SORT BY:

LIST ORDER
THREAD
AUTHOR
SUBJECT


SEARCH

IPS HOME


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

    Re: iSCSI Boot Last Call - Technical Comments



    Mark,
    
    >I don't think most
    > users will ever want (or need) to see more than a 16-bit LUN.
    
    Having a 32/48/64-bit LUN, I believe, doesn't mean that there are
    more than 2^16 LUNs.  IIRC, the 64-bit LUN address space
    can be sparsely populated, depending on how many levels of
    addressing a target supports and which specific LUs are distributed
    into which level.  So, I cannot agree with this statement.
    
    >>so the boot
    > > client can simply copy the value in the iSCSI PDUs as-is.
    >
    > What customer or user looks at iSCSI PDUs?
    
    A boot client is not a customer, but a very thin code probably running
    on the iSCSI NICs, and working off the DHCP responses.
    --
    Mallikarjun
    
    Mallikarjun Chadalapaka
    Networked Storage Architecture
    Network Storage Solutions
    Hewlett-Packard MS 5668
    Roseville CA 95747
    cbm@rose.hp.com
    
    
    ----- Original Message -----
    From: "Mark Bakke" <mbakke@cisco.com>
    To: "Mallikarjun C." <cbm@rose.hp.com>
    Cc: "Prasenjit Sarkar" <psarkar@almaden.ibm.com>; "Constantine Sapuntzakis" <csapuntz@stanford.edu>;
    <ips@ece.cmu.edu>
    Sent: Wednesday, September 11, 2002 3:29 PM
    Subject: Re: iSCSI Boot Last Call - Technical Comments
    
