Musings

The following is solely authored by Jonathan Ferguson, and is henceforth referred to as Congruent Networks℠


Congruent Networks

All that is old is new again
Segment Routing, in particular SRv6, owes its existence to 'Source Demand Routing' (RFC 1940), May 1996; and 'Explicit Routing Protocol (ERP) for IPv6' (draft-ietf-sdr-erp-00), October 1994. Segment Routing like its ancestor makes the same mistake by not including End Systems and therefore preventing end user self-determination. - For this omission, a workaround will be possible when Segment Routing is incorporated into Quagga and everyone installs Quagga on their Computers.

Border Gateway Protocol
There needs to be a standards compliant scheme to remove duplicate consecutive ASNs in the AS_SEQUENCE attribute, to unravel the AS(s) prepended by administrative domains whom elected to manipulate the AS path. - Conversely there is not a need to remove duplicate consecutive ASNs in the AS_CONFED_SEQUENCE because that is within self-control.
More generally there has to be a way to un-mangle anything that a neighbour's neighbouring customer, peer or transit Autonomous System has done.

Commercialisation
Wavelength Division Multiplexing provides the benefit of being able to increase the utilisation of an expensive duplex-pair of optical fibres by way of a Multiplexing Splitter(Planar Lightwave Circuit), and a Demultiplexing Splitter [another PLC] and Wavelength Filters. - Its advantage is also its disadvantage which is fate-sharing.

  • By combining WDM with TRansparent Interconnection of Lots of Links the fate-sharing can be disseminated by the wavelengths reported by the pluggable modules installed into the Bridge, as duplicate wavelengths cannot be connected to the same WDM Splitter. At the other end of the link TRILL will be able to determine which interfaces share fate by the RBridge Nickname contained in Protocol Data Units. The gain is in the knowledge of confirming that the interfaces of the TRILL Switches are patched to the intended Multiplexer and Demultiplexer.

The WDM RBridge needs to be available to the mass-market coinciding with the SFP56 reaching economic critical-mass, to take advantage of PAM-4(Pulse-Amplitude Modulation 22) which will enable operating in a coherent spectrum with a denser channel spacing than legacy CWDM. These WDM RBridges could be implemented with such simplicity that it would facilitate the ability of the near-end and far-end WDM RBridges being owned and operated by different entities. - The WDM RBridge could become the customary means by which entities interconnect at various scales of SFP56 and SFP112 occupancy, by the ability of an arbitrary topology and a variable number of multiplexing and filtered-demultiplexing Splitter pairs. The economics of this new paradigm will enable the correct deployment of Wavelength Multiplexing/Demultiplexing, which is by the population of all wavelengths from the beginning, with each wavelength being measured separately at the ingress fibre point to the demultiplexing Splitter, and the light levels being balanced by passive attenuators at each transmitter. - With the future availability of less lossy wavelength filters and variable transmit power Pluggables, the WDM RBridges would be able to implement closed-loop feedback active transmitter attenuation, obviating the aforementioned tedious manual process.
There is nothing precluding WDM RBridges from doing Simplex CWDM, which is where two Splitters are connected back-to-back with a single fibre without filters and these are known as Power Splitters. The Pluggables installed have different transmit and receive wavelengths, where the neighbouring RBridges have the complementary inverse wavelengths of its neighbour. - This is possible due to the Pluggables having an integral receive wavelength filter. The reason for Simplex CWDM being uncommon is that the requisite transmit attenuation magnitude has to be compromised as it also affects the received power for these transceivers, and this process influences the desired transmitter attenuation at the other end and its receive power.
The next technological step for WDM RBridges is to use Pluggables which have a selectable transmit wavelength and etalon receive filter, enabling the use of back-to-back Power Splitters in either Simplex or Duplex configuration. - This is an established and cost effective technology used in Optical Transport Hierarchy/Optical Transport Network, where it is known as 'Colourless, Directionless and Contentionless'.

