SORT BY:

LIST ORDER
THREAD
AUTHOR
SUBJECT


SEARCH

IPS HOME


    [Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

    RE: iSCSI: keys/parameter dependence



    --- pat_thaler@agilent.com wrote:
    
    > The part that might cause interoperability problems
    > is different 
    > interpretations of which parameters action or
    > acceptance is dependent on
    > others. For instance, does the sentence mean that
    > InitialR2T should be sent
    > before FirstBurstSize and that FirstBurstSize
    > doesn't need a response when 
    > the response to InitialR2T was Yes? 
    
    This is exactly why I wrote that leaving this
    paragraph in the draft is dangerous. I would
    really hate to have to start checking the dependency
    graphs of parameters in order to figure out whether
    I may signal readiness to commit or whether I should
    keep waiting for a response to some key. Please
    let's not go down this road. Boolean omissions
    are relatively easily to deal with compared to
    what this would entail.
    
    > If the sentence is left in, then it should be
    > clarified by adding something 
    > such as:
    > "The definition of a key specifies whether the key
    > is dependent on any other 
    > keys."
    > and if any of our current keys are dependent, add to
    > those keys
    > "This key is dependent on <key names>."
    > If none are currently dependent, it would be
    > reasonable to add to 11 "Currently,
    > no Login/Text Operational keys are dependent."
    
    I don't mind them being dependent much, but if
    values of some may be omitted due to values given
    to others, it makes the negotiation much more
    complicated. Let's not do that. "Each key that
    has been sent should be received" is a nice and
    simple rule that I'm voting for (boolean omissions
    are an exception that can be overcome w/o too much
    effort). 
    
    > -----Original Message-----
    > From: John Hufferd [mailto:hufferd@us.ibm.com]
    >
    > I do not see a problem with the Text, the previous
    > key=value pairs are ones
    > that were in previous PDUs or key=value pairs that
    > were scanned previous to
    > the current pair being scanned.
    
    You just tried to define it, but I don't think
    that the original meaning of "after" in the draft
    was anything close to this. Besides, you're talking
    about "scanning" order. I know that I'll be doing
    batch processing of keys, i.e., first accumulate
    them from all the PDUs that have the C-bit set
    plus the next one that doesn't. So, you're saying
    that "after" is defined wrt my scanning order?
    How is the other side supposed to know how I'm
    planning to "scan" the key=value pairs? I may find
    it easier to do it from the other end, or sort
    first, then process, etc.
    
    Furthermore, even before the C-bit, there has never
    been a requirement to answer one key before the next,
    so specifying that in the case when two dependent
    keys are sent in one command (PDU?) a response to
    one might mean something about the other looks
    very suspicious because logically the same should
    be true then regardless of how the keys were
    received.
    
    > It is not clear
    > that we have to write any
    > thing else to help people understand the concept of
    > previous.
    
    If previous can have two different interpretations,
    then IMHO a little more precision is appropriate.
    You just bundled both together and introduced
    yet another undefined thing called "scan". That
    isn't much help either. Plus I wasn't shooting
    for clarification of "after" as much as for
    getting rid of ordering requirements.
    
    > I think we
    > are straining at a nat here, and it is not worth the
    > effort.
    
    I'm not straining yet, the issue really isn't that
    difficult here. The key reception order should be
    irrelevant because there are no requirements on
    processing order. It should also be irrelevant
    because there are no dependencies currently for
    the operational parameters, nor should they be
    introduced, because they would signifficantly add
    to complexity. Finally, the last sentence is
    dangerous because it seems to imply that there
    could be more legal response omissions than just
    for booleans, and I doubt that anybody wants that.
    
    Therefore I said that I would like the whole
    paragraph removed. Removing is not much of an
    effort, AFAIK.
    
    > I do not see the problem.
    
    I do and I'm getting scared if others don't.
    
    Martins Krikis, Intel Corp.
    
    Disclaimer: these opinions are mine and may not be
                those of my employer.
    
    
    __________________________________________________
    Do You Yahoo!?
    Yahoo! - Official partner of 2002 FIFA World Cup
    http://fifaworldcup.yahoo.com
    


Home

Last updated: Sat Jun 01 03:18:56 2002
10449 messages in chronological order