Conference aosg::osf_tuning

	Correct me if I'm wrong, but the 4/230 is one of the Motorola
	68000 based machines?  If the program were CPU bound, the
	8200 could be expected to be much, much faster.  So, given
	the sparse information, the program is either not CPU bound
	or is doing something to neutralize the advantage.� I'd
	guess the program is I/O or memory bound.

	Much information about the program is needed to offer more
	advice.

	As for the program size; pointers on the 8200 are twice as
	large, longs are twice as large, the standard libraries have
	quite a few years more of features added.  How are you measureing
	the program size?  Is it measureing the same thing on both
	systems?

    Thx for reply. 
    I am trying to get the program from customer. But because of some
    security reasons he is not able to give me.
    
      The program size on DU is 2 times larger than SUN OS. SUN also 
    claims that their OS is 64 bit. So pointer size would be same !
    The program size also should be same becasue that machine also
    might have 64 bit instructions?
    
     thx & regards
    Nagaraj Hegde

    This is all very fishy.
    
    A Sun 4/280 is an age-old machine, among the first Sparcs that existed.
    It should be about 30 times slower than a Turbolaser CPU. Second,
    no version of Solaris that ships right now is 64-bit (2.6 would be).
    But this is SunOS and that decidedly is 32-bits.
    
    Anyway, there is virtually no way that these two machines can be
    the same speed, regardless of what the program is. 
    (cross-answered in DIGITAL_UNIX, where you posted the same).
    

	Even if you can't get the program, a description of its
	function may be enough to offer some clue what it is
	doing.  If I/O bound for example, a poor configuration
	on the 8200 could make it merely as fast as the Sun.
	If the customer has an I/O bound application, then a
	little configuration work can go a long way on a system
	like the 8200 to make much faster.

    Hello Thx for reply.
    
     I have a rough Idea of what this program does. What  the program does
    is it loads the entire oracle database into memory and  some search &
    operation is done on this data. 
    
     The machine is 8200 & the disks are connected in BA356 bix through 
    kzpsc controller.
    
     The benchmarking is done in customer place and I am not able to get
    source code and kernel parameters.
    
     Now probably when they plan for next time benchmarking I can 
    try to get the details.
    
     Thx & regards
    Nagaraj Hegde

	It depends on where you start timing the operation.  If you
	start from the search, then it probably is memory bound.  So
	put more memory on the 8200, it should support more than that
	particular model of Sun can.  If they start timing from when
	the database load starts, then part of it will be I/O bound.

    Hello !
     Thx for reply. Now Since TL is online customer tried on following 
    configuration.
    
      1. DEC 4100 Dual CPU 300Mhz
         2MB cache, 512MB RAM, 2mbps FWD SCSI (on RAID)
         5 * 4GB fast & wide SCSI disk
    
      2. Software Oracle 7.2.3.2
       
     Here the performance on DEC4100 and SUN Ultra 3000 dual CPU system
    with 512 MB RAM 20GB disk on RAID controller of 2MBPS. Oracle Version 
    7.2.3.2 are equal upto 200 users. But when users exceeded 200 
    digital unix peformance is much slower then SUN machine.
      
     I could not get the source from customer.
    
     The benchmark program consists of a driver program which carries out
    the following tasks. 
     
     * Generates a number of input files  with random queue numbers and
       cycle codes for input to some other program.
     * Queries input for concurrent number of processes to be executed.
     * Fires as many child processes (instances) of the application program
       as required, based on input received.
     * Proviedes input from randomly alloted files to each of the forked
       child processes.
     * Logs into results file, time taken  by each of the child processes/
     * Logs into results file, the CPU, memory & IO statistics at the same
       points in time.
    
    
     Overlay feature is not utilized.
    
     The testing is done for I/O intensity.
                       
    
     Thx & Regards
    Nagaraj Hegde  
                                                        

Hi

 You need more system information to look at the tuning?

Is the system swapping?

Is a disk spindle over loaded...

Is the network a bottleneck?

What patches and or tuning parameters are set?

it will be hard for anyone to help with such little data.

    Hello !
     Thx a lot for reply. The customer has agreed to allow us to have 
    a look during his benchmark. After that I will try to tune the system.
    and update in here.
    
     Currently notuning is done & no patches are applied.
    
    
     Thx & Regards
    Nagaraj Hegde