Free Nokia NRS I Exam Flashcards

OSI Layers 7–5 (Application, Presentation, Session; PDU: Data/Message) map to the TCP/IP Application layer. Layer 4 (Transport; PDU: TCP Segment or UDP Datagram) maps to Transport. Layer 3 (Network; PDU: Packet) maps to Internet. Layers 2–1 (Data Link Frame and Physical Bit) map to the TCP/IP Network Interface (Link) layer.

Layer 2 inspects the 16-bit Ethernet II EtherType (0x0800 = IPv4, 0x86DD = IPv6, 0x0806 = ARP, 0x8100 = 802.1Q, 0x8847 = MPLS unicast). Layer 3 inspects the 8-bit IPv4 Protocol or IPv6 Next Header field (1 = ICMP, 6 = TCP, 17 = UDP, 58 = ICMPv6, 89 = OSPF). TCP/UDP use 16-bit ports (0–65535); delivery to a socket also depends on protocol and address/connection context, not destination port alone. IPv6 Next Header can identify an extension header before the transport header.

TCP (RFC 9293) is connection-oriented with a 20–60 byte header, a 3-way handshake (SYN, SYN-ACK, ACK), byte-stream sequence and cumulative ACK numbers, retransmission of lost segments, and sliding-window flow control (rwnd). UDP (RFC 768) is stateless and connectionless with a fixed 8-byte header (Source Port, Destination Port, Length, Checksum) and no handshake, flow control, or transport retransmission.

BGP uses TCP port 179. LDP uses UDP port 646 for multicast/unicast Hello discovery and TCP port 646 for session establishment and label exchange. SSH uses TCP 22, Telnet TCP 23, DNS UDP/TCP 53, DHCPv4 Server UDP 67 / Client UDP 68, TFTP initial request UDP 69 (the transfer uses negotiated UDP transfer IDs), and NTP UDP 123.

The control plane (on the CPM) runs OSPF, IS-IS, BGP, and LDP to build the RIB. The data plane (FP silicon on IOMs/IMMs/XMAs) forwards transit packets at line rate using the programmed FIB and label tables. The management plane handles CLI, SSH, SNMP, and NETCONF. In-band management traverses production data links (e.g., to the system IP), whereas out-of-band management uses the dedicated CPM management Ethernet port on a separate network.

The Nokia 7750 Service Router (SR) is a multiservice Provider Edge (PE), aggregation, and Broadband Network Gateway (BNG) platform supporting Layer 2/Layer 3 VPNs and IP/MPLS routing. The 7450 Ethernet Service Switch (ESS) focuses on high-density Carrier Ethernet aggregation and metro switching. The 7950 Extensible Routing System (XRS) is a high-capacity core (P) and peering router designed for multi-terabit backbone switching.

The CPM runs control and management software and can have active/standby redundancy. On modular 7750 SR designs, IOMs provide forwarding resources and accept MDAs for physical ports; IMMs integrate media/forwarding functions. The 7950 XRS uses XCM carriers with XMAs, rather than making XMA a universal 7750 component. Fabric components connect forwarding modules; their placement and redundancy vary by chassis. Identify the actual platform before applying a slot or module diagram.

Physical ports use slot/mda/port notation (e.g., 1/2/4 denotes chassis slot 1, MDA 2, port 4; breakout/connector ports append a connector or channel suffix such as 1/1/c1/1). The system interface is a persistent logical loopback interface configured with a /32 IPv4 (and /128 IPv6) address that is not tied to one physical link; remote reachability still requires a working path and advertised/installed routes, serving as the default router-id and transport endpoint for OSPF, IBGP, and LDP.

In the traditional SR OS boot model, bootstrap/boot-loader processing locates the Boot Option File, then the runtime image and startup configuration. Exact filenames and storage locations depend on platform and release; cf3: is a common location, not universal. The BOF specifies the out-of-band management port IP address and static routes, primary/secondary/tertiary runtime TiMOS software image locations, primary/secondary/tertiary configuration file paths, console speed, and, where supported, the persist (index persistence) setting before loading the SR OS runtime image and startup configuration.

