Elisabeth Stahl on Benchmarking and IT Optimization

Oracle’s New T5 TPC-C: Where’s the SPARC?, Part II

with 5 comments

With Oracle’s new SPARC server announcement today, we are all still waiting in anticipation (take your pick of Rocky Horror or Carole King) for something exciting. The just released TPC-C benchmark result surely is not.

Here are some reasons why:

  • The performance of the Oracle T5-8 (even with the use of Oracle database partitioning) is downright lackluster. An IBM POWER6 result from 2008, 2 generations ago, is 42% higher per core. An IBM POWER7 result from 2010, 1 generation ago, is 2.2x better performance per core than the Oracle result. (1)
  • The price for all Oracle software support used in computing the price/performance for this benchmark is $2300/year. I can only guess what you get for that.
  • The Oracle database software is not even available until September. Yes, September.
  • It’s keeping me wa a a a aiting . . .


    (1) IBM Power 780 (2 chips, 8 cores, 32 threads) with IBM DB2 9.5 (1,200,011 tpmC, $.69/tpmC, configuration available 10/13/10); IBM Power 595 (5 GHz, 32 chips, 64 cores, 128 threads) with IBM DB2 9.5 (6,085,166 tpmC, $2.81/tpmC, configuration available 12/10/08); vs. Oracle SPARC T5-8 (8 chips, 128 cores, 1024 threads – 8,552,523 tpmC, $.55/tpmC, configuration available 9/25/13).
    Source: Results current as of 3/26/13.
    TPC-C ,TPC-H, and TPC-E are trademarks of the Transaction Performance Processing Council (TPPC).

    The postings on this site solely reflect the personal views of the author and do not necessarily represent the views, positions, strategies or opinions of IBM or IBM management.

    technorati tags: , , ,,,,,,,,,,,

    Written by benchmarkingblog

    March 26, 2013 at 2:23 pm

    Posted in Oracle, SPARC T5, TPC-C

    Tagged with , , ,

    5 Responses

    Subscribe to comments with RSS.

    1. I think you are not considering the fact Oracle T5-8 price/performance result is cheaper ($.55/tpmC) than IBM Power 780 ($.69/tpmC) and IBM Power 595 ($2.81/tpmC). Although it is performing better by achieving the result of (8,552,523 tpmC) against IBM Power 780 (1,200,011 tpmC), IBM Power 595 (6,085,166 tpmC).


      March 26, 2013 at 4:36 pm

      • But of course… what else were you expecting? When you lose the straight fight then you start to slice and dice till you look good in some parameters… 🙂 e.g. hmm.. my 300BHP V6 car is less powerful than your 500BHP V12 car… but wait!!!! your car delivers less power per cylinder than my car… so my car is more powerful… Brilliant!!!

        Tan Ah Beng

        March 27, 2013 at 3:53 am

    2. I’m sorry but following the links you’ve provided I can’t understand your claim.
      What I’m seeing is a much better overall result for T5-8.


      April 1, 2013 at 7:52 am

    3. Show us IBM Power 780 full config with 8 sockets TPC-C performance !!!! Why IBM only has 2 chips benchmark on 780 ? Is it because 8 chips performance is NOT good ????


      April 16, 2013 at 7:41 pm

    4. Where to start with the Oracle bandwagon crowd. You seem to follow Elizabeth around making claims as if repeating them makes them true. Your claims are lacking significant fact.

      1) The Power6 595 result achieved a 6 M TPC result with 64 cores. This was roughly 50% better than the Power5 595 result. I’ll state the obvious here. 64 cores achieved 6 M whereas the 128 core T5 only achieved 8.5 M. If you had to guess….just work with me here….if you have to guess what a 128 core P6 595 (doesn’t exist but we are in pretend mode here) result would be, what do you think?

      2) The reason you see IBM and their sellers (which I am one – IBM BP) extrapolate is because Power servers actually scale fairly linearly. So, if I have a 8 core 4.14 GHz 780 achieve 1.2 M TPC then by extrapolation I would expect a 64 core to achieve somewhere around 9.6 M TPC (maybe down by 6% or so making it 9.1 M). That would mean a 64 core 780 would achieve more TPC than the 128 core T5. What would you say if IBM did a TPC test with their 256 core 4.0 GHz Power7 795? I’m sure you would agree it would be higher than the T5-8 8.5 M number wouldn’t you? If we could be honest with each other and set aside our desire to intellectually ‘one up’ each other right? If the 64 co P6 595 is 6M, Oracles 128 co T5-8 is 8.5 M and let’s just say the 256 co P7 795 is 22 M (let’s just say although I would expect something around 24 M)……what would you then say about Power performance? If you used your Oracle licensing factor advantage of .5 compared to the 1.0 for Power then you could extrapolate to say with 512 (4 x 128 co)) would yield roughly 34 M and your licensing equivalent would only be 256 cores. Do you agree? That is what I would say. Draw attention from the fact it takes 512 cores and the licensing manipulation by a ISV to equalize the inefficiency of my SPARC technology. Oh wait, Oracle did that already with that crazy clustered 30 M T3 supercluster result using 27 servers with 1728 cores.

