Aggregate Storage And Multi-Threading/Multi-Core Systems

Please pardon if this question has been asked before, but the Forum search is not returning any relevant results.
We are in the process of purchasing hardware for an 11.1.2 Essbase environment. We are going 64-bit, on Windows 2008, with either 32 GB or 64 GB of system RAM. The debate we are having is the number of CPUs and cores per CPU. We have not built any ASO databases as of yet, but we plan to launch a major BSO to ASO conversion project once 11.1.2 is off the ground here.
Historically, with BSO, we did not see performance improvements significant enough to justify the cost of additional CPUs when we ran calcs on multi-CPU systems vs. single or dual CPU systems, even when the settings and design should have taken the most advantage of BSO's multi-threading capabilities. However, it would seem that ASO's design may be able to make better use of multi-core systems.
I know that there are a lot of factors behind any system's performance, but in general, is ASO in 11.1.2 written well enough to make it worthwhile to consider, say, a four CPU, total 16 core system vs. a 2 CPU, total four core system?

Grand central dispatch - infancy, not really doing its job, and I don't think apps have to be specifically written for HT, but they do have to not do things that they use to - prevent threads from going to sleep! or be parked.
high usage is not necessarily high efficiency. often the opposite.
Windows 7 seems to be optimized for multi-core thanks to a lot of reworking. Intel wants and knows it isn't possible to hand code, that the hardware has to be smarter, too. But the OS has a job, and right now I don't think it does it properly. Or handle memory.
Gulftown's 12MB cache will help, and over all should be 20% more efficient doing its work.
With dual processors, and it doesn't look like there are two quick path bridges, data shuffling has led to memory thrashing. Use to be page thrashing with not enough memory. Then core thrashing but having the cores, but not integrated (2008 is often touted as being greatest design so far, but it was FOUR dual-cores, 2009 was the first with a processor that really was new design and (native) 4-core.
One core should be owned by the OS so it is always available for its own work and housekeeping.
The iTunes audio bug last year showed how damaging and not to implement code and how a thread could usurp processing and add a high cpu temperature while basically doing nothing, sort of a denial of service attack on the processor - those 80*C temps people had.
All those new technology features under development and not like OpenCL, GCD and even OpenGL are tested, mature but rather 1.0 foundation for the future. A year ahead of readiness.

Similar Messages

  • SAP XI on a multi-core, multi-thread server?

    Hi Everyone!
    Can anyone tell me whether SAP XI can run on a Server whose processor is multi-core, multi thread (4 cores, 8 threads per core, 32 threads)?
    Thanks in advance!
    Warm regards,
    Glenn

    Hi Ravi,
    Thanks for your reply! The server in question is SUN Solaris. What sort of confguration needs to be done on both the OS side and on XI? Is there a SAP note I can follow?
    Warm regards,
    Glenn

  • Indesign and multi threaded cpu support??????????

    When the indesign will finally support multi-core processors?????? It is tragic that adobe dont have any support for multi-core cpu and that this products cost thousends $$$$$!!!!!!!!!!!!!!! Only premiere and some photoshop function support multi-core cpu.

    If you are good freelancer you can get money spent on CS back in a month. So I would not call this software expensive. What concerns multi threaded cpu support I believe that programming tools as well as operating systems need to evolve. I am sure that we will get better support in this matter in close future.

  • Multi-Thread application and common data

    I try to make a multi-Thread application. All the Threads will update some common data.
    How could I access the variable �VALUE� with the Thread in the following code:
    public class Demo {
    private static long VALUE;
    public Demo(long SvId) {
    VALUE = 0;
    public static class makeThread extends Thread {
    public void run() {
    VALUE++;
    public static long getVALUE() {
    return VALUE;
    The goal is to get the �VALUE� updated by the Thread with �getVALUE()�
    Thanks for your reply
    Benoit

