Loading ...
Sorry, an error occurred while loading the content.

static registry format

Expand Messages
  • Sherman Pun
    In section 8.4.6 Other static registry format , there are provisions for other platforms to specify their static registry format when their ref. impl. is
    Message 1 of 10 , Apr 25 1:12 PM
    View Source
    • 0 Attachment
      In section 8.4.6 "Other static registry format", there are provisions
      for other platforms to specify their static registry format when
      their ref. impl. is completed.

      However, in section 8.4.5, the word UNIX OS is referred there.
      The description also implies the current format of the dat.conf is fix,
      i.e. it will start with IA name, API version, ...., and ends with
      a string containing instance data.

      I am questioning the use of the word Unix in section 8.4.5. Is that
      intended for Linux?

      What about OS's that is a derivative of Unix like Solaris but is still
      under development. Section 8.4.6 does provide rooms, e.g. to add elements
      if really necessary to the static registry format.

      I therefore propose to replace the word Unix in section 8.4.5 to Linux.

      --- Sherman Pun
      Sun Microsystems
    • Sherman Pun
      Repost as I could not find this in the dat-discussions alias. ... X-Unix-From: Sherman.Pun@sun.com Fri Apr 25 13:12:28 2003 Date: Fri, 25 Apr 2003 13:12:27
      Message 2 of 10 , Apr 30 9:20 AM
      View Source
      • 0 Attachment
        Repost as I could not find this in the dat-discussions alias.
        ------------- Begin Forwarded Message -------------

        X-Unix-From: Sherman.Pun@... Fri Apr 25 13:12:28 2003
        Date: Fri, 25 Apr 2003 13:12:27 -0700 (PDT)
        From: Sherman Pun <Sherman.Pun@...>
        Subject: static registry format
        To: dat-discussions@yahoogroups.com
        Cc: sherman.pun@...
        MIME-Version: 1.0
        Content-MD5: rGPgZb1Eh4bp81aeTFU0PA==

        In section 8.4.6 "Other static registry format", there are provisions
        for other platforms to specify their static registry format when
        their ref. impl. is completed.

        However, in section 8.4.5, the word UNIX OS is referred there.
        The description also implies the current format of the dat.conf is fix,
        i.e. it will start with IA name, API version, ...., and ends with
        a string containing instance data.

        I am questioning the use of the word Unix in section 8.4.5. Is that
        intended for Linux?

        What about OS's that is a derivative of Unix like Solaris but is still
        under development. Section 8.4.6 does provide rooms, e.g. to add elements
        if really necessary to the static registry format.

        I therefore propose to replace the word Unix in section 8.4.5 to Linux.

        --- Sherman Pun
        Sun Microsystems


        ------------- End Forwarded Message -------------
      • Steffen Persvold
        ... I have also a couple of questions regarding the static registry format. The spec states that the each entry starts with the IA name, but when that provider
        Message 3 of 10 , May 9, 2003
        View Source
        • 0 Attachment
          On Fri, 25 Apr 2003, Sherman Pun wrote:

          > In section 8.4.6 "Other static registry format", there are provisions
          > for other platforms to specify their static registry format when
          > their ref. impl. is completed.
          >
          > However, in section 8.4.5, the word UNIX OS is referred there.
          > The description also implies the current format of the dat.conf is fix,
          > i.e. it will start with IA name, API version, ...., and ends with
          > a string containing instance data.
          >
          > I am questioning the use of the word Unix in section 8.4.5. Is that
          > intended for Linux?
          >
          > What about OS's that is a derivative of Unix like Solaris but is still
          > under development. Section 8.4.6 does provide rooms, e.g. to add elements
          > if really necessary to the static registry format.
          >
          > I therefore propose to replace the word Unix in section 8.4.5 to Linux.
          >

          I have also a couple of questions regarding the static registry format.
          The spec states that the each entry starts with the IA name, but when that
          provider is loaded how can you be sure that it registers itself to the
          dynamic registry with exactly that name ?