In Classic CLI, commands entered in configuration mode modify the active running configuration immediately line-by-line (and 'info' displays config while 'show' displays operational state). In MD-CLI, edits are staged in a transactional candidate datastore derived from YANG models; you inspect pending diffs with 'compare', verify syntax/dependencies with 'validate', and apply all changes atomically to the running datastore with 'commit' (or abandon them with 'discard').

In MD-CLI, '(ex)' indicates Exclusive configuration mode (locking out other editors), '(pr)' indicates Private candidate mode, and '(ro)' indicates read-only configuration mode. In MD-CLI, a leading asterisk '*' means the candidate datastore contains uncommitted edits, and '!' warns that another user committed changes since your candidate baseline was opened. In Classic CLI, a leading '*' means the running configuration has unsaved changes since the last 'admin save'.

It depends on auto-config-save. Commit activates candidate changes in the running datastore. With auto-config-save enabled, a successful commit also saves the configuration; with it disabled, explicitly use admin save to persist it. Verify the BOF’s startup configuration location and save result. Activation and durable startup configuration are separate concepts; neither “commit never saves” nor “every commit always survives reboot” is correct.

For an untagged frame, it inspects the 2-byte field immediately following the 6-byte Destination MAC and 6-byte Source MAC addresses. If the value is >= 0x0600 (1536 decimal), the frame is Ethernet II and the field is an EtherType identifying the upper-layer protocol (e.g., 0x0800 IPv4). If the value is <= 1500 (0x05DC), the frame is IEEE 802.3 and the field indicates the payload byte length, followed by an IEEE 802.2 LLC (and optional SNAP) header.

An untagged Ethernet frame is 64 to 1518 bytes: 6-byte Destination MAC + 6-byte Source MAC + 2-byte EtherType/Length + 46 to 1500 bytes of payload (padded to the minimum payload length if necessary) + 4-byte FCS (CRC-32). Frames shorter than 64 bytes are discarded as runts. The 7-byte Preamble, 1-byte Start Frame Delimiter (SFD 0xD5, transmitted LSB-first as 10101011), and Inter-Packet Gap (IPG), normally 12 byte-times of idle time rather than a frame field, are excluded from the frame size.

For a commonly used globally assigned MA-L/EUI-48 address, the allocation contains a 24-bit Organizationally Unique Identifier (OUI, first 3 bytes) and a 24-bit vendor-assigned NIC identifier (last 3 bytes). In the first transmitted octet (Byte 0), Bit 0 (least significant bit) is the Individual/Group (I/G) bit: 0 = unicast, 1 = multicast/broadcast. Bit 1 is the Universal/Local (U/L) bit: 0 = universally administered, 1 = locally administered. Locally administered addresses do not have to preserve a manufacturer OUI layout.

On classic 10/100 Mb/s shared half-duplex Ethernet, simultaneous transmissions collide; CSMA/CD listens before sending, detects collisions within the 512-bit slot time (requiring at least a 64-byte frame), transmits a 32-bit jam signal, and waits a random truncated binary exponential backoff delay. Full-duplex point-to-point links provide independent transmit and receive paths with zero collision possibility, so CSMA/CD is completely disabled. Gigabit half-duplex Ethernet instead uses a 4096-bit slot with carrier extension; do not generalize the 512-bit timing to every half-duplex speed.

A Layer 1 hub places all connected ports into a single shared collision domain and a single broadcast domain. A Layer 2 switch isolates each port into its own collision domain (microsegmentation) but floods broadcasts across all ports in the same VLAN. VLANs partition a switch into separate logical broadcast domains, and Layer 3 router interfaces terminate Layer 2 frames to bound broadcast domains.

1) The receiving interface validates frame integrity