    That code is so wrong in so many ways......
    I know you're just experimenting here, learning what can and can't be done with Threads, but bad habits start early, and get harder to kick as time goes on. I am going to give a little explanation here about what's wrong, and what's right.. If you're going to do anything serious though, please, read some books, and don't pick up bad habits.
    Alright, The "answer" code. You don't use Thread.sleep() to wait for Threads to finish. That's just silly, use the join() method. It blocks until the threads execution is done. So if you have a whole bunch of threads in an array, and you want to start them up, and then do something once they finish. Do this.
    for(int k=0; k<threads.length; k++) {
      threads[k].start();
    for(int k=0; k<threads.length; k++) {
      threads[k].join();
    System.out.println("All Threads Done");Now that's the simple problem. No tears there.
    On to the java memory model. Here where the eye water starts flowing. The program you have written is not guarenteed to do what you expect it to do, that is, increment VALUE some amount of time and then print it out. The program is not "Thread Safe".
    Problem 1) - Atomic Operations and Synchronization
    Incrementing a 'long' is not an atomic operation via the JVM spec, icrementing an int is, so if you change the type of VALUE to an int you don't have to worry about corruption here. If a long is required, or any method with more then one operation that must complete without another thread entering. Then you must learn how to use the synchronized keyword.
    Problem 2) - Visiblity
    To get at this problem you have to understand low level computing terms. The variable VALUE will NOT be written out to main memory every time you increment it. It will be stored in the CPUs cache. If you have more then one CPU, and different CPUs get those threads you are starting up, one CPU won't know what the other is doing. You get memory overwrites, and nothing you expect. If you solve problem 1 by using a synchronized block, you also solve problem 2, because updating a variable under a lock will cause full visiblity of the change. However, there is another keyword in java.. "volatile".. A field modified with this keyword will always have it's changes visible.
    This is a very short explaination, barely scratching the surface. I won't even go into performance issues here. If you want to know more. Here's the resources.
    Doug Lea's book
    http://java.sun.com/docs/books/cp/
    Doug Lea's Site
    http://g.cs.oswego.edu
    -Spinoza

  • Enterprise User and Multi Thread Server

    Hi,
    We are going to build a system which uses a configuration:
    10g2
    Enterprise User
    Multi Thread Server
    Client apps accesses db over JDBC/SSL
    Could some please share experience about issues regarding
    using Enterprise User and Multi Thread Server ?
    Is MTS transparant for Enterprise User authentication ?
    Regards,
    Cezary

    If you build simpserv with -t, and set MIN and MAXDISPATCHTHREADS, you
    should have an example of a multithreaded server.
         Scott
    Johan Philippe wrote:
    >
    We have been going to the documentation on multi-threading and contexting in servers
    on Tuxedo 7.1.
    And the impression so far is that apart from the function definitions there is
    not much information around.
    Didn't even find a simple example of a multi-threaded server in the Bea refs yet.
    Does anyone know/have such an example?
    And is anyone using multi-contexting in a server, because the limitation that
    only server-dispatched
    thread get a context puts quite a limitation on its usefullness.

  • "SQL-02103 Inconsistent cursor cache." error on dual processor multi threaded system

    I have a C++ with Pro*C++ COM object running on a Win2000 OS on
    a dual processor Middleware. The program is being called by a
    service in a multi threaded way. An oracle error like ORA-2103
    and SQL-02103 will be generated after running the program. We've
    tried putting mutex on every SQL statement in Pro*C and it works
    but the systems became very very slow. We concluded that the
    error might be cause by a collision of threads accessing the
    same Table or different tables having foreign constraints. This
    is so because the error keeps coming from a statement accessing
    Table A and following errors comes from a Table B having a
    foreign key on Table A.
    Our only solution now is to put Mutex on Statements accessing
    same tables or tables with foreign constraints but this will
    make our system run very very slow. I dont know if the problem
    is in our Oracle Database because we already upgraded our system
    from 8.1.6 to 8.1.7. The wierd thing on this is that the system
    is running smoothly without oracle errors on a SINGLE PROCESSOR
    Middleware. Hope you can give me any insights on this problem.