          As an example, Scali has made a DAT 1.0 compliant kernel bypass module for
          ethernet. This provider may have mulitple IAs corresponding to the actual
          ethernet interfaces in the machine (for example det0 => eth0, det1 =>
          eth1 and det2 => eth0+eth1) and creation of such devices is not in control
          of the static registry file. Therefore with the current format, one would
          have to have an entry in the dat.conf file for each ethernet device and I
          think this could get messy :

          det0 u1.0 nonthreadsafe default /opt/scali/lib/libdetdapl.so 1.0 ""
          det1 u1.0 nonthreadsafe default /opt/scali/lib/libdetdapl.so 1.0 ""
          det2 u1.0 nonthreadsafe default /opt/scali/lib/libdetdapl.so 1.0 ""
          det3 u1.0 nonthreadsafe default /opt/scali/lib/libdetdapl.so 1.0 ""

          Should we instead open up for wildcards in the static registry ? For
          example :

          det* u1.0 nonthreadsafe default /opt/scali/lib/libdetdapl.so 1.0 ""

          Regards,
          --
          Steffen Persvold | Scali AS
          mailto:sp@... | http://www.scali.com
          Tel: (+47) 2262 8950 | Olaf Helsets vei 6
          Fax: (+47) 2262 8951 | N0621 Oslo, NORWAY
        • Kanevsky, Arkady
          Sherman, 8.4.5 defined the static registry format for all Unixes and Windows. This includes all Unixes: linux, solaris, aix, bsd, hp-ux and others. 8.4.6 is
          Message 4 of 10 , May 9, 2003
          View Source
          • 0 Attachment
            Sherman,
            8.4.5 defined the static registry format for all Unixes and Windows.
            This includes all Unixes: linux, solaris, aix, bsd, hp-ux and others.
            8.4.6 is for embedded and other non-unix and windows platforms.

            If/When somebody develop DAPL for embedded platform and
            if they, for some reason, need a different format we will consider it.

            For now, the expectation is that defined static registry format
            is sufficient and DAPL RI implements it.

            Arkady

            Arkady Kanevsky email: arkady@...
            Network Appliance phone: 781-768-5395
            375 Totten Pond Rd. Fax: 781-895-1195
            3rd Floor http: www.netapp.com
            Waltham, MA 02451-2010 general phone: 781-768-5300

            > -----Original Message-----
            > From: Sherman Pun [mailto:sherman.pun@...]
            > Sent: Fri, April 25, 2003 4:12 PM
            > To: dat-discussions@yahoogroups.com
            > Cc: sherman.pun@...
            > Subject: [dat-discussions] static registry format
            >
            >
            > In section 8.4.6 "Other static registry format", there are provisions
            > for other platforms to specify their static registry format when
            > their ref. impl. is completed.
            >
            > However, in section 8.4.5, the word UNIX OS is referred there.
            > The description also implies the current format of the
            > dat.conf is fix,
            > i.e. it will start with IA name, API version, ...., and ends with
            > a string containing instance data.
            >
            > I am questioning the use of the word Unix in section 8.4.5. Is that
            > intended for Linux?
            >
            > What about OS's that is a derivative of Unix like Solaris but
            > is still
            > under development. Section 8.4.6 does provide rooms, e.g. to
            > add elements
            > if really necessary to the static registry format.
            >
            > I therefore propose to replace the word Unix in section 8.4.5
            > to Linux.
            >
            > --- Sherman Pun
            > Sun Microsystems
            >
            >
            > ------------------------ Yahoo! Groups Sponsor
            > ---------------------~-->
            > Rent DVDs from home.
            > Over 14,500 titles. Free Shipping
            > & No Late Fees. Try Netflix for FREE!
            > http://us.click.yahoo.com/BVVfoB/hP.FAA/uetFAA/W6uqlB/TM
            > --------------------------------------------------------------
            > -------~->
            >
            > To unsubscribe from this group, send an email to:
            > dat-discussions-unsubscribe@yahoogroups.com
            >
            >
            >
            > Your use of Yahoo! Groups is subject to
            http://docs.yahoo.com/info/terms/
          • Kanevsky, Arkady
            Steffen, IA_Name is what Consumer gets from dat_registry_list_providers that extract information from the static Registry. This call need to list all available
            Message 5 of 10 , May 9, 2003
            View Source
            • 0 Attachment
              Steffen,
              IA_Name is what Consumer gets from dat_registry_list_providers
              that extract information from the static Registry.
              This call need to list all available providers that consumer can
              open.