Misconceptions
Referring to an Ethernet as being multiplexed due to the use of non-locally-significant VLAN tags. Multiplexing and demultiplexing does exist in non-DIX Ethernet by way of the 'Destination Service Access Point' and 'Source Service Access Point'.
Treating an Ethernet as if it were a Time Division Multiplexer because of a Bridge having Z interfaces with speed N, and N interfaces with speed Z. The success of Ethernet is evidential by it outlasting Plesiochronous Digital Hierarchy and Synchronous Digital Hierarchy [despite attempts to assimilate Ethernet 1, 2] which have subrates and tributaries.
Explaining IP multicast communication as being one-to-many by means of replication [implying that there can only be one transmitter and any number of receivers], when in fact the benefits are much greater, as it is many-to-many with any host able to be a source to the group address at any time.
Specifying that a network has a backbone, bearer, core, edge or trunk, while the technologies in use do not impose a hierarchy. - This nomenclature should not be promulgated as it is a personal construct to abstract a portion of the network. This is an easy method to separate the wheat from the charlatans.
Referring to a Computer as something other than an End System, such as a Client or Server. End Systems are not architecturally limited to any role.
Using Open Systems Interconnection, especially X.200 the seven layer reference model and peer protocols, as a teaching or troubleshooting aid for technologies which were not devised in or with the organisation. - This has been a long standing issue, 'A PERSPECTIVE ON THE ARPANET REFERENCE MODEL' (RFC 871), September 1982.
A Datacentre may not be what you think it presently is as it is a continually redefined term. Originally Datacentres processed programmes on Punched Cards, then progressed to providing access to Time-sharing systems. The next iteration was the provisioning of storage mediums and the replacing of storage media, then came the installation of your own equipment at another party's premises which is known as co-location. The present definition of a Datacentre is equivalent to an Internet eXchange Point, there are legacy Datacentres which are foremost an Internet eXchange Point and then predominately the host of End Systems in a co-locating fashion.
Describing Broadband as analogous to speed is correct when referring to the modulation bandwidth of a carrier signal but is incorrect when relating to Computer Networks. - Broadband is a synonym of the Primary Rate Interface and is known as an E1 and is 2Mbps, whereas its complement Narrowband is the Basic Rate Interface and is 64kbps.

Multi-Protocol Label Switching
The motivation for MPLS was as an alternative to IP longest prefix matching for Routers, with influence from the deficiencies of Asynchronous Transfer Mode and to 'show up' The ATM Forum. Had ATM of been targeted as the Connection Oriented Network Service replacement for DCE-DCE(Data Circuit-terminating Equipment to Data Circuit-terminating Equipment) as in Frame relay, X.25 and X.75 network interfacings, ConnectionLess Network Protocol could have supplanted the Internet Protocol; but with only an NSAP address and not a full implementation of CLNP [this situation also befell Synchronous Digital Hierarchy], Cell Relay was doomed from the outset and this is despite the decision of cell size. The dead end was the reliance on a single Network-Network Interface(ITU-T: Network-Node Interface) which was SDH. Since circa 1996 label switching has become a means for an entity to provide private interconnection(backhaul in archaic lexicon) of another organisation's incongruent network. As of this writing MPLS does not benefit the source or destination host in the administrative domain that has deployed it.
For MPLS to be useful it must be transitive beyond an AS and the label-space needs to be globally unique and without fiscal encumbrance. [Historic recurrence: 'A Framework for Multiprotocol Label Switching' (draft-ietf-mpls-framework-00), May 12, 1997 – 4.1.1.3 Other Label Allocation Methods, paragraph Ⅱ]
To ensure the efficacy of MPLS the hosts would utilise source-based routing originating from the reliable flooding of a label representing the host itself and for each of its interfaces by a TLV from an equivalent of PNNI within the routing domain [hence a 'Forwarding Equivalence Class'], for congruence these labels, the MTU of all links for each End System, and the SPF graph, would be conveyed via the IDRP/BGP between Autonomous Systems.
It is from the padding of IS-IS Hellos that the maximum MTU can be determined per link for the eventual calculation of whether the route is feasible. When the frame is to be populated with a payload the lowest denominator MTU is used.
With my proposal there is not a need for a 'resolution protocol' as each link-label is an EUI-64, for which an IPv6 Global Unicast Address by Stateless Address Autoconfiguration can be resolved to a hardware address and vice versa. The label that the host is to use for itself is equal to the numerically-lowest of its link-labels. - This is the only instance where duplicate labels do not signify a loop in the graph; this also infers that every host along with its numerically-lowest link must appear in the graph.
To future-proof MPLS will require that adequate bit fields are reserved for versioning, capability advertising and further extensibility, prior to any scale of deployment.

  • Datagram service would pop each link-label on ingress and pop each host-label on egress — if the label becomes unknown the PDU is dropped.


