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

Conference mvblab::alphaserver_4100

Title:AlphaServer 4100
Moderator:MOVMON::DAVISS
Created:Tue Apr 16 1996
Last Modified:Fri Jun 06 1997
Last Successful Update:Fri Jun 06 1997
Number of topics:648
Total number of notes:3158

428.0. "Mixing memory options..." by HGOVC::NANDAN () Thu Jan 23 1997 06:59

T.RTitleUserPersonal
Name
DateLines
428.1Memory RulesVYGER::HENDERSONLThu Jan 23 1997 09:1613
428.2You can mix the orderable part numbersJULIET::HATTRUP_JAJim Hattrup, Santa Clara, CAThu Jan 23 1997 18:215
428.3Thanks, and...HGOVC::NANDANThu Jan 23 1997 20:1114
428.4NPSS::GLASERSteve Glaser DTN 226-7212 LKG1-2/W6 (G17)Thu Jan 23 1997 23:3249
    The 4100 uses the size of the first memory option to decide the overall
    "system configuration" size.  Any smaller memory options appear
    replicated some number of times depending on the ratio of the small
    option size to the "system configuration" size.
    
    For example, if you have a single 1 Gbyte option and two 512 Mbyte
    options, then the 512 Mbyte options will show up twice.
    
    In this example:
    
    location 0         to 1.0 Gbyte will be in the big option
    location 1.0 Gbyte to 1.5 Gbyte will be in the first small option
    location 2.0 Gbyte to 2.5 Gbyte will be in the second small option
    
    but in addition
    
    location 1.5 Gbyte to 2.0 Gbyte will be a duplicate of locations
    location 1.0 to 1.5 Gbyte so that writes to location 1.0 Gbyte will
    also change location 1.5 Gbyte.
    
    Similarly location 2.5 Gbyte to 3.0 Gbyte will be a duplicate of
    locations 2.0 to 2.5 Gbyte in the second small option.
    
    The operating system already has support for "bad pages" to avoid using
    memory that it thinks is broken.  In the scenario above, this mechanism
    is used to mark the duplicate memory as "bad" so the operating system
    won't try to use it.  It's not really bad, of course, it just
    represents non-existant memory.
    
    I'm not sure whether the system will have two holes of "bad memory" or
    just one.  One version would be to notice memory from 0 to 2.5 Gbyte
    with one hole from 1.5 to 2.0 Gbyte. Another option would be to notice
    memory from 0 to 3.0 Gbyte with two holes (1.5..2.0 and 2.5..3.0).
    I'm not sure how much memory will be reported (2.0 Gbyte, 2.5 Gbyte or
    3.0 Gbyte depending on whether holes are ignored and whether a hole at
    the end is treated differently).
    
    Note that if the "small" option were even smaller, say 256 Mbyte then
    it would get replicated 4 times (1 usable and 3 duplicate copies) if
    the large option were still 1 Gbyte.
    
    I'm not a 4100 expert and am extrapolating from other stuff in the
    note and of how system's I've seen before work.  I may have got some of
    the details wrong.
    
    P.S. in the 4100 case an "option" is really two cards.  They are
    installed in pairs and both boards of a pair must be the same size.
    
    Steveg
428.5LANDO::CUMMINSFri Jan 24 1997 09:302
    Please check out note 92 and its replies for a lengthy discussion on
    memory holes on Rawhide.