Oracle: Beyond Benchmarksmanship -- SPARC T3-1 takes JD Edwards "Day In the Life" benchmark lead, beats IBM Power7 by 25%.
This one is for the books! It appears Sun has passed the “report obscure benchmark you do well on tradition” to Oracle. Sun used to report success with the Manugistics NetWORKS Fulfillment Benchmark. Good luck finding the latest results for that “industry standard” benchmark.
While the JD Edwards "Day In the Life" is an active benchmark, it certainly is not in the category of industry standard, such as the SPEC suits or TPC-C. It is so obscure that Oracle didn't bother to provide a direct reference in their announcement for the reader to make sense of the results. A googling of “JD Edwards "Day In the Life" benchmark” produced an IBM white paper that provided the following reference in Appendix B:
Minimum Technical Requirements (MTRs) for JD Edwards EnterpriseOne are hosted on the MyOracle Web site: https://support.oracle.com/
- Log in to Oracle’s Partner Connection with your userid and password:
- After signing in, click on More, then click on Certifications
- Double click on JD Edwards EnterpriseOne Certifications
- Double click on the Note Link for MTRs, such as Note:745831.1
- Then scroll down to the MTR needed and double click it
We are not going to register as an Oracle Partner to analyze Oracle's claims. We need only view what Oracle has published in its announcement to examine their claim. Oracle claims:
- The SPARC T3-1 server is 25% faster and has better response time than the IBM P750 POWER7 system, when executing the JD Edwards EnterpriseOne 9.0.1 Day in the Life test, online component.
- The SPARC T3-1 server had 25% better space/performance than the IBM P750 POWER7 server.
- The SPARC T3-1 server is 5x faster than the x86-based IBM x3650 M2 server system, when executing the JD Edwards EnterpriseOne 9.0.1 Day in the Life test, online component.
- The SPARC T3-1 server had 2.5x better space/performance than the x86-based IBM x3650 M2 server.
What Oracle did not say was that:
- For a single socket SPARC T3 to have 25% better results than a single socket POWER7, Oracle needed twice the number of cores and four times the threads as the IBM POWER7.
- Oracle compared their just released SPARC T3-1 results with that of an IBM POWER6, a product announced almost FOUR YEARS ago. This is very disingenuous of Oracle and assumes their customers will not bother to check if Oracle is making apples-to-apples comparisons. Sun used to assume this.
- Oracle then compared their new SPARC T3-1 server results to an IBM x3650M2, 2x2.93 GHz X5570, with 64GB of memory – half the RAM of the Oracle T3 machine. It is incumbent on Oracle to compare their new machine with a comparably configured IBM x86 server, that is, one with 128MB, or provide results for SPARC T3-1 server with 64GB of RAM. Neglecting to do so will result in more of Oracle's performance claims coming under increased scrutiny.
Oracle claims their SPARC T3-1 is 5X faster than the IBM IBM x3650M2. This claim is not conclusive. A server cannot be 5X faster simply because the benchmark reports it serviced 5X the number of users. Moreover, the IBM x3650 M2's response time is .29 secs compared with the latest SPARC T3-1 of .523 sec. If response time is more important to the user, the year and half old IBM x3650 M2 with half the RAM and half the core count is about 2X as responsive as the latest SPARC T3-1 servers. In fact, the amount of available system RAM usually has a direct relationship on the number of users. It will be interesting see what the latest IBM X3650 M3 with 128GB of RAM will have for results.
If core licenses costs are important, IBM's year old POWER7 750 uses half the number of cores for about the same benchmark performance as the the just released Oracle SPARC T3-1 server. This Oracle benchmark announcement demonstrates that the latest SPARC T3-1 server, at a minimum, has a business application suit licensing cost of 2x over that for IBM POWER7 750 servers. Oracle is telling us that SW costs based on cores could double by using their HW over POWER7 servers from IBM.
Oracle should not assume that the readers of its benchmark results will believe their claims without investigation!