    
    > Mallikarjun-
    >
    > My comments are embedded.
    >
    > --
    > Mark
    >
    > "Mallikarjun C." wrote:
    > >
    > > > consider an application where a host will be able to be booted from
    > > > a small set of different LUNs, perhaps to change O/S levels in a
    > > > test environment
    > >
    > > How does the DHCP server distinguish which O/S level the client
    > > wants to boot from, to provide the right LUN in the Root Path?
    >
    > The DHCP server doesn't.  In this type of environment, the DHCP server
    > is generally under the control of the same test folks who are booting
    > the servers.  One would manually change the LUN in the DHCP server
    > GUI or config file, and reboot.  The DHCP server doesn't have to do
    > anything special, but it's one more scenario where the LUN field is
    > directly manipulated by a user, which is why I'd like to keep it
    > simple.
    >
    >
    > > If it's a well-known LUN->O/S_rev mapping that the client boot
    > > code uses by convention, then there's no issue with always leaving
    > > the LUN field blank (for 0) in the DHCP response.  What am I missing?
    > >
    > > > I still think that allowing a 16-bit LUN in the string is the
    > > > simplest, least error-prone approach.
    > >
    > > AFAICT, this assumes that the user knows how to figure out which
    > > 16-bits of the 64-bit LUN are to be entered into the DHCP server
    > > file.  I'm not sure that's a valid assumption.
    >
    >
    > Actually, the user only has to worry about this when the 64-bit LUN
    > field is used.  The 16-bit version always ends up in the high
    > bytes of the LUN field in the protocol, and the user never sees
    > this structure, which is really not pretty.  I don't think most
    > users will ever want (or need) to see more than a 16-bit LUN.
    >
    >
    > > My preference is to require 64-bits always (if non-blank), so the boot
    > > client can simply copy the value in the iSCSI PDUs as-is.
    >
    > What customer or user looks at iSCSI PDUs?  Would they copy and
    > paste one from a network analyzer?  This doesn't make sense to
    > me.  End users don't care about PDUs.
    >
    >
    > > --
    > > Mallikarjun
    > >
    > > Mallikarjun Chadalapaka
    > > Networked Storage Architecture
    > > Network Storage Solutions
    > > Hewlett-Packard MS 5668
    > > Roseville CA 95747
    > > cbm@rose.hp.com
    > >
    > > ----- Original Message -----
    > > From: "Mark Bakke" <mbakke@cisco.com>
    > > To: "Prasenjit Sarkar" <psarkar@almaden.ibm.com>
    > > Cc: "Constantine Sapuntzakis" <csapuntz@stanford.edu>; "Jim Hafner" <hafner@almaden.ibm.com>; "David
    Black"
    > > <Black_David@emc.com>; "Duncan Missimer" <duncan_missimer@hp.com>; "Elizabeth Rodriguez"
    > > <erodrigu@Brocade.COM>; "IPS" <ips@ece.cmu.edu>; "John Hufferd" <hufferd@us.ibm.com>;
    <owner-ips@ece.cmu.edu>
    > > Sent: Tuesday, September 10, 2002 2:39 PM
    > > Subject: Re: iSCSI Boot Last Call - Technical Comments
    > >
    > > > It's true that in many cases, zero is all that is needed.  However,
    > > > consider an application where a host will be able to be booted from
    > > > a small set of different LUNs, perhaps to change O/S levels in a
    > > > test environment, or to make a different application available
    > > > on the host.  Having these LUNs mapped behind the same iSCSI target
    > > > and editing the LUN in the root-path string is a lot simpler than
    > > > having several different targets, and changing the iSCSI name in
    > > > the root-path.
    > > >
    > > > I still think that allowing a 16-bit LUN in the string is the
    > > > simplest, least error-prone approach.
    > > >
    > > > --
    > > > Mark
    > > >
    > > > Prasenjit Sarkar wrote:
    > > > >
    > > > > Mark and Costa,
    > > > >
    > > > > If you are so very concerned with entering the value wrong, you can leave
    > > > > the lun value blank. The lun value defaults to zero and you can always use
    > > > > lun mapping in the target to make the boot lun to be zero for the initiator.
    > > > >
    > > > > Analogous to the other issue you brought up, I think this is going to be the
    > > > > common case.
    > > > >
    > > > > Sincerely,
    > > > > Prasenjit
    > > > >
    > > > > ----- Original Message -----
    > > > > From: "Constantine Sapuntzakis" <csapuntz@stanford.edu>
    > > > > To: "Mark Bakke" <mbakke@cisco.com>
    > > > > Cc: "Prasenjit Sarkar" <psarkar@almaden.ibm.com>; "Jim Hafner"
    > > > > <hafner@almaden.ibm.com>; "David Black" <Black_David@emc.com>; "Duncan
    > > > > Missimer" <duncan_missimer@hp.com>; "Elizabeth Rodriguez"
    > > > > <erodrigu@Brocade.COM>; "IPS" <ips@ece.cmu.edu>; "John Hufferd"
    > > > > <hufferd@us.ibm.com>; <owner-ips@ece.cmu.edu>
    > > > > Sent: Tuesday, September 10, 2002 11:17 AM
    > > > > Subject: Re: iSCSI Boot Last Call - Technical Comments
    > > > >
    > > > > > Many people also use emacs/vi for their configuration too, so they have
    > > > > > the same problem Mark has with the GUI. I would agree with Mark on his
    > > > > > point about entry being error-prone.
    > > > > >
    > > > > > -Costa
    > > > > >
    > > > > >
    > > > > > Mark Bakke wrote:
    > > > > > > It would be nice if the GUI could handle this.  However, a DHCP
    > > > > > > GUI (Microsoft, for example), only provides a place to paste in
    > > > > > > the entire root-path option; it won't give you separate fields
    > > > > > > to type in the IP address, LUN, and target name.  So the user is
    > > > > > > left with the job of correctly typing in the root-path option.
    > > > > > > We found (through experience) that the LUN field was particularly
    > > > > > > frustrating, expecially since a 16-bit LUN value actually appears
    > > > > > > in the top characters of the LUN field.  For example, LUN 12 (hex
    > > > > > > 0x0c) would appear as:
    > > > > > >
    > > > > > > 000c000000000000
    > > > > > >
    > > > > > > in the option string.  Getting the wrong number of zeroes before
    > > > > > > or after either causes the option to be refused by a network boot
    > > > > > > program, or causes the wrong LUN to be tried.  Either way, the
    > > > > > > user has to decide what the problem is, fix the LUN string,
    > > > > > > and try again.  It's also a bit non-intuitive since most users
    > > > > > > would expect that the LUN should have been entered as
