Oracle continues its record of having 0 (i.e. ZERO) credibility. How many times has Oracle been called out for publishing & making statements about competitors solutions that were just flat out wrong leading one to wonder if it is more than the standard competitor FUD and benchmark exaggerations but purposefully meant to mislead.
Take a recent Oracle blog post by @Brian-Oracle at https://blogs.oracle.com/BestPerf/entry/20160317_sparc_t7_1_oltp. Oracle is hell bent to produce a TPC-C benchmark on a POWER8 server since IBM has not. I do not work for IBM and have not heard any official reason but have heard they do not see TPC-C as a good benchmark of platform performance which is why you do not see any entries since 2013 by any vendor.
There is a Oracle Marketing troll who posts as @PlinkerTind for this El Reg article http://forums.theregister.co.uk/forum/1/2016/04/07/open_power_summit_power9/?post_received=2834982#c_2834982 who asks what is IBM scared of. What he fails to disclose is that IBM would love to run a Oracle benchmark on POWER8 but the software license agreements state if the product is being used for benchmark purposes it requires the approval of the owning company. Well, Oracle doesn’t permit it except for Oracle benchmarks such as Oracle EBS. Why would Oracle not want to let IBM conduct a benchmark using Oracle DB? Because it would show what customers see who run it and what prospects wee who evaluate it that POWER controls software licensing; Oracle, DB2, EnterpriseDB, and every other product. Fewer servers, fewer sockets, fewer cores, etc.
Back to the Oracle blog, author writes “On a per chip basis, the SPARC T7-1 server demonstrated nearly 5.5 times better performance compared to an IBM Power System S824 server”. Let’s break it down using the authors “vernacular”. A T7-1, although it is actually 8 x quad core chiplets and not a single 32 core chip as Oracle likes to claim, yet this author along with the other Oracle marketing trolls refer to a 2 socket S824 & S822 as a 4 chip system – essentially for the Scale-Out servers, they call each socket two chips because of how IBM builds the chip using a Dual Chip Module (DCM) vs the Single Chip Module (SCM) in the Enterprise servers. Now, IBM who engineered the chips says it works functionally as 1 chip in 1 socket. (Ref: IBM POWER8 S824 Redbook http://www.redbooks.ibm.com/redpapers/pdfs/redp5097.pdf).
So, the Oracle blog authors says the SPARC 32 core chip, by HIS words is 5.5X better performance than the POWER8 S824 server. Thus, 6 cores vs 32 cores or 5.33X. 6 cores because Oracle is comparing Chip to Chip. Thus, they consider it reasonable and credible to compare their 32 core SPARC T7 chip vs a 6 core POWER8 chip. That is so disingenuous.
Next, the author chose the 24 core S824 running at 3.52 GHz vs the 4.13 GHz T7-1. Why didn’t Oracle pick the 16 core S824 running at 4.15 GHz? That way they could call it a 4 chip, 2 socket with 4 cores per chip system vs Oracle’s 32 core, single socket, single chip (that is really 8 x quad chiplets) system. Using the 16 core S824 would give the SPARC T7-1 server 2X more cores allowing for easier extrapolation? Oracle did set the ASMI mode to favor performance but it is the 2nd best option to use (there is a better option but since Oracle doesn’t know the platform they either chose not to or didn’t know what to select). Even with this set, their is still a clock frequency discrepancy. Now, I’m not hung up on the clock frequency because most Intel servers have lower clock frequency. Unlike Intel which cannot run all of their cores at the higher clock frequency like POWER there are times when the clock frequency comparison being made is vastly different can’t be helped. That’s not the case though with POWER. IBM offers servers ranging from 3.0 to 4.35 Ghz making it easy for them to choose one that makes the comparisons as close as possible.
Next the author says “On a per core basis the SPARC T7-1 server demonstrated nearly 3% better performance per core compared to an IBM Power System S824 server.”. I’ll just refer to the above paragraph where the T7 clock frequency was running 17% higher. Then he says “At the system level, the SPARC T7-1 server demonstrated nearly 1.4 times better performance compared to the IBM Power System S824 server.”. 24 cores times 1.4 = 33.6……hmmm. You tested a 1 socket vs a 2 socket server running with a 17% higher clock. I point out the 1 vs 2 socket server because with POWER processors I would expect a single socket to be slightly better performing than a 2 socket than a 4 socket than a > 4 socket server. Taking into account other factors that do not require the memory & I/O provided by those extra sockets. Again, if they want to test on a per core basis that can be done with any system. If they wanted to test on a per socket basis then do it with the T7-1 and even use the S824 but only use 12 cores. Otherwise, compare a T7-2 vs a S824 which would be 64 vs 24 cores.
Read the comments for an article about POWER9 at AnandTech http://www.anandtech.com/show/10230/ibm-nvidia-and-wistron-develop-new-openpower-hpc-server-with-power8-cpus-nvlink. Look for comments by @Brutalizer who is an Oracle Marketing troll. He gets crushed by the the commenters as they rightly point out that Oracle ran the Oracle benchmark on the POWER8 server with no disclosure on the full details of how the server was configured; How many DIMMs were used? Filling all 16 DIMM slots makes a difference than just 8 or 4 DIMMs since all 3 configuration options can achieve the configured 512 GB Ram. Although a few tunables were disclosed as if to demonstrate that Oracle made an effort to give the POWER server a fair shake, I question the ones used. I won’t disclose what I would have done differently as I like that Oracle looks like petty fools in their effort to show Oracle on POWER8 performance. If they were interested they would authorize IBM to run their own benchmark – or accept my POWER challenge where we run a customers workload on each of our servers. Alas the cowards have yet to acknowledge it let alone accept it.
The lesson I hope readers learn from my blog is not that Oracle software and hardware products are bad, they are not at all. However, they have this seemingly uncontrollable need to overstate and mislead customers by any effort to get customers to consider and god forbid buy their products. Yes, in this poker game we all play there is bluffing and smart play but with most vendors it is within the rules. With Oracle they always seem to have a card stuffed in their sleeve as they seem incapable of competing fairly making them resort to unscrupulous behavior such as the Oracle blog.
P.S. I do not accept any of the results obtained by Oracle testing the S824. If IBM conducted the same test running Oracle with AIX on POWER8 I am confident the results would make the S824 look far better than stated by Oracle. One can only draw the conclusion that Oracle optimized their T7-1 to achieve the most favorable results. IBM should have the same opportunity. Of course, my Power challenge to Oracle is the real-world test using actual customer data. Maybe this blog will get Oracle to man-up!
Power The Enterprise 