SORT BY:

LIST ORDER
THREAD
AUTHOR
SUBJECT


SEARCH

IPS HOME


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

    iSNS, SLP, and FCIP



    The interim meeting notes indicate that we need to
    make progress on this.  Let me start by injecting
    a couple of things that either weren't discussed at
    or may not have been clear from the interim meeting:
    
    (1) SLP scopes are allowed to overlap in a fashion
    similar to Fibre Channel zones -- Service, User,
    and Directory Agents can each be in multiple scopes.
    
    (2) RFC 2610 specifies DHCP options for configuring
    SLP Directory Agent locations and SLP scopes for both
    User and Service Agents.  This would allow DHCP to be
    used for centralized administration of an SLP
    infrastructure, and works even when DHCP is not used
    for IP address assignment.
    
    One caveat about the analogy in (1) above - Fibre Channel
    zones are above the level of FCIP and completely transparent
    to it. SLP and iSNS for FCIP would be used only for tunnel
    setup - telling each FCIP device about peers to whom it
    should establish TCP connections for FCIP.  Also, both of
    these have some impact on iSCSI use of iSNS - I'll pick up
    that topic in a separate message.
    
    Going back over the meeting notes on this subject, I see
    the following topics:
    
    (1) Change notification (RFC 3082)
    (2) Scope vs. discovery domains
    (3) Periodic re-registration timing values
    (4) SLP only supports strings
    (5) Centralized administration
    
    Taking these up in order ...
    
    -- (1) Change notification
    
    There are two cases here, initial discovery and loss of
    connectivity.
    
    For initial discovery, FCIP may not need change notification
    *because* it's a peer to peer protocol (as opposed to client-
    server), so it doesn't matter who sets up the initial TCP
    connection.  For the situation in which some FCIP devices
    are up and running and a new one comes up, SLP-based discovery
    at the new device is sufficient to set up everything.
    
    The FCIP draft doesn't appear to contain any words about how
    to deal with concurrent setup - i.e., both peers initiate setup
    at the same time, resulting in two parallel TCP connections that
    aren't associated with a single FC link.  Some text on this topic
    should be added to the FCIP draft.
    
    For loss of connectivity, either FCIP or something running over
    it (e.g., FSPF) is going to heartbeat each connection, and so
    no external mechanism seems to be necessary.
    
    The one situation that doesn't seem to be covered here is a
    partial loss of connectivity (FCIP device loses touch with one
    of its peers, but not all of them).  These scenarios tend to be
    messy, and I'm not sure how iSNS compares to SLP in dealing
    with them (loss of connectivity to an iSNS server seems to
    have more severe impacts than loss of connectivity to an SLP
    DA).
    
    -- (2) Scope vs. discovery domains
    
    For FCIP, these appear to be essentially the same concept since
    SLP scopes can overlap in the same fashion as iSNS discovery domains.
    iSNS is somewhat better at information hiding - with SLP, each agent
    knows what scope(s) it's in, whereas with iSNS, that information is
    private to the nameserver (similar to Fibre Channel).  This may make
    a big difference if static configuration and more than a handful
    of scopes are involved, but appears to be less of an issue if
    DHCP is used.
    
    -- (3) Periodic re-registration
    
    SLP registrations with a Directory Agent carry a lifetime
    that defines both how long the DA can cache the registration
    and the period within which the registration must be
    refreshed.  This results in additional traffic to do the
    refreshes.  Lifetimes are measured in seconds, and RFC 2608
    recommends a default of 5 minutes before an idle TCP connection
    is closed.  This suggests that a few minutes for a re-registration
    period may be reasonable, and doesn't sound like a cause for
    a lot of traffic for moderate-sized configurations.  iSNS
    usage should generate less traffic, regardless.
    
    -- (4) SLP only supports strings
    
    As does iSCSI.  OTOH, turning an X.509 cert into a string is
    not exactly pleasant, but OTOH FCIP doesn't seem to have a
    requirement for distributing non-string parameters.  This
    may change as FCIP security is more tightly specified.
    
    -- (5) Centralized administration
    
    RFC 2610 provides a way for DHCP to be used to centrally
    administer SLP (and do so in a fashion where the only
    multicast-like things are DHCP's link-level broadcasts).
    On the assumption that FCIP connectivity among N devices
    is considerably simpler in structure than typical FC zoning
    among N ports, this seems to scale up to reasonable-sized
    configurations (several dozen FCIP devices should not be a
    problem, I'm not sure about more than 100).
    
    This is all for further discussion and comment, although
    it does not appear to present a strong case for use of
    iSNS with FCIP.
    
    Thanks,
    --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 01:04:38 2001
6315 messages in chronological order