    A reliable service would keep all labels intact and would indicate whether the label is traversing in the forward or reverse direction — if a succeeding labelled-link is down an indicator bit is inverted so that the label is then delivered to the source host — if during the reverse journey the label cannot return to the preceding label the PDU is dropped. [This theory is too simplistic as there is not a means for the source host to know at which label the PDU could not proceed] The label space would need to include an additional 1 bit field for each label to indicate at which label the PDU could not proceed, alternatively there could be a 10 [sufficiently sized to accommodate the diameter of the future Internet] bit field [the most significant bit is for the direction] akin to a hop limit that is decremented only in the forward direction per label and never incremented. - Both approaches can be the method of signifying that these labels are not a datagram service.

If an overlay network is desired at Open Systems Interconnection model pseudo-layer 2 ½ then we could go back to the future and deploy CLNP, with its benefits of ES-IS redirects and the need of a host to have only one address for all of its interfaces, by developing IPinCLNS.
MPLS remains intangible for many and despite 'Encapsulating MPLS in IP or Generic Routing Encapsulation (GRE)' (RFC 4023), March 2005 and 'Encapsulation of MPLS over Layer 2 Tunnelling Protocol Version 3' (RFC 4817), March 2007, IPinIP and IPinGRE have instead prevailed.
The eventuality of MPLS was that it was too late, was initially under-specified, was too simplistic of an encapsulation, and that it continues to be re-specified. 'Comparison of Proposals for Next Version of IP' (RFC 1454), May 1993
Regarding Overlay Networks, since the IANA exhaustion of IPv4 there has been a shift from connecting IPv6 islands with IPinIP and IPinGRE, to the connecting of Ethernets by way of L2TPv3 PWE3, and more recently VXLAN and Geneve, this growth is in part due to the larger host address space of IPv6 and will accelerate coinciding with the availability of TRILL RBridges.

Security Or Insecurity
You are compelled to due process in the jurisdiction that you are present and in the locale of every person, Intermediate System and End System you have an impact upon.
Before intercepting, altering, restricting or logging packets between End Systems, you must have consent from the user of the source address and the user of the destination address for every permutation of source to destination flow. When a user changes at either the source or destination host you must again be granted consent before intercepting, altering, restricting or logging their first packet.
You must not do anything unbeknownst to the user, nor anything that modifies or constrains the end-to-end nature of the Internet.

  • When source-based routing or Segment Routing is made available a new paradigm will be possible during path computation, by the creation of an equivalent of the Wireless Regulatory Database containing the legal requirements of each country and jurisdiction, a route can be deduced that takes into account bilateral accord with your locale and countries' laws that are conducive to your own.

Significance And Simplification
In a Computer Network, or a network comprising Computer Network technologies, there are three elemental roles: End System, Bridge, and Intermediate System.
Bridges connect to Bridges and in between of the End Systems and Intermediate Systems. End Systems do not directly connect to Intermediate Systems, nor do Intermediate Systems connect directly to Intermediate Systems, as each would defeat an advantage of using Ethernet as a medium.
The choices to be made for Bridges match the physical topology which corresponds with the choice to use Wavelength Division Multiplexing, or not, and if so which wavelength is to terminate at which location.
There is not a choice to be made for End Systems as they can be anywhere and be doing anything.
The choices to be made for Intermediate Systems are for whether the Intermediate System is a level 1 Intermediate System, a level 2 Intermediate System, or a level 1 and level 2 Intermediate System. Level 1 Intermediate Systems connect [via Bridges] to End Systems, whereas Level 2 Intermediate Systems can also run Border Gateway Protocol. To be frank, significant networks do not use Open Shortest Path First due to the interconnections of inter-area [non-stub] areas being more constrained compared to the versatility of the Level 2 Routeing Domain. - Intermediate System-Intermediate System has further benefits such as the ability to have up to 256 Network Service Access Point addresses [hence up to 256 areas, whether different or the same] simplifying the migration of Intermediate Systems between areas, and is aided by the conventional Level 2 adjacency formations when using a single Network Entity Title.
The choices to be made for Inter-Domain Routing concerns the logical hierarchy of Internal Border Gateway Protocol for whether the Level 2 Intermediate Systems are in any combinatorial gestalt of full mesh, Route Reflector, and Confederation.

Virtual Local Area Networks
It must be by volition of the user of an End System whether and which VLANs they participate in.
IEEE 802.1Q does not include a mechanism for an Ethernet Switch to signal to a host for permission to assign a VLAN to the port that the host is attached. In the case that the VLAN tag is transitive, subsequent Switches and hosts need to be permissive of the VLANs they receive by stripping all VLAN tags on ingress and then on egress being courteous to their LAN by transmitting all frames untagged.

