SORT BY:

LIST ORDER
THREAD
AUTHOR
SUBJECT


SEARCH

IPS HOME


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

    RE: sector alignment for DataOut PDUs?



    Given that the TCP issues of reassembly are so much more complicated than
    the concatenation issues necessary for target data alignment, is there
    really a problem here?
    
    If there is a problem, can someone explain that in more detail?
    
    Eddy
    
    -----Original Message-----
    From: Buck Landry [mailto:blandry@crossroads.com]
    Sent: Wednesday, February 27, 2002 12:15 AM
    To: Rod Harrison; Paul Koning
    Cc: ips@ece.cmu.edu
    Subject: RE: sector alignment for DataOut PDUs?
    
    
    >>> Also, I think we should consider making the "alignment" any
    multiple of 512 the target wants
    <<<
    
    Well.. if we put no restrictions on it, and I decided my target should only
    handle one data-out pdu per scsi write cmd, I might abuse this and force an
    alignment of (512 * ReallyHugeNumber).  This might demand more resources or
    complexity from the initiator than is wished for, although I can't think of
    any reasons off the top of my head (perhaps I should think harder).  OTOH I
    am fully aware that the target is the entity that is supposed to be cheap n'
    easy to implement...
    
    Perhaps you could limit the alignment size by the target's MaxRecvPDULength?
    Or...  
    
    I believe this "multiple-of" idea also needs to be reconciled with
    MaxBurstSize.  If MaxBurstSize is negotiated by the initiator to 512 bytes,
    then by my understanding the target still must send 512-byte R2Ts for writes
    (and hence, expect 512-byte data-out pdus), even if the target's sector size
    is 4k.
    
    --buck 
    
    
    
    
    -----Original Message-----
    From: Rod Harrison [mailto:rod.harrison@windriver.com]
    Sent: Tuesday, February 26, 2002 5:51 PM
    To: Paul Koning
    Cc: ips@ece.cmu.edu
    Subject: RE: sector alignment for DataOut PDUs?
    
    
    
    	I was talking about DATA-OUT PDUs.
    
    	I would suggest we also restrict any target that wants to
    negotiate for the 512 byte DATA-OUT option to not doing anything
    weird with FirstBurstSize or MaxRecvPDULength. I think it would
    unreasonable to allow a target to require 512 byte aligned
    DATA-OUTs and have those other two parameters not multiples of
    512 bytes.
    
    	How do we handle the case where the iSCSI target node has
    multiple LUNs and not all of them are disks? Perhaps the rule for
    transfers lengths that are not multiples of 512 bytes works OK
    here?
    
    	Also, I think we should consider making the "alignment" any
    multiple of 512 the target wants, not just 512. Otherwise it
    seems to me you are back where you started if you have a disk
    with 4096 byte sectors and the initiator chooses to send you
    DATA-OUT PDUs that are not a multiple of 4k.
    
    	- Rod
    
    -----Original Message-----
    From: Paul Koning [mailto:ni1d@arrl.net]
    Sent: Tuesday, February 26, 2002 11:05 PM
    To: rod.harrison@windriver.com
    Cc: ips@ece.cmu.edu
    Subject: RE: sector alignment for DataOut PDUs?
    
    
    >>>>> "Rod" == Rod Harrison <rod.harrison@windriver.com> writes:
    
     Rod> I have no objection in principle to this but I have a
    concern
     Rod> over non-media access commands. Things like mode select
     Rod> payloads, i.e. defect lists etc are not necessarily 512
    byte
     Rod> multiples. The wording would need to allow "odd" sizes when
    the
     Rod> SCSI transfer length is not a 512 byte multiple.
    
    Are those DataOut or DataIn?  The target can arrange inbound PDU
    sizes
    to be whatever multiple is helpful if it wants to, that isn't the
    case
    I was talking about.
    
    In any case, if there are cases where the total transfer length
    for
    data out isn't a multiple of 512, then the rule would have to
    allow
    the last transfer to have an odd length.  The earlier transfers
    could
    still be constrained if the constraint is enabled.
    
          paul
    


Home

Last updated: Wed Feb 27 13:18:03 2002
8904 messages in chronological order