      3) A Power7+ 780 running at 4.42 GHz is roughly 26% for 8 cores and 19.5% for 64 cores better performing than the Power7 780 running at 3.86 GHz. So, from 8 to 64 there is a 6% decrease – ready, set, go! IBM publishes this information on a public website – there for everybody to see. Where can I get published performance results for a 16, 32, 64 and 128 core T5-8? I don’t mean a benchmark but the sizing information that a Oracle SE would use? We could chart both to see the linear scaling for both the T5-8 vs a 780. Here is why the P7+ 780 would perform well – 1) 4.42 GHz – fast is fast! 2) Server can be run with 1 – 1000 LPARs 3) Server could have 1 – 64 cores allocated to any LPAR 4) Server could allocate up to 1280 virtual processors to any LPAR. 5) Power7 (and 7+) have 4 way SMT taking these 1280 virtual processors out to 5120 logical processors. 6) Power7+ has DPO aka Dynamic Platform Optimizer designed to work with the Power Hypervisor (which really is drawn from the company that invented virtualization on the mainframe) to constantly keep processors and memory “local”.

      4) I’ll be honest with you about Oracle’s price/performance claims. They are “used car” like. “Snake oil salesman” like. Said another way – simply lacking credibility. Result after result after result misrepresents fact. I’ve seen this since I joined IBM (I’m now at a BP) from Sun. They would say things like “Our 2 processor outperforms IBM’s 2 processor”. When you looked under the covers that would be something like 2 x 8 core T1/2 cores (that is two processors by Sun’s definition where the Power result is a single socket using a DCM or a dual core module which IBM often refers to as 2 processors. Do you see where I’m going with this? Yes, your two vs my two is really only the same by name only and not the actual end result. Oracle likes to use short term licenses like 1, 2 or 3 year with web support or other unrealistic options. I work with A LOT of Oracle customers and they use either perpetual licenses based on core or named user. I looked at the IBM result for the 595 and 780 – yes, they were using perpetual licenses for DB2. Maybe that is IBM’s problem for handicapping themselves, I don’t know. But for Oracle cheerleaders to jump up and down using data that is non-standard at best and purposefully misrepresenting at worst……you simply lose credibility.

      What benchmarks and all of these “my server is bigger and badder than your server” misses out on is that this is good theater but what is good for customers is what the Total Cost of Ownership. It is hard for competitors to like each other let alone acknowledge the competitors have a good product. My ten years at Sun gave me an appreciation for the people and the technology. Having worked the past 7 years with AIX and Power technology has opened my eyes that there are other compelling technologies. What you don’t realize is the flexibility and efficiency available with the Power Hypervisor and the virtualization suite called PowerVM. It is consistent across the entire family. Yes, they run fast but the secret sauce is how Power is able to consolidate a massive number of workloads from x86, Itanium and SPARC due to the efficiency. This has the benefit and affect of consolidating and thus lowering software licensing not to mention the overall TCO.

      Oracle technology is simply not in the same league. Doesn’t mean you are bad. Doesn’t mean you aren’t better than x86 – I respect HP-UX, Solaris and AIX – all have a great and proud history. Solaris & SPARC sellers and customers are the result of management that long lost it’s direction at Sun. They now have an executive team that is marketing driven who believe they can holler and make claims of truth but in fact it is all in a effort to drive software licensing for the core of the company – software! Now, SPARC & Solaris has become the “legacy” products they always sold against for years. It is clear in the marketplace that Oracle markets the Solaris / SPARC mindshare in a effort to sell more x86 products (ala Exa*. It is these products that drive the licensing / maintenance revenues desired by Larry and team. I get it….it’s a business to them. To some of us though it is a passion!

      Brett Murphy

      April 30, 2013 at 7:47 pm

    Leave a Reply

    Fill in your details below or click an icon to log in: Logo

    You are commenting using your account. Log Out / Change )

    Twitter picture

    You are commenting using your Twitter account. Log Out / Change )

    Facebook photo

    You are commenting using your Facebook account. Log Out / Change )

    Google+ photo

    You are commenting using your Google+ account. Log Out / Change )

    Connecting to %s

    %d bloggers like this: