Looking at the draft more carefully (when I am more awake, sorry), I
think what you are asking is that more text be put in section 4.1, maybe
a subsection? That additional text would talk about placement on a
switch that is not directly attached to the client, but is not as far
upstream as the first hop router? That text would include mention of
the effect of spanning tree variation and LACP path changes.
Sorry I was dense,
Joel
On 5/15/2011 9:45 PM, [email protected] wrote:
Joel,
If LACP is running active-passive across links to two different switches, there
may be no physical network node common to the currently active and currently
passive paths for some traffic. If SAVI is set up poorly, a switch to the
currently passive link could remove validation checks or black-hole traffic
that was flowing on the previously active link. For example, binding an IP
address to a single port or ports on a single switch is insufficient in such an
environment (address validation binding needs to span switches).
The material in 5.2.4 on Multi-LAN Hosts looks like a starting point - wrt
SAVI, active-passive link aggregation across multiple switches can be a way to
get into that sort of trouble on a single LAN (!). Beyond that, I think the
notion that SAVI checks should be consistent across alternate paths for the
same traffic is applicable at other scopes, e.g., all egresses from a
multi-homed network to the ISP networks involved.
Thanks,
--David
-----Original Message-----
From: Joel M. Halpern [mailto:[email protected]]
Sent: Friday, May 13, 2011 11:28 AM
To: Black, David
Cc: [email protected]; [email protected]; [email protected];
[email protected];
[email protected]; [email protected]; [email protected]
Subject: Re: [savi] Gen-ART review of draft-ietf-savi-threat-scope-05
Thank you David.
I think I understand the second part of your comment (about
multi-instance hosts), and will have to look at the document more to see
how to address that.
I am not sure I understand what the question is regarding the LACP /
link group issue. Is this really different from the issues that arise
whenever the SAVI functionality is not on the leaf switch? Where in the
document do you think it would help to discuss this?
Thank you,
Joel
On 5/13/2011 1:03 AM, [email protected] wrote:
I am the assigned Gen-ART reviewer for this draft. For background on
Gen-ART, please see the FAQ at
<http://wiki.tools.ietf.org/area/gen/trac/wiki/GenArtfaq>.
Please resolve these comments along with any other Last Call comments
you may receive.
Document: draft-ietf-savi-threat-scope-05
Reviewer: David L. Black
Review Date: 12 May 2011
IETF LC End Date: 18 May 2011
Summary: This draft is on the right track, but has open issues, described in
the review.
This draft discusses the threats and deployment environment for IP source
address validation with particular attention to finer-grain validation that
could be used within a network to validate IP addresses closer to the sources
of network traffic than ingress to an ISP's network.
Major issues:
There is no discussion of link teaming or aggregation (e.g., via LACP); this
may affect source address validation functionality by requiring the same
validation checks on all aggregated ports. An important case to discuss
is where the aggregated host links are connected to ports on different switches
(e.g., in an active/passive configuration).
The discussion of multi-instance hosts in section 5.2.3 is incomplete
in several important aspects:
(1) Some of the software switch implementations are single instance switches
whose implementation is distributed across multiple physical servers. This
results in concerns similar to the link aggregation discussion above.
(2) Live migration of virtual machines among physical servers causes
relocation of MAC addresses across switch ports. A so-called "gratuitous ARP"
is often used to inform the network of the MAC address move; port-based
source address validation information needs to move in response to such ARPs.
(3) MAC address relocation is also used as a failure recovery technique; the
surviving hardware element (e.g., host in a cluster) takes over the MAC
addresses of the failed hardware; like the previous case, a "gratuitous ARP"
is a common means of informing the network that the MAC address has moved,
and source address validation information needs to move in response to it.
Minor issues:
There doesn't seem to be much discussion of dynamic network reconfiguration,
which may change traffic egress points. VRRP may be a useful example to
discuss beyond the typical routing protocol updates to forwarding tables.
Nits/editorial comments:
idnits 2.12.11 ran clean.
Thanks,
--David
----------------------------------------------------
David L. Black, Distinguished Engineer
EMC Corporation, 176 South St., Hopkinton, MA 01748
+1 (508) 293-7953 FAX: +1 (508) 293-7786
[email protected] Mobile: +1 (978) 394-7754
----------------------------------------------------
_______________________________________________
savi mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/savi
_______________________________________________
Gen-art mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/gen-art
