Unix Strikes Back in latest TPC-C Benchmark Results

Discussions

News: Unix Strikes Back in latest TPC-C Benchmark Results

  1. IBM managed to take UNIX (AIX) to the top of published TPC-C benchmark's non-clustered list. Running on BEA Tuxedo 8.0, IBM ran at 680,613 transactions per minute, beating out Com+ on Windows Server which holds the 2nd and 3rd places. TPC Benchmark C is an on-line transaction processing (OLTP) benchmark.

    IBM used only half of the processors of the previous leader HP Superdome with Windows Server 2003 (IBM 32 Power 4 1.7 GHz, HP 64 Itanium 2 1.5 GHz).

    Read full report at http://www.tpc.org/tpcc/results/tpcc_result_detail.asp?id=103050901

    TPC-C non-clusered results:
    http://www.tpc.org/tpcc/results/tpcc_perf_results.asp?resulttype=noncluster&version=5

    Threaded Messages (28)

  2. What's amazing is how long the previous record (Fujitsu) stood ... almost two years! The Windows 2003 / SQL Server 2000 results were impressive, but only stood for two weeks!

    Peace,

    Cameron Purdy
    Tangosol, Inc.
    Coherence: Easily share live data across a cluster!
  3. Key words to note: IBM used BEA's Tuxedo ...

    another testimony to Tuxedo's dominance in the field!
  4. Based on throughput Unix certainly did Good. But looking at cost factor IBM system costs about $7,574,961(US) and HP system costs about $6,453,432(US). So Windows certainly beat them in cost.
  5. Balaji >>
     Based on throughput Unix certainly did Good. But looking at cost factor IBM system costs about $7,574,961(US) and HP system costs about $6,453,432(US). So Windows certainly beat them in cost.
    <
    Well, all is relative. The Windows system had twice as many processors. So, if you go and buy something that is priced by the number of processors, then I bet the numbers look a bit different.
  6. IBM processor has a dual core[ Go to top ]

    What that means is there are two CPU cores inside a single CPU packaging.

    - An industry first.

    Sun & Intel are also on this path in near future.

    IBM L3 cache is also humongous.

    This is a very good CPU.

    Chameleon
  7. <tero>
    Well, all is relative. The Windows system had twice as many processors. So, if you go and buy something that is priced by the number of processors, then I bet the numbers look a bit different.
    </tero>
    How? The report lists total server cost. There is no relative pricing. According to that report, A 64 processors machine with Windows and other software costs much less compared to IBM's Box with 32 processors and other software.
  8. Relative costs[ Go to top ]

    I believe Tero was referring to the fact that many middleware products (appservers, TPs) are licenced per CPU, not per machine. In other words, if you're running a wintel box with 2 processors you need to buy 2 licences for the middleware. If you're running a UNIX box with 1 processor, you only need to buy 1 licence. See the difference?
  9. In this case, there are no savings.

    Each CPU is counted as two for determining licensing costs.

    Chameleon
  10. Even though the IBM Power 4 processor has two cores it is actually a single processor and it is count as one in terms of licensing
  11. it doesn't matter[ Go to top ]

    at this (or any realistic) level, the database is on a different machine than the appserver. In fact, if they are not, then there really is no point in paying for a full J2EE application server, because if transactions and persistence are local, then there are easier, faster, and more robust ways to do the things that makes a J2EE server worthwhile on a distributed application.
  12. Re: Relative costs[ Go to top ]

    This is a database benchmark and it includes the costs of all software. So SQL EE x 64 + Datacenter x 64 is included in the Superdome price and DB2 x 32 + Tuxedo x 32 is figured into the price.
  13. IBM processors really good[ Go to top ]

    I think those 1.7 gig ibm coppermine procs are much better than intel 1.5 gig xeons.
  14. I notice there is no one that claims that the TPC-C Benchmark (or all benchmarks) is worthless today!

    Regards
    Rolf Tollerud
  15. <rolf>
    I notice there is no one that claims that the TPC-C Benchmark (or all
    benchmarks) is worthless today!
    <rolf>

    Man, those TPC-C results are worthless!
  16. On the contrary, my position hasn't changed. Even though the new IBM result is refreshing in that it demonstrates that there are still vendors interested in publishing "alternative" platforms ;), the comparative data available is hardly representative of all the major options available to the conscientious technologist.

    In fact, I would go so far as to say that the current state of the standard benchmarks is actually a detriment to Microsoft's public perception: A benchmark publishing predominantly Microsoft results can be easily dismissed by the religious on the other side as unbalanced. However, if Microsoft continues to come back and up the ante over all other major platforms (with a significant representation of those platforms in the results), it gives them a greater legitimacy.

    cramer
  17. Is it worthless for you????
  18. it is so predictable..[ Go to top ]

    My dear Roberto,

    The TPC-C benchmark and the TPC (Transaction Processing Performance Council) is one of the most respected bodys in the business. Just look the member list..they are above suspicion.