  • VLANs are near the end of their useful life, the predominant usage is to facilitate an Ethernet interface to have more than one unicast IP address, in this situation it is one IPv4 address per broadcast domain. With IPv6 being the present Internet and IPv6 allowing for more than one unicast address per physical interface, a pseudo-private LAN can be accomplished by using the Unique Local Address fd00::/8 block and dividing it into ::/64.

When I transitioned from academia to industry I noticed differences and similarities between Computer Networks and Telecommunications Networks — some are obvious, many were not
Telecommunications Networks use Computer Networking technologies. The major holdout is Dense Wavelength Division Multiplexing requiring in-band management for amplification gain feedback and for regeneration passthrough.
Computer Networks have a single upfront expense [excluding the insourced or outsourced administrative expense and electricity], whereas Telecommunications Networks have a periodic expense [usually monthly] and may have an additional expense for setup.
Computer Networks are deliberately resilient because the users are not notified when a single interconnection is disconnected due to the use of TRansparent Interconnection of Lots of Links, or in antiquated networks the use of Spanning Tree Protocol and Link Aggregation Control Protocol.
Telecommunications Networks advise their customers [Change Management] of when a disconnection is planned, because the connectivity is being paid for and because the connectivity is most likely provided as a stub [spur in Carriers' vernacular] including the rare instances where LACP is used. Carriers are reliant upon the MTBF(Mean Time Between Failures) of their active equipment and the physical diversity of their passive equipment.
When Telecommunications Networks provide an Ethernet as a private interconnection, the carrier disables SNPA(SubNetwork Point of Attatchment) address learning, Spanning Tree Protocol, Link Layer Discovery Protocol and other protocols to make the Ethernet appear transparent to the customer, as no frames are present when the customer's equipment is not connected, to ensure that the customer's equipment is unable to communicate with the carrier's equipment.
Computer Networks do not hobble the speed of network interfaces, whereas Telecommunications Networks can limit the interface throughput to be less than the native speed to facilitate an additional revenue stream and because it is physically and economically infeasible to provide a fat-tree between their customer's locations. One method of an additional revenue stream is to install a piece of equipment [Network Termination device, NT1, in ISDN parlance] at the customer's premises to deliver different private interconnections from each interface, where one of could be connectivity to the Internet.
The Metro Ethernet Forum exists solely for Telecommunications Networks to be able to outsource to other carriers the provision of private interconnection of their customer's Computer Networks in locations that they are unable to economically serve, in this instance they outsource to another carrier whom they interconnect [Point of Interconnect] with and it is this interconnection method that the Metro Ethernet Forum standardises. It is to be noted that MEF themselves are not a standalone technology standards body, the MEF is merely a Consumer Advocate [for member carriers] to the standards bodies. In instances where there is a dominant Carrier in a location, the MEF will be redundant, as it is the preeminent party who will dictate how the private interconnection is to occur. Additionally, do not confuse the Metro Ethernet Forum with Metropolitan Ethernet implementations of ITU-T Automatic Protection Switching G.8032/Y.1344 Ethernet Ring Protection Switching.
In Computer Networks each user sets up their own tunnels by means such as IPsec or WireGuard, whereas Telecommunications Networks provide a method for the administrator of the Computer Network to setup tunnels across the carrier's network using the so called 'SD-WAN' to create such tunnels as pseudowires over L2TPv3 or MPLS. - If the administrator of the Computer Network is proficient, they can obviate SD-WAN by creating tunnels between their disjoint equipment by bridging the physical interfaces and logical tunnel interface of their RBridges by such means as L2TPv3 or VXLAN. The downside to any method of tunnelling is either of the requirement to reduce the Maximum Transmission Unit for it to be encapsulated into the native packet's payload, or for fragmentation/segmentation and reassembly to be performed along with reception reordering.
The formal education for qualifications of which I am laureled, did not anticipate that carriers would install equipment [active or passive] in customers' premises. The expectation was that carriers would meet at the boundary of the customer's premises and that the entirety of the premises cabling would be the sole domain of Registered Cablers. - This has a benefit that a fault can be isolated, by the property owner, the carrier, or an independent party, at the neutral point, to either the customer's property or the carrier's infrastructure. In Australia, the National Broadband Network was specified to be installed this way, in addition to this all fibres would be spliced in pits going direct to the exchange, without using pillars or roadside cabinets, where each customer would be patched to the Passive Optical Network Splitter and Optical Line Terminal of their chosen of carrier.


Jonathan Ferguson, Ubuntu Wiki, Musings, https://wiki.edubuntu.org/JonathanFerguson/Musings


JonathanFerguson/Musings (last edited 2024-03-27 10:11:31 by jonathan-ferguson)