are amortized over the system costs.

The problem is that write-performance of Flash is dramatically worse than a decent HDD with DRAM.

Here's the proof, from a company that is NOT trying to pump it's newly acquired hardware business!!!

http://portal.acm.org/citation.cfm?id=1519081