Changes between Version 8 and Version 9 of OpenFlow/FOAM


Ignore:
Timestamp:
01/04/12 17:04:54 (12 years ago)
Author:
Josh Smift
Comment:

--

Legend:

Unmodified
Added
Removed
Modified
  • OpenFlow/FOAM

    v8 v9  
    4848
    4949The following sections are mostly of interest to FOAM admins.
     50
     51== Advisories ==
     52
     53Here are some things that FOAM admins should be aware of and watch out for.
     54
     55=== Rspecs that include both cross-connect ports ===
     56
     57In the current standard [wiki:OpenFlow/CampusTopology campus topology], a local !OpenFlow-controlled VLAN (often VLAN 1750) is cross-connected to multiple inter-campus VLANs, and a given sliver typically only includes one cross-connect port. FOAM admins should be careful not to approve rspecs that include both cross-connect ports (either explicitly or implicitly, e.g. by requesting the entire local DPID for VLAN 1750).
     58
     59Some experimenters may have good reasons for wanting to use both cross-connects; if they do, make sure they understand that this is a risky topology before you approve their sliver.
     60
     61=== Rspecs that include match rules with no protocol specified ===
     62
     63Rspecs that include match rules at one layer typically need to include match rules specifying the protocol at a lower layer. For example:
     64
     65 * An rspec that specifies a layer 3 match, like nw_src=A.B.C.D/N, also needs to specify dl_type to indicate what the layer 3 protocol is (e.g. dl_type=0x800 for IP and/or 0x806 for ARP).
     66 * An rspec that specifies a layer 4 match, like tp_port=N, also needs to specify nw_proto to indicate what the layer 4 protocol is (e.g. nw_proto=1 for ICMP, or 6 for TCP, or 17 for UDP).
     67
     68The important thing is that if the protocol ''isn't'' specified, some switches will end up with flowtable entries that match more traffic than they should, interfering with other experiments at your site. FOAM admins should be careful not to approve rspecs that include a match at one layer without specifying the protocol.
     69
     70We don't currently think that there's a valid use case for an experimenter wanting to specify a match at one layer without specifying the protocol at a lower layer, but will amend this advisory if we encounter one.
    5071
    5172== Switching from Expedient to FOAM ==