    I have a C++ with Pro*C++ COM object running on a Win2000 OS on
    a dual processor Middleware. The program is being called by a
    service in a multi threaded way. An oracle error like ORA-2103
    and SQL-02103 will be generated after running the program. We've
    tried putting mutex on every SQL statement in Pro*C and it works
    but the systems became very very slow. We concluded that the
    error might be cause by a collision of threads accessing the
    same Table or different tables having foreign constraints. This
    is so because the error keeps coming from a statement accessing
    Table A and following errors comes from a Table B having a
    foreign key on Table A.
    Our only solution now is to put Mutex on Statements accessing
    same tables or tables with foreign constraints but this will
    make our system run very very slow. I dont know if the problem
    is in our Oracle Database because we already upgraded our system
    from 8.1.6 to 8.1.7. The wierd thing on this is that the system
    is running smoothly without oracle errors on a SINGLE PROCESSOR
    Middleware. Hope you can give me any insights on this problem.

  • 64-bit and Support for Multi-Core Computers

    I recently bought one of the new 12-core Mac Pro's from Apple, but was disappointed to see the lack of 64-bit support and optimization for multi-core computers within InDesign. When producing magazines with hundreds of pages and hundreds of fonts, the need for multi-core support and 64-bit processing are vital.
    Can anyone at Adobe confirm that they will implement this in CS6? It's long overdue.

    InDesign is a really complicated program. 64 bit support is of marginal use of InDesign. It would be faster (as is the case using ID Server 64 bit on Windows), but I'm not sure how much of a difference that would make for the average desktop user.
    Adobe has done a lot of multi-threading work under the hood for CS5 (and pdf export is one of the first fruits of that). What further support they might add for CS+, is anyones guess, but it'll be easier now that a lot of the preliminary work is done. Each feature would be a separate effort, so if there's something SPECIFIC you'd like to see multi-threaded, I suggest you write that in a feature request. The more detailed you can be about it, the better!
    Harbs

  • How should implement multi-thread in single-threaded Operating system using

    How should implement multi-thread in single-threaded Operating system using java?
    Java supports "Multi-threading".Is there is any way run the multiple threads (Implementing multi threading) using java in a Single-threaded Operating system (That is the operating system does not support for multi-threading).

    Previous questions from OP suggest they are using J2ME, so the question might be possible.
    806437 wrote:
    How should implement multi-thread in single-threaded Operating system using java?
    What is the actual question/problem?
    A java app doesn't do threads or not do threads. It uses classes. The VM does threads.
    So if you have a platform that does not have threads and you want to support the thread class then the VM, not a java app, must provide some pseudo mechanism, such as green threads, to support that.
    If your question is about java code and not the VM then you must build a task engine and insure that the tasks are of short enough duration that it is an effective use for your system.

  • Multi thread and deadlocks

    I am running dataload from a text file to MS SQL SRVR database. The jobs are running in multi thread mode. I am getting deadlock errors with threads.
    each thread is supposed to delete records.
    with single thread mode log file order is as below
    1. get product to be deleted A
    2. delete A
    3. get product to be deleted B
    4. delete B
    5. get product to be deleted C
    6 delete C
    with multi thread mode
    1. get product to be deleted A
    2. get product to be deleted B
    3. get product to be deleted C
    4. get product to be deleted D
    5. delete A
    6. get product to be deleted E
    7. get product to be deleted F
    8. deadlock while deleteing the product B
    Any clues what is going wrong here?
    Here is the code for the delete method:
    ================================
    PreparedStatement prepStatement = null;
    System.out.println("get product to be deleted " + product);
    prepStatement = conn.prepareStatement("delete from MYTABLE where product = ? ");
    prepStatement.setString(1, product);
    prepStatement.executeUpdate();
    ================================
    Please note that only unique data is sent to each thread.
    Any help is greatly appreciated.
    Thanks

    Thanks for the responses. I need to do update/insert new records after delete operation is completed. Later at the end of the db changes, i close the connection.
    From my understanding deadlocks happen when both threads try to touch the same record same time. But, i am sending unique records for each thread. Unless these is somethign wrong with the query, it should always return unique values.
    This is the query that sends unique records for each thread in sql srvr.
    select distinct top ? columnA, columnB from MyTable where processed = 'FALSE' and columnB = ? "
    This should be equivalent of oracle query
    select distinct columnA, columnB from MyTable where columnB=? rownum > ? and rownum <= ?
    if the top in sql srvr is equivalent to rownum in oracle then i always get distinct records and there should be no deadlocks.
    Any help?
    Thanks

  • System() from a multi-threaded app

    The man page for system(3S) states that its MT-Level is "unsafe." Nonetheless we need to call system() from a multi-threaded application. At present we "just go ahead and do it," and of course it doesn't work well. I'm not entirely sure of the details of the failure -- it's more a colleague's baby than mine -- but I believe the system() call often fails to complete ("hangs").
    Is there a thread-safe equivalent-or-replacement for system()? Surely the need to launch other applications from a multi-threaded application is commonplace today? How is it done safely-and-correctly?

