[Search for users] [Overall Top Noters] [List of all Conferences] [Download this site]

Conference spezko::cluster

Title:+ OpenVMS Clusters - The best clusters in the world! +
Notice:This conference is COMPANY CONFIDENTIAL. See #1.3
Moderator:PROXY::MOORE
Created:Fri Aug 26 1988
Last Modified:Fri Jun 06 1997
Last Successful Update:Fri Jun 06 1997
Number of topics:5320
Total number of notes:23384

5264.0. "OVMS V6.2-1h3 and 8400 5/440 6gb memory questions." by SALEM::HARRIS_M () Fri Mar 21 1997 14:10

    Hi,
    
    I've got a couple of questions about a cluster I'm configuring.
    
    I'm running OVMS Alpha V6.2-1H3 in a cluster environment. I've got
    5 8400's (2 cpus each, model 5/440s) and 2 8400s (4 cpus each, model
    5/440s). Each 8400 has 6GB of memory. There are 4 16node scoo8s, 12
    pair of HSJ50s.
    
      1) What is the maximum size system disk that can be configured off
        of the HSJ50 to be used as a single cluster disk.
          
      2) What is the maximum memory size supported under V6.2-1h3.
    
      3) The SPD for V6.2-1H3 states that VMS doesn't support dumping more
         than 4GB of memory. Does this mean that there is a 4GB memory
         limit for V6.2-1H3 or is this just for Full Dumps.
    
    			
    		         Any input would be greatly appreciated.
    	
    					Matt
    
T.RTitleUserPersonal
Name
DateLines
5264.1BACHUS::VERCAMMENYou never walk alone...Mon Mar 24 1997 07:2419
>      1) What is the maximum size system disk that can be configured off
>        of the HSJ50 to be used as a single cluster disk.

	HSOF V5.0 allows you to use disk drives with capacities of up to 120 GB

>      2) What is the maximum memory size supported under V6.2-1h3.

	For the Alphaserver 8400 - 14GB of memory is supported by OpenVMS

>      3) The SPD for V6.2-1H3 states that VMS doesn't support dumping more
>         than 4GB of memory. Does this mean that there is a 4GB memory
>         limit for V6.2-1H3 or is this just for Full Dumps.

	That's indeed what the SPD says, but I don't know why.



	\Jan
5264.2And here's why...MILORD::BISHOPThe punishment that brought us peace was upon HimMon Mar 24 1997 14:0951
>>      3) The SPD for V6.2-1H3 states that VMS doesn't support dumping more
>>         than 4GB of memory. Does this mean that there is a 4GB memory
>>         limit for V6.2-1H3 or is this just for Full Dumps.
>
>	That's indeed what the SPD says, but I don't know why.
    
    
    Yes indeed, and ehere's the reason:
    
    BUGCHECK uses the bootdriver to perform its I/O. Bootdrivers use a data
    structure that defines the virtual address and size of the buffer,,
    plus the physical address and size (typeically a page!) of each
    fragment of the buffer.
    
    Now the size of each field in this structure is a longword, so
    addresses and sizes can't exceed 4GB. Sizes aren't a problem, since
    I/Os are done by BUGCHECK in 128-block segments. 
    
    The VA is never a problem either, as BUGCHECK has some PTEs reserved at
    boot time that give it addresses in S0/S1 space. It's just the PA that 
    hurts. 
    
    Now different bootdrivers use different parts of this data structure.
    CNBTDRIVER (the most common) uses the VA, so 4GB is not a problem
    there. KDMBTDRIVER uses the PA, so your dump isn't going to be much
    use.
    
    When those words were put into the SPD, it was late in the day for
    V6.2, and all I knew was that dumps over 4GB were broken. I didn't then
    know exactly which combinations of full vs selective and bootdriver
    gave or avoided the problem. For safety's sake, we simply said none
    were supported (but like many things we say aren't supported, YMMV).
    
    In V7.1, I changed BUGCHECK to handle systems over 4GB by copying all
    data to a buffer that I knew was below 4GB and writing it from there.
    Later in the V7.1 development cycle, as DOSD was being designed, we
    were able to define the exact combination of dumps and drivers that hit
    the problem, hence today I can sound like I know what I'm talking
    about.
    
    In a future release, the bootdrivers will be changed to pass 64 bits of
    address around, since new platforms with discontiguous memory could
    exist with *no* memory below 4GB. At that time, BUGCHECK will take
    advantage of those changes.
    
    Richard Bishop
    OpenVMS Alpha BUGCHECK/SDA maintainer
    
    ps: I don't normally frequent this conference (Steve H. forwarded the
    base note to me.) So if my answer raises more questions, someone will
    need to alert me. :-)