[arin-ppml] What is a "host"?

Owen DeLong owen at delong.com
Thu Sep 9 14:37:38 EDT 2010


On Sep 9, 2010, at 10:50 AM, Loki Jorgenson wrote:

> Pardon for butting in mid-conversation - but this particular bit of semantics has aggravated us before and so it is topical.
> 
> "Host" tends to be rather over-loaded semantically and in general has connotation that ties it to a physical or virtual machine.  It will likely be problematic if you don't get your key terms defined, particularly as it appears in policy and will need to be interpreted.  "Host" doesn't map well to IP addresses except in the simplest case.  And in an increasingly sensor network, mobile, and cloud-based world, this won't fit well.
> 
Host maps particularly well in this context, actually. The term host, as I would intend it in this policy
would be to count each item which requires a link-local IPv6 address other than loopback interfaces
as a host.

Thus, 100 IP addresses on a single interface would be one host.
100 interfaces on a single machine would be 100 hosts.
The interfaces of 100 virtuals running on a single machine would be 100 hosts.
A sensor that has an IPv6 address would be a host.

etc.

> Alternately, I could suggest something like "node" (or "network node") and subsequently "node IP end-point" or "IP node" or "IP end-point".  Here "end" relates to a point-to-point IP connection, not an end-to-end network path.  Sensors (and anything else with an address) then are simply IP nodes (or comprise at least one IP node).  A host then may have a plurality of IP nodes (either physical or virtual).
> 
I would argue that the terms host and node are synonymous, at least for purposes of this discussion,
and I don't see a meaningful difference in the terminology.

Owen




More information about the ARIN-PPML mailing list