    Hi,
    system() is not thread safe because it internally uses vfork() system call which in turn is not MT-safe . When you do a vfork()/fork1() in threaded applications only the thread which issues the call gets duplicated in the child process , which could result in dead lock problems . Solution of running system() under the protection of a lock may not help to avoid this problem.
    If you're using Solaris thread library , fork() + exec() is a safe substitute for system() . If you are using Posix library , use pthread_atfork() to make sure that you handle the locks properly before/after forking.
    Hope this helps.
    Thanks,
    Prajeesh

  • Memory leaks and multi threading issues in managed client.

    In our company we use a lot of Oracle, and after the release of the managed provider we migrated all applications to it. First the  things were very impressive : the new client was faster, but after some days applications that uses 100MB with old client goes to 1GB and up. The memory is not the only issue, we use a lot of multi threading, and we experience connection drops and not disposal, after 1 days working one of the application had over 100 sessions on the server. I think there is something wrong with connection pool and multi threading.
    Is someone experience same problems.
    Yesterday we went back with unmanaged provider. Now things are back to normal.

    connection drops: did you try to use "Validate Connection=true" parameter in your connection string?
    the new client was faster: are you sure with this statement? Even in 64bit environment? I got quite serious performance problems when running application under 64bit process: https://forums.oracle.com/thread/2595323

  • Can Acrobat 9 Multi-Thread (use all my cpu cores)

    i have several 700+ page thick PDF documents i want to run OCR on. i can batch run them just fine, however i cannot get adobe to use more than one core of my 8 core iMac. so the process takes needlessly long because 87% of my CPU power is just sitting idle. i also cannot do anything with adobe while it is running its OCR program.
    is there a way to force acrobat to be multi-threaded?
    is there a way to use it for other things while its working?

    Nope, no version of Acrobat is a multithreaded application. See the first bullet point: http://livedocs.adobe.com/acrobat_sdk/9.1/Acrobat9_1_HTMLHelp/DevFAQ_Introduction.21.16.ht ml

  • Oracle Multi-threaded libraries and Forte R3

    Hello all,
    I heard through the grapevine that it would be possible
    to use Oracle 7.3 multi-threaded support with Forte 3.0G.
    (thus making Load-Balancing obsolete in this case)
    Has anyone given a try at this assumption ?
    If yes, on what platform and what versions ?
    Thanks,
    Vincent Figari
    You don't need to buy Internet access to use free Internet e-mail.
    Get completely free e-mail from Juno at http://www.juno.com
    Or call Juno at (800) 654-JUNO [654-5866]
    To unsubscribe, email '[email protected]' with
    'unsubscribe forte-users' as the body of the message.
    Searchable thread archive <URL:http://pinehurst.sageit.com/listarchive/>

    Hi,
    It is available In R3.0.F.2 also. I tested it this release and it works
    very well. But, you need to make a design pattern on it : A Oracle
    Session can not be shared between several transactions throught threads.
    So You need to manage a pool of DBSessions.
    Hope this Helps,
    Daniel Nguyen
    Freelance Forte Consultant
    Vincent R Figari wrote:
    >
    Hello all,
    I heard through the grapevine that it would be possible
    to use Oracle 7.3 multi-threaded support with Forte 3.0G.
    (thus making Load-Balancing obsolete in this case)
    Has anyone given a try at this assumption ?
    If yes, on what platform and what versions ?
    Thanks,
    Vincent Figari
    You don't need to buy Internet access to use free Internet e-mail.
    Get completely free e-mail from Juno at http://www.juno.com
    Or call Juno at (800) 654-JUNO [654-5866]
    To unsubscribe, email '[email protected]' with
    'unsubscribe forte-users' as the body of the message.
    Searchable thread archive <URL:http://pinehurst.sageit.com/listarchive/>-
    To unsubscribe, email '[email protected]' with
    'unsubscribe forte-users' as the body of the message.
    Searchable thread archive <URL:http://pinehurst.sageit.com/listarchive/>

