Thanks for your feedback. Answers inline: -----
Section 2: > MIMO can occur with refraction or partial reflection; it does not
need total internal reflection - Singlecore fiber transmission is essentially
equivalent to SISO. Multicore is equivalent to MIMO. Both are based on TIR.
The reference to TIR is wrt to spartial frequency and not temporal frequency. >
MIMO requires multiple transmitters, whereas multipath routing can be leveraged
using only one network interface - If a multicore optic fiber has multiple
cores similar to MIMO then the network interface also must have multiple
sub-interface or multiple addresses for multipathing. Taking this point
further, in fact Multipath TCP might require multiple addresses on an
interface. In fact this multipath TCP connections must be formed through the
virtual address space with anycast ipv6 addressing. then the RR cluster to
cluster communication is possible. IEEE documentation online show that
research is going on over the suitability of multicore for MIMO.
Section 3: > “machine” and “host” imply more similarity then
difference- Hence the reference to a step back. Host to host communication is
end-to-end and happens at the Transport layer. Then taking a step back and
observing that the machine to machine routing communication happens at the
Internet layer then the node to node communication can also include cluster to
cluster communication in the middle.A cluster being a cluster of nodes as
explained in the MIMO routing principle. > the Von Neumann bottleneck
would be an artifact of a machine implementation; non-Von architectures, such
as dataflow, do not suffer this fate (and are used in high-performance
routers)- Von neumann architecture is simple and based on data and program
simultaneously accessing a common memory space. The harvard architecture has
separate address space for data and program with common access for memory
(cache) (modified harvard).If the routing control plane can be considered as a
centralized control plane then the routing control plane must have a common
memory space. This concept can be stretched and we can consider the whole
routing plane of the internet as a centralized control plane.If a centralized
control plane is needed for the Internet then a common memory address space
needs to be facilitated. Hence a Virtual address space is the only option.
While the High performance routers may individually not be impacted at the
Internet layer, the need for cluster to cluster communcation and a common and
more advanced control plane between all the high performance routers cannot be
ignored. The benefit is not exactly higher data flow but better control of the
data flow through multipath.Anyway, primarily the reference to Van neumann
bottleneck is wrt finding where the data is stuck or congested in the Internet.
And the data can be stuck in transit between high performance routers.
Section 4 (which isn’t even written out as paragraphs): - Were divided
in sections. I have mentioned additional points below.
> OSPF is incremental computation of a Ford-Fulkerson graph based on local
information; whether code uses a heap, a stack, or any other data structure is
an implementation detail- I was referring to both EIGRP and OSPF. Specifically
in EIGRP a local computation does not resolve any global routing challenges.
Hence the feasibility condition.However this is applicable to OSPF as well. The
Ford-Fulkerson graph is the max-flow-min-cut algorithm wherein the max flow is
through the min-cut path. While an alternate path is called an augmenting
path. This is again mathematically anologous to path connectedness.
> classful routing hasn’t been used in IPv4 since the early 1990s and CIDR;
>it was never part of IPv6- Implication is that classful boundaries will exist
>due to classful summarization in routing protcols. While there may be no
>impact due to high performance routers, there will be traffic impact due to
>congestion and lack of control on summarization.This appears to be more of a
>concern without a solution in IPv6.
> BGP runs OSPF on a graph transformation, where the networks of Autonomous
>Systems collapse to nodes in the transformed graph; nodes are not “inserted”,
>but computed- Correct. BGP cannot insert routes.. I need to change this and
>put the "insertion" bit as one of the possibilities through the virtual
>address space and virtual redistriibution.Another factor is that in OSPF a
>virtual link does not allow data transfer. Analogously virtual links through
>the network of autonomous systems can be interlinked to form a virtual
>address space and a central control plane. In turn, the virtual AS boundaries
>of the central control planes can be determined by the best effort
>reachability as calculated through k-Nearness algorthms.
> a tree is a UAG where each node can have only one parent; while all trees
> are UAGs, not all UAGs are trees > A UAG whose underlying graph is a tree is
> — a tree. > BGP computes a graph determined by AS interconnectivity; whether
> this graph is strongly connected or not depends on the AS connectivity alone
This is mostly the duality that I have been referring. your comments are
opposite to my comments.There is a duality between program control and state
machine.This is analogous to the duality between an SDN controller and BGP PBB
EVPN. this is derived by applying the Von neumann bottleneck to the TCP IP
layers. The Ford fulkerson algorithm is based on max-flow min cut theorem which
is in turn based on a duality theorem.
The IBGP mesh within an AS is strongly connected. If the interconnected AS is
not strongly connected an attempt can be made to make it strongly connected by
utilizing the IBGP full mesh within an AS.If the graph is restricted to being a
UAG. An attempt must be made to make it a DAG for increased utilization of path
vector attributes.Hence the polytree structure (the virtual address space on
top interconnecting autonomous systems)
While the virtual address space may appear to be a best effort traffic
optimization solution it may be critical in solving future congestion and
traffic bottleneck problems due to mathematical limitations.
On Wednesday, 10 October 2018, 8:56:24 am GMT+5:30, Joe Touch
<[email protected]> wrote:
On Oct 8, 2018, at 7:23 AM, vineet deshpande <[email protected]> wrote:
I have re-written the document removing the buzz words
The buzzwords are still there.
They are exceeded only by the amount and extent of factual errors - a *partial*
list of which is noted below.
Joe
-----
Section 2: - MIMO can occur with refraction or partial reflection; it does not
need total internal reflection - MIMO requires multiple transmitters, whereas
multipath routing can be leveraged using only one network interface
Section 3: - “machine” and “host” imply more similarity then difference - the
Von Neumann bottleneck would be an artifact of a machine implementation;
non-Von architectures, such as dataflow, do not suffer this fate (and are used
in high-performance routers)
Section 4 (which isn’t even written out as paragraphs): - OSPF is incremental
computation of a Ford-Fulkerson graph based on local information; whether code
uses a heap, a stack, or any other data structure is an implementation detail -
classful routing hasn’t been used in IPv4 since the early 1990s and CIDR; it
was never part of IPv6 - BGP runs OSPF on a graph transformation, where the
networks of Autonomous Systems collapse to nodes in the transformed graph;
nodes are not “inserted”, but computed - a tree is a UAG where each node can
have only one parent; while all trees are UAGs, not all UAGs are trees - A UAG
whose underlying graph is a tree is — a tree. - BGP computes a graph determined
by AS interconnectivity; whether this graph is strongly connected or not
depends on the AS connectivity alone
---- _______________________________________________
Int-area mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/int-area