    > > > > > >
    > > > > > > 000000000000000c
    > > > > > >
    > > > > > > (BTW, I missed a zero counting that last one, and I caught it
    > > > > > > because the length looked wrong comparing it to the previous
    > > > > > > one).
    > > > > > >
    > > > > > > Anyway, this is an extremely easy thing to do to make our users'
    > > > > > > lives easier, and we have no control over adding iSCSI user
    > > > > > > interface capabilities to DHCP servers.
    > > > > > >
    > > > > > > I really think that this is the right thing to do.
    > > > > > >
    > > > > > > --
    > > > > > > Mark
    > > > > > >
    > > > > > > Prasenjit Sarkar wrote:
    > > > > > >
    > > > > > >>
    > > > > > >>
    > > > > > >>
    > > > > > >>
    > > > > > >>Mark,
    > > > > > >>
    > > > > > >>I tend to agree with Jim that machines can help with 8-byte entities.
    > > > > > >>
    > > > > > >>Regarding the prioritization, I do not doubt that your assertion will
    > > > > > >>generally hold true, but at the same time, I feel that we should not
    > > > > make
    > > > > > >>any assumptions about the order of updates to the Discovery and DHCP
    > > > > > >>databases in a boot environment. However, if people insist, I can make
    > > > > the
    > > > > > >>change.
    > > > > > >>
    > > > > > >>Thanks,
    > > > > > >>
    > > > > > >>   Prasenjit Sarkar
    > > > > > >>   Research Staff Member
    > > > > > >>   IBM Almaden Research
    > > > > > >>   San Jose
    > > > > > >>
    > > > > > >>
    > > > > > >>                      Jim Hafner
    > > > > > >>                      <hafner@almaden.i        To:       Mark Bakke
    > > > > <mbakke@cisco.com>
    > > > > > >>                      bm.com>                  cc:       Prasenjit
    > > > > Sarkar <psarkar@almaden.ibm.com>, Duncan
    > > > > > >>                      Sent by:                  Missimer
    > > > > <duncan_missimer@hp.com>, Costa Sapuntzakis
    > > > > > >>                      owner-ips@ece.cmu         <csapuntz@stanford.edu>,
    > > > > Elizabeth Rodriguez
    > > > > > >>                      .edu                      <erodrigu@Brocade.COM>,
    > > > > David Black <Black_David@emc.com>,
    > > > > > >>                                                John Hufferd/San
    > > > > Jose/IBM@IBMUS, IPS <ips@ece.cmu.edu>
    > > > > > >>                                               Subject:  Re: iSCSI Boot
    > > > > Last Call - Technical Comments
    > > > > > >>                      09/10/2002 07:59
    > > > > > >>                      AM
    > > > > > >>
    > > > > > >>
    > > > > > >>
    > > > > > >>Mark,
    > > > > > >>
    > > > > > >>I would rather not go the route your going with the LUN number issue.
    > > > > GUIs
    > > > > > >>can always do the translation to more nibbles under the covers.
    > > > > Besides,
    > > > > > >>if you have a LUN that has only 16bits or less of significant (i.e.,
    > > > > > >>nonzero) digits, you'll need to carefully specify where these digits go
    > > > > in
    > > > > > >>the 8 byte defined SCSI LUN and that depends on the LUN number
    > > > > convention
    > > > > > >>of the target.   So, to avoid that can of worms and the extra
    > > > > descriptions
    > > > > > >>required, I'd suggest leaving this as an 8byte (16 hex nibble) field.
    > > > > > >>
    > > > > > >>On the other point, I have no opinion.
    > > > > > >>
    > > > > > >>Jim Hafner
    > > > > > >>
    > > > > > >>Sent by:        owner-ips@ece.cmu.edu
    > > > > > >>
    > > > > > >>To:        Prasenjit Sarkar <psarkar@almaden.ibm.com>, Duncan Missimer
    > > > > > >><duncan_missimer@hp.com>, Costa Sapuntzakis <csapuntz@stanford.edu>,
    > > > > > >>Elizabeth Rodriguez <erodrigu@Brocade.COM>, David Black
    > > > > > >><Black_David@emc.com>, John Hufferd/San Jose/IBM@IBMUS, IPS
    > > > > > >><ips@ece.cmu.edu>
    > > > > > >>cc:
    > > > > > >>Subject:        iSCSI Boot Last Call - Technical Comments
    > > > > > >>
    > > > > > >>I have only a few technical comments on the boot draft.  The editorials
    > > > > > >>have been sent to the authors.  The draft looks good!
    > > > > > >>
    > > > > > >>Page 4
    > > > > > >>
    > > > > > >>
    > > > > > >>>   The "LUN" field is a hexadecimal representation of the 8-byte LU
    > > > > > >>>   number.  Digits above 9 may be either lower or upper case, and all
    > > > > 16
    > > > > > >>>   nibbles must be present. If the LUN field is blank, then LUN 0 is
    > > > > > >>>   assumed.
    > > > > > >>
    > > > > > >>Since most LUNs are just 16 bits (and many of these are even smaller),
    > > > > > >>I'd like to relax this a bit.  Typing 14 or 15 zeroes plus the actual
    > > > > > >>LUN value into a field in the DHCP GUI is quite error-prone, since in
    > > > > > >>a DHCP server's user interface or /etc/dhcpd.conf, this string will be
    > > > > > >>entered directly by the end user.
    > > > > > >>
    > > > > > >>So in addition to the rules above, how about:
    > > > > > >>
    > > > > > >>- If the LUN field contains four or fewer hex digits, these digits
    > > > > > >>constitute the LUN number from which to boot.
    > > > > > >>
    > > > > > >>Page 5
    > > > > > >>
    > > > > > >>
    > > > > > >>>   It is possible that the port number obtained from the Discovery
    > > > > > >>>   Service may conflict with the one obtained from the DHCP service. In
    > > > > > >>>   such a case, the implementor has the option to try both port numbers
    > > > > > >>>   in the Boot stage.
    > > > > > >>
    > > > > > >>In this case, I think that we should pick one to take precedence,
    > > > > instead
    > > > > > >>of leaving it up to the implementor.  Since the discovery service should
    > > > > > >>have the most up-to-date IP address and port number information, I think
    > > > > > >>that it should be the one that is used.
    > > > > > >>
    > > > > > >>--
    > > > > > >>Mark A. Bakke
    > > > > > >>Cisco Systems
    > > > > > >>mbakke@cisco.com
    > > > > > >>763.398.1054
    > > > > > >
    > > > > > >
    > > > > >
    > > > > >
    > > > > >
    > > > > >
    > > >
    > > > --
    > > > Mark A. Bakke
    > > > Cisco Systems
    > > > mbakke@cisco.com
    > > > 763.398.1054
    > > >
    >
    > --
    > Mark A. Bakke
    > Cisco Systems
    > mbakke@cisco.com
    > 763.398.1054
    >
    
    


Home

Last updated: Thu Sep 12 16:19:00 2002
11826 messages in chronological order