              The dynamic registry has nothing to do with it. It is just for
              redirection and is process specific to keep track of open
              IAs and direct Consumer calls to the right library.

              Arkady

              Arkady Kanevsky email: arkady@...
              Network Appliance phone: 781-768-5395
              375 Totten Pond Rd. Fax: 781-895-1195
              3rd Floor http: www.netapp.com
              Waltham, MA 02451-2010 general phone: 781-768-5300

              > -----Original Message-----
              > From: Steffen Persvold [mailto:sp@...]
              > Sent: Fri, May 09, 2003 3:17 AM
              > To: dat-discussions@yahoogroups.com
              > Subject: Re: [dat-discussions] static registry format
              >
              >
              > On Fri, 25 Apr 2003, Sherman Pun wrote:
              >
              > > In section 8.4.6 "Other static registry format", there are
              > provisions
              > > for other platforms to specify their static registry format when
              > > their ref. impl. is completed.
              > >
              > > However, in section 8.4.5, the word UNIX OS is referred there.
              > > The description also implies the current format of the
              > dat.conf is fix,
              > > i.e. it will start with IA name, API version, ...., and ends with
              > > a string containing instance data.
              > >
              > > I am questioning the use of the word Unix in section 8.4.5. Is that
              > > intended for Linux?
              > >
              > > What about OS's that is a derivative of Unix like Solaris
              > but is still
              > > under development. Section 8.4.6 does provide rooms, e.g.
              > to add elements
              > > if really necessary to the static registry format.
              > >
              > > I therefore propose to replace the word Unix in section
              > 8.4.5 to Linux.
              > >
              >
              > I have also a couple of questions regarding the static
              > registry format.
              > The spec states that the each entry starts with the IA name,
              > but when that
              > provider is loaded how can you be sure that it registers
              > itself to the
              > dynamic registry with exactly that name ?
              >
              > As an example, Scali has made a DAT 1.0 compliant kernel
              > bypass module for
              > ethernet. This provider may have mulitple IAs corresponding
              > to the actual
              > ethernet interfaces in the machine (for example det0 => eth0, det1 =>
              > eth1 and det2 => eth0+eth1) and creation of such devices is
              > not in control
              > of the static registry file. Therefore with the current
              > format, one would
              > have to have an entry in the dat.conf file for each ethernet
              > device and I
              > think this could get messy :
              >
              > det0 u1.0 nonthreadsafe default /opt/scali/lib/libdetdapl.so 1.0 ""
              > det1 u1.0 nonthreadsafe default /opt/scali/lib/libdetdapl.so 1.0 ""
              > det2 u1.0 nonthreadsafe default /opt/scali/lib/libdetdapl.so 1.0 ""
              > det3 u1.0 nonthreadsafe default /opt/scali/lib/libdetdapl.so 1.0 ""
              >
              > Should we instead open up for wildcards in the static registry ? For
              > example :
              >
              > det* u1.0 nonthreadsafe default /opt/scali/lib/libdetdapl.so 1.0 ""
              >
              > Regards,
              > --
              > Steffen Persvold | Scali AS
              > mailto:sp@... | http://www.scali.com
              > Tel: (+47) 2262 8950 | Olaf Helsets vei 6
              > Fax: (+47) 2262 8951 | N0621 Oslo, NORWAY
              >
              >
              > ------------------------ Yahoo! Groups Sponsor
              > ---------------------~-->
              > Rent DVDs from home.
              > Over 14,500 titles. Free Shipping
              > & No Late Fees. Try Netflix for FREE!
              > http://us.click.yahoo.com/BVVfoB/hP.FAA/uetFAA/W6uqlB/TM
              > --------------------------------------------------------------
              > -------~->
              >
              > To unsubscribe from this group, send an email to:
              > dat-discussions-unsubscribe@yahoogroups.com
              >
              >
              >
              > Your use of Yahoo! Groups is subject to
              http://docs.yahoo.com/info/terms/
            • Caitlin Bestler
              ... In fact, I d venture that 8.4.5 is appropriate for *any* OS that supports standard IO and has a local file system suitable for holding configuration files.
              Message 6 of 10 , May 9, 2003
              View Source
              • 0 Attachment
                On 5/9/03, Kanevsky, Arkady wrote:

