SORT BY:

LIST ORDER
THREAD
AUTHOR
SUBJECT


SEARCH

IPS HOME


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

    RE: iSCSI: More on naming and discovery



    David,
    
    > With my WG co-chair hat off:
    >
    > -- DNS
    >
    > I'm uncomfortable with passing hostnames inband
    > in the basic iSCSI protocol with the exception of
    > 3rd party commands.  As long as the iSCSI initiator
    > and target can talk to each other via IP addresses,
    > use of DNS or LDAP to get those addresses is reasonable
    > provided that we have the ability to make sure booting
    > works reliably, but I'm uncomfortable specifying the
    > protocol in a way that requires resolution of DNS
    > hostnames as part of connection setup.  The current
    > CONNECT discussion seems to be consistent with this.
    
    I would hope that even this connect scheme disappear and instead allow
    gateways to handle much of the effort without trying to re-engineer network
    routing.  This of course would depend more fully on the information obtained
    from a LDAP server.
    
    > IMHO, requiring BOOTP for a server
    > that gets all its storage from iSCSI is not going to
    > be accepted (lots of machines don't boot via BOOTP,
    > and won't anytime in the near future), and requiring
    > a DNS resolution to find the storage required to get
    > the server up to the first point at which it can be
    > worked on (e.g., single user mode in Unix) is an
    > invitation to trouble.
    
    BOOTP and DHCP should be viewed as virtually the same protocol.  DHCP is the
    modern version and is present on your system.  You are right about DNS
    possibly being a problem but at the same time, that is an option that IT
    managers can make.  Perhaps their DNS services are in-house and reliable and
    provide the secondary fail-over resources which improve on the reliability
    of other resources. Even MS has space for indicating three DNS server
    locations.  I would work on a scheme that did not rely on DNS, but not every
    case is it bad.  Perhaps I over-stated my objections to dissuade invention
    of a SCSI name server.  To scale a name server on a large scale requires a
    massive amount of ram as these services can not be placed on a hard drive
    for dynamic retrieval.
    
    > 3rd party commands are a different matter.  The
    > current naming structure of 3rd party commands that
    > requires the Initiator to understand how the Target
    > of the 3rd party command resolves names makes 3rd
    > party commands difficult to use and fragile.
    > Requiring DNS resolving functionality in the 3rd
    > party command target in return for robust global
    > names for the storage involved in the 3rd party
    > command seems like a reasonable tradeoff.
    
    There should be two clearly defined address spaces, IP and SCSI.  These
    spaces should not be mixed in anyway.  Also, the SCSI address, if it must
    transverse IP, should be done in a transparent fashion.  Such configuration
    would be in the realm of the provider and not something that the client
    should be allowed to deal with or influence.  This type of configuration
    should happen at the time of authentication where the SCSI Portal is
    signaled as routes are determined.  It would also be wise to lease the
    authority.
    
    > -- CONNECT and target naming
    >
    > One thing to keep in mind is that there are traffic
    > analysis tools and both current and forthcoming
    > implementations of QoS that understand TCP ports.
    > If every iSCSI target has its own <IP address, TCP
    > port>, these tools work and can differentiate the
    > iSCSI targets without any further enhancements.
    > If multiple iSCSI targets are behind a common
    > <IP address, TCP port> pair, these tools will
    > not be able to separate out traffic to the
    > targets for different analysis or QoS treatment
    > without iSCSI-specific enhancements, although
    > they can do so based on different source addresses.
    
    QOS is a system wide configuration and far beyond requirements for
    connecting a SCSI transport.  Again, LDAP and a signal upon authentication
    would be helpful.  Most of these details should be hidden from the
    transport.  If they can be handled outside of the connection, they should
    be.
    
    Doug
    
    
    > --David
    >
    > ---------------------------------------------------
    > David L. Black, Senior Technologist
    > EMC Corporation, 42 South St., Hopkinton, MA  01748
    > +1 (508) 435-1000 x75140     FAX: +1 (508) 497-8500
    > black_david@emc.com       Mobile: +1 (978) 394-7754
    > ---------------------------------------------------
    >
    
    


Home

Last updated: Tue Sep 04 01:06:48 2001
6315 messages in chronological order