  • Java multi-thread Applet and Multi-processor

    Hello,
    I have a JAVA applet which work fine on a mono-processeur machine and not fine at all on multi-processors.
    It use multi thread (for reading on socket) and some times, request are lost.
    Is that a matter of programming or a JVM bug ?
    This happens on Linux and Windows NT bi-processors.
    We hare using JAVA 1.3.0_02
    Thanks for your help
    [email protected]

    I have already have a look to javax.swing.SwingUtilities.invokeLater() but (I think) it don't work for my need.
    A graphic applet which create 2 threads : one for reading and one for writing on a socket.
    Request sent to the server are memorized in a vector, and once answer is received, the request is remove from vector and treated. Access to the list is protected by synchronize.
    Everything works fine one several plateforme (linux, windows 98/2000/NT, Solaris). The only problem is on multi-processor : request are lost !

  • Multi-Instrument + Multi-Thread problem found

    Right.....I have done a lot of research on why logic was not multi-threading correctly and have narrowed it down to one problem...using Kontakt as a multi instrument.
    My Mac is a 2 x 2.26 GHz Quad-Core Intel Xeon. This means that it is capable of distributing work loads over the multiple cores of the CPU to prevent core audio overloads from happening.
    If I run any plug ins in stereo or mono mode, the CPU acts as it should; I will load up a program, add processing and when the load is becoming a strain for the CPU, I use a bus send and work on that, then the workload is redistributed across more of the cores by logic, allowing me to continue working.
    However, I need to use Kontakt as a multi instrument. I usually have 3 instances of Kontakt open and on each one of these, I usually have a minimum of 10 outputs being sent to the aux channels of logic. Therefore using Kontakt as a stereo instrument is not an option, as I would have to open around 30 instances!
    Here is the problem; say for example, in Kontakt I send a kick drum out of Kontakt into AUX 3 + 4. The AUX channel in logic will being to show output. I can now EQ this single kick drum seperate from any other sound in logic.
    If I began to add processor heavy plug ins to this AUX output, rather than share the workload across the cores as it should and does with stereo instruments, logic now only uses a single core and it will core audio overload very quickly.
    As I said, this only happens when I set up Kontakt as a multi instrument. The main problem is when I begin to have many different outputs from my 3 instances of Kontakt, as by this stage, many different sounds will have processing. All 3 multi instruments and their processing are then controlled by the single processor and the other 7 are almost inactive! It therefore becomes impossible to use Kontakt and I have to switch to stereo instances. This happens for all multi-instruments not just Kontakt.
    I have just purchased Kontakt 4, I am upgrading to OSX 10.6 (Snow Leopard) and am updating Logic to v9.1. (From v9.0) Will this help at all? Is there an answer to this problem or will I just have to open 30 stereo and mono instances of Kontakt to continue working!
    I have included a link to a ZIP file with two photos of the problem. In the photo named MASSIVE. I have one instance of massive running as a stereo instrument. It is sent to a BUS where I applied the WAVES L316 Limiter. I did this multiple times and if you can see on the system performance (Top right) Logic is sharing the workload over the processors as it should.
    In the second picture named KONTAKT MULTI - I have done the same thing, but I have opened Kontakt as a multi instrument, sent a kick drum to AUX 2 and then forwarded this to a send where I applied the same WAVES L316. You can see on the system performance meter (Top Right) that Logic now does not re-distribute the workload, it is using one core and anymore plug ins applied to the AUX or bus channels relating to Kontakt will now make it core audio over load, even though now I am using far less processing.
    PHOTOS: http://www.sendspace.com/file/og8a94

    Ditto
    Using Pro 8 Studio, I had this problem after installing Snow Leopard.
    Invested £159 in Pro 9 update, same problem.
    This has stopped most of my work since my external WK4 sounds are superior to the included software sounds.
    As suggested, will now report since since it obviously seems to be a general problem.

Maybe you are looking for