                >Sherman,
                >8.4.5 defined the static registry format for all Unixes
                >and Windows. This includes all Unixes: linux, solaris,
                >aix, bsd, hp-ux and others.
                >8.4.6 is for embedded and other non-unix and windows
                >platforms.
                >
                In fact, I'd venture that 8.4.5 is appropriate for *any*
                OS that supports standard IO and has a local file system
                suitable for holding configuration files. Given that you
                can fit a fairly large "/etc" directory in a NVRAM part
                my guess is that only the most specialized embedded
                platforms would ever need to consider an alternate format.

                Caitlin Bestler
                http://asomi.com/CaitlinBestler/
              • Walker, Ken
                In the DAT 1.1 specification, look at the dat_provider_init() API (section 8.2.4.1). To be loaded by a 1.1 Registry, a Provider must support this call.
                Message 7 of 10 , May 9, 2003
                View Source
                • 0 Attachment

                  In the DAT 1.1 specification, look at the dat_provider_init() API (section 8.2.4.1).  To be loaded by a 1.1 Registry, a Provider must support this call.  Notice that part of the behavior you are to implement is to register with the dynamic registry only the IA Name passed to you in dat_provider_init().

                   

                  As a side effect, Providers and the Registry must be somewhat matched up—both must be DAT 1.0 or both must be DAT 1.1-or-higher.

                   

                  Regarding wildcards: If by "messy" below you mean the static registry will contain many entries, this is true but desirable.  The point to having separate entries for each IA is to give the administrator the ability to override a single entry at any time.  Wildcarding would give us a shorter static registry file, but complicate the specification of individual overrides.

                   

                  Ken

                   

                   

                  -----Original Message-----
                  From: Steffen Persvold [mailto:sp@...]
                  Sent: Friday, May 09, 2003 2:17 AM
                  To: dat-discussions@yahoogroups.com
                  Subject: Re: [dat-discussions] static registry format

                   

                  I have also a couple of questions regarding the static registry format.
                  The spec states that the each entry starts with the IA name, but when that
                  provider is loaded how can you be sure that it registers itself to the
                  dynamic registry with exactly that name ?

                  As an example, Scali has made a DAT 1.0 compliant kernel bypass module for
                  ethernet. This provider may have mulitple IAs corresponding to the actual
                  ethernet interfaces in the machine (for example det0 => eth0, det1 =>
                  eth1 and det2 => eth0+eth1) and creation of such devices is not in control
                  of the static registry file. Therefore with the current format, one would
                  have to have an entry in the dat.conf file for each ethernet device and I
                  think this could get messy :

                  det0 u1.0 nonthreadsafe default /opt/scali/lib/libdetdapl.so 1.0 ""
                  det1 u1.0 nonthreadsafe default /opt/scali/lib/libdetdapl.so 1.0 ""
                  det2 u1.0 nonthreadsafe default /opt/scali/lib/libdetdapl.so 1.0 ""
                  det3 u1.0 nonthreadsafe default /opt/scali/lib/libdetdapl.so 1.0 ""

                  Should we instead open up for wildcards in the static registry ? For
                  example :

                  det* u1.0 nonthreadsafe default /opt/scali/lib/libdetdapl.so 1.0 ""

                  Regards,
                    --
                    Steffen Persvold   |       Scali AS     
                  mailto:sp@... |  http://www.scali.com
                  Tel: (+47) 2262 8950 |   Olaf Helsets vei 6
                  Fax: (+47) 2262 8951 |   N0621 Oslo, NORWAY

                   

                • Kanevsky, Arkady
                  If 8.4.6 is confusing may be in the next version of the spec let remove it. This will keep only a single Static registry format! Arkady Arkady Kanevsky
                  Message 8 of 10 , May 9, 2003
                  View Source
                  • 0 Attachment
                    If 8.4.6 is confusing may be in the next version of the spec
                    let remove it. This will keep only a single Static registry format!
                    Arkady

                    Arkady Kanevsky email: arkady@...
                    Network Appliance phone: 781-768-5395
                    375 Totten Pond Rd. Fax: 781-895-1195
                    3rd Floor http: www.netapp.com
                    Waltham, MA 02451-2010 general phone: 781-768-5300

                    > -----Original Message-----
                    > From: Caitlin Bestler [mailto:cait@...]
                    > Sent: Fri, May 09, 2003 11:25 AM
                    > To: dat-discussions@yahoogroups.com
                    > Subject: RE: [dat-discussions] static registry format
                    >
                    >
                    > On 5/9/03, Kanevsky, Arkady wrote:
                    >
                    > >Sherman,
                    > >8.4.5 defined the static registry format for all Unixes
                    > >and Windows. This includes all Unixes: linux, solaris,
                    > >aix, bsd, hp-ux and others.
                    > >8.4.6 is for embedded and other non-unix and windows
                    > >platforms.
                    > >
                    > In fact, I'd venture that 8.4.5 is appropriate for *any*
                    > OS that supports standard IO and has a local file system
                    > suitable for holding configuration files. Given that you
                    > can fit a fairly large "/etc" directory in a NVRAM part
                    > my guess is that only the most specialized embedded
                    > platforms would ever need to consider an alternate format.
                    >
                    > Caitlin Bestler
                    > http://asomi.com/CaitlinBestler/
                    >
                    > ------------------------ Yahoo! Groups Sponsor
                    > ---------------------~-->
                    > Rent DVDs from home.
                    > Over 14,500 titles. Free Shipping
                    > & No Late Fees. Try Netflix for FREE!
                    > http://us.click.yahoo.com/BVVfoB/hP.FAA/uetFAA/W6uqlB/TM
                    > --------------------------------------------------------------
                    > -------~->
                    >
                    > To unsubscribe from this group, send an email to:
                    > dat-discussions-unsubscribe@yahoogroups.com
                    >
                    >
                    >
                    > Your use of Yahoo! Groups is subject to
                    > http://docs.yahoo.com/info/terms/
                    >
                    >
                    >
                  • Caitlin Bestler
                    ... /opt/scali/lib/libdetdapl.so 1.0 ... /opt/scali/lib/libdetdapl.so 1.0 ... /opt/scali/lib/libdetdapl.so 1.0 ... /opt/scali/lib/libdetdapl.so 1.0
                    Message 9 of 10 , May 9, 2003
                    View Source
                    • 0 Attachment
                      On 5/9/03, Steffen Persvold wrote:


                      >
                      >det0 u1.0 nonthreadsafe default
                      /opt/scali/lib/libdetdapl.so 1.0 ""
                      >det1 u1.0 nonthreadsafe default
                      /opt/scali/lib/libdetdapl.so 1.0 ""
                      >det2 u1.0 nonthreadsafe default
                      /opt/scali/lib/libdetdapl.so 1.0 ""
                      >det3 u1.0 nonthreadsafe default
                      /opt/scali/lib/libdetdapl.so 1.0 ""
                      >
                      >Should we instead open up for wildcards in the static
                      >registry ? For example :
                      >
                      >det* u1.0 nonthreadsafe default
                      /opt/scali/lib/libdetdapl.so 1.0 ""


                      That option was kicked around last year when we had a
                      subgroup focusing on Naming and other supporting services.

                      We decided against it, mostly on the presumption that
                      some form of differentiation would be needed anyway.

                      For example, each of those "det"s needs to have an
                      entry in the routing tables, don't they?

                      If you provide no differentiating data, how is any
                      application supposed to determine which one to open?

                      Even when assigning ports to IA Names dydnamically you do
                      necessarily have to update dat.conf with each reboot. You
                      can decide that there are up to 8 IAs on this physical
                      device, and then dynamically assign the ports to IAs
                      after each reboot. The dat.conf file can remain the same.

