Oracle has been accused of stifling the publication of an HP/Oracle DBMS benchmark that indicates its own SPARC SuperCluster world-record benchmark system cost almost 60 per cent more per transaction than a similar test on an HP Proliant system.
The record TPC-C benchmark result is held by a $30.53m, 108 processor, SPARC Supercluster, which achieved 30,249,688 transactions per minute (TPM) at a cost of $1.01/TPM.
When this was announced in December last year, Oracle boss Larry Ellison trashed HP, quoting a March 2007 Superdome (Itanium-based) TPC-C result of 4,092,799 TPM, which had a cost of $2.93/TPM. It was way too slow and pricey by comparison.
Our information, coming via customers, is that, at that time, Oracle was allegedly sitting on an HP/Oracle DBMS benchmark, run to TPC-C standards, showing a single Proliant DL980 G7 did 3,388,535 TPM at a cost of $0.63/TPM – this would be much more cost-effective than Oracle's unrealistic behemoth benchmark system, albeit slower.
Sources tell us this result has not been made "official" by the TPC-C because Oracle allegedly refuses to sign it off for submission to the benchmarking body. They also tell us that HP estimates that an Exadata machine – which Oracle, according to our source, allegedly refuses to benchmark – would have a performance cost of $2.40/TPM.
HP and Oracle used to be very friendly. An HP/Oracle system leads the top 10 TPC-C price/performance list, with two other HP/Oracle combos also in the list.
Back in 2007, the since-trashed HP/Oracle Superdome system held the world TPC-C record at just over 4 million TPM, and Oracle and HP were bosom buddies. In June last year HP had a new DL980 server coming out and wanted to boost its chances with a good benchmark result, but not going for any TPC-C record.
HP got together with Oracle, as before, and both companies put five months' worth of man hours into helping the DL980/Oracle system go as fast as possible. It performed well and the results of a benchmark run were audited and made ready for release and submission to the TPC-C. Oracle had to formally sign off on this as per a legal framework governing the Oracle and HP relationship in this matter.
Meanwhile, HP boss Mark Hurd resigned his position at HP on 6 August, 2010, after an internal investigation uncovered expense-account irregularities, then joined Oracle with Larry E trashing HP's board for being idiots. HP and Oracle stopped being friends and now exist in a state of virtual marketing warfare.
When it was time for Oracle to sign off on the use of the DL980/Oracle TPC-C benchmark in November, it did nothing. HP couldn't understand why there was a delay.
Then, surprise, surprise, the SPARC SuperCluster world record TPC-C result was announced in December and HP dissed big time by Larry E. In a launch webcast he said HP "got up to 4 million transactions per minute on one big honking Superdome ... and that's the best HP has ever done: 4 million ... That's HP's claim to fame."
He then mentioned IBM's 10 million TPC-C result in August last year, alluding back to the HP result with a "that's a little embarrassing" comment, before announcing Oracle's own massive 30 million TPM result. This whole presentation sequence would not have been anywhere near as effective if there was an Oracle/HP TPC-C benchmark with 60 per cent better price/performance out there. Oracle solved that problem by seemingly keeping the result private.
Oracle is now running a "Cash for Clunkers" campaign on its website, saying HP's Superdome is a clunker and should be replaced by an Oracle Sun server box. HP is busy telling its customers about the online transaction processing performance (OLTP) of the flash-assisted DL980 G7 server and quietly gnashing its teeth that it cannot say this is an official TPC-C benchmark, which it would be were it not for Oracle's apparent intransigence.
That's the story we are hearing. Marketing is warfare by other means. Indeed. Both HP and Oracle as well as the TPC have been asked for comments. An HP spokesperson said: "HP is declining to comment. For information regarding any benchmark tests involving Oracle software, please contact Oracle." Oracle too decided not to comment. There has been no comment from the TPC either. ®