    Nevertheless, no matter, it seems to be just a time before MS and Sql Server take all first 10 places on all the benchmarks.

    When that happens, as the way of the world is, everyone will then say that "this TPC results is worthless" or "they are bought by Microsoft", and so on.

    It will not matter to them the least that before MS started to get high scores they often quoted TPC scores to marketing their products.

    Regards
    Rolf Tollerud
  19. but the Price/tpmC[ Go to top ]

    what is cheapest?
  20. but the Price/tpmC[ Go to top ]

    what is cheapest?

    AMD + Linux + open source Java solution

    Peace,

    Cameron Purdy
    Tangosol, Inc.
    Coherence: Easily share live data across a cluster!
  21. what is cheapest?[ Go to top ]

    If you look Top Ten TPC Results by Price/Performance, MS/SQL Server already has all 10 positions for that benchmark..

    Regards
    Rolf Tollerud
  22. it is so predictable..[ Go to top ]

    In short: Do you agree with the TPC result? Do you agree with the results for IBM/Unix that was tested? Yes or No?
  23. it is so predictable..[ Go to top ]

    But I though I have made myself clear?
    Of course I believe the IBM/Unix test!

    But how long time do you think it will stand?

    And how long time do you think it will take until MS take all first 10 places on all the benchmarks?

    Regards
    Rolf Tollerud
  24. it is so predictable..[ Go to top ]

    That's not the question. You are introducing a new element in the discussion. very tipical of you.
  25. What about real app ?[ Go to top ]

    Rolf,

    I don't know what is your database experience with the big three databases ( Oracle, DB2, SQL Server ). Are you DBA for any of those ? Have you worked with all of them ? Are you a database professional at all .If you are you will understand that TPC-C test is not a good benchmark for real systems. In TPC-C you have just couple of tables as far as I remember the index access is just by primary key. Not really your typical enterprise app ? One test that is based on real app is SAP benchmark . MS doesn’t perform very well there . Why is that ? If their stuff is good why it cannot get really good results with the most popular ERP system ? It’ real app after all.

    Regards,
    Alex
  26. What about real app ?[ Go to top ]

    <borisov>
    I don't know what is your database experience with the big three databases ( Oracle, DB2, SQL Server ). Are you DBA for any of those ? Have you worked with all of them ? Are you a database professional at all .If you are you will understand that TPC-C test is not a good benchmark for real systems. In TPC-C you have just couple of tables as far as I remember the index access is just by primary key. Not really your typical enterprise app ? One test that is based on real app is SAP benchmark . MS doesn’t perform very well there . Why is that ? If their stuff is good why it cannot get really good results with the most popular ERP system ? It’ real app after all.
    </borisov>

    This link contradicts what you are saying about Windows with SQL Server 2000
    Check it out
    http://www.sap.com/benchmark/
  27. How ??[ Go to top ]

    How ??

    SAP - Standard Application Benchmark Two-tier Client/Server Release 4

    Rank 1 - 8 . Oracle 9i - 3,942,000.00 - 1,263,000.00
    Rank 7. Db2 - 1,250,000
    Rank 11. - SQL server . - 835,000.00

    What is measured is Steps Per Hour.

    SAP - Standard Application Benchmark Three-tier Sales & Distribution Release 4

    rank 1 - 2 - DB2 - 14,398,000 - 14,139,000
    rank 3 - SQL Server - 7,818,000
    Whar is measured is Steps Per Hour.

    and etc.

    I guess you are looking at the date of certification ( This is the default order ). So you should sort by Steps per Hour because this is what is measured.

    If you go at : http://www.ideasinternational.com/benchmark/bench.html

    You will see all the bechmarks TPC-C , SAP and etc. Check out the SAP bechmarks. Here they are sorted by what us measured in each of them not by the date of certification.

    Regards,
    Alex
  28. Customers Win[ Go to top ]

    Amazing results. What a set of options all these
    enterprises going to have. Run HP/Windows/SQL
    achive great performance or run IBM/Unix/DB2(Oracle)
    get greater or comparable speed.

    The bottom line <QUOTE>backend rocks</QUOTE>,
    DO NOT under utilize database specific features by using slower
    standard middle tier frameworks J2EE,.Net,
    unless you really want portability across databases.

    If you are company's business is not software and only
    interested in getting things done faster and ready to pay
    hefty money to BIG3 database vendors, use the best of
    them.
  29. times they are a-changin[ Go to top ]

    Well, it seems HP/SQLserver/COM+ is back in top again...
    but I am sure unix will strike back, again and again and...
    (PS Have a look at TPC-H and TPC-W.)