                      There are very few environments where the number and purpose
                      each RDMA capable interface is truly dynamic. Unless you're
                      at the core of the network, the topology that a host is
                      immediately connected to tends to be fairly stable. When
                      that is not true, network/device/fabric management is
                      likely to be automated. How "messy" dat.conf gets needs
                      to be measured by whether it is "messy" to management
                      software as opposed to a human reader. Boringly repetitive
                      entries never bother management software.


                      Caitlin Bestler
                      http://asomi.com/CaitlinBestler/
                    • Kanevsky, Arkady
                      Do not forget we would still need to define what IA_Name will be generated and how many for dat_registry_list_providers from the * . ARkady Arkady Kanevsky
                      Message 10 of 10 , May 9, 2003
                      View Source
                      • 0 Attachment
                        Do not forget we would still need to define what IA_Name will be generated and how many
                        for dat_registry_list_providers from the "*".
                        ARkady
                         

                        Arkady Kanevsky          email: arkady@...
                        Network Appliance        phone: 781-768-5395
                        375 Totten Pond Rd.      Fax:   781-895-1195
                        3rd Floor                http: www.netapp.com
                        Waltham, MA 02451-2010   general phone: 781-768-5300

                        -----Original Message-----
                        From: Walker, Ken [mailto:kwalker@...]
                        Sent: Fri, May 09, 2003 11:28 AM
                        To: 'dat-discussions@yahoogroups.com'
                        Subject: RE: [dat-discussions] static registry format

                        In the DAT 1.1 specification, look at the dat_provider_init() API (section 8.2.4.1).  To be loaded by a 1.1 Registry, a Provider must support this call.  Notice that part of the behavior you are to implement is to register with the dynamic registry only the IA Name passed to you in dat_provider_init().

                         

                        As a side effect, Providers and the Registry must be somewhat matched up—both must be DAT 1.0 or both must be DAT 1.1-or-higher.

                         

                        Regarding wildcards: If by "messy" below you mean the static registry will contain many entries, this is true but desirable.  The point to having separate entries for each IA is to give the administrator the ability to override a single entry at any time.  Wildcarding would give us a shorter static registry file, but complicate the specification of individual overrides.

                         

                        Ken

                         

                         

                        -----Original Message-----
                        From: Steffen Persvold [mailto:sp@...]
                        Sent: Friday, May 09, 2003 2:17 AM
                        To: dat-discussions@yahoogroups.com
                        Subject: Re: [dat-discussions] static registry format

                         

                        I have also a couple of questions regarding the static registry format.
                        The spec states that the each entry starts with the IA name, but when that
                        provider is loaded how can you be sure that it registers itself to the
                        dynamic registry with exactly that name ?

                        As an example, Scali has made a DAT 1.0 compliant kernel bypass module for
                        ethernet. This provider may have mulitple IAs corresponding to the actual
                        ethernet interfaces in the machine (for example det0 => eth0, det1 =>
                        eth1 and det2 => eth0+eth1) and creation of such devices is not in control
                        of the static registry file. Therefore with the current format, one would
                        have to have an entry in the dat.conf file for each ethernet device and I
                        think this could get messy :

                        det0 u1.0 nonthreadsafe default /opt/scali/lib/libdetdapl.so 1.0 ""
                        det1 u1.0 nonthreadsafe default /opt/scali/lib/libdetdapl.so 1.0 ""
                        det2 u1.0 nonthreadsafe default /opt/scali/lib/libdetdapl.so 1.0 ""
                        det3 u1.0 nonthreadsafe default /opt/scali/lib/libdetdapl.so 1.0 ""

                        Should we instead open up for wildcards in the static registry ? For
                        example :

                        det* u1.0 nonthreadsafe default /opt/scali/lib/libdetdapl.so 1.0 ""

                        Regards,
                          --
                          Steffen Persvold   |       Scali AS     
                        mailto:sp@... |  http://www.scali.com
                        Tel: (+47) 2262 8950 |   Olaf Helsets vei 6
                        Fax: (+47) 2262 8951 |   N0621 Oslo, NORWAY

                         




                        To unsubscribe from this group, send an email to:
                        dat-discussions-unsubscribe@yahoogroups.com



                        Your use of Yahoo! Groups is subject to the Yahoo! Terms of Service.
                      Your message has been successfully submitted and would be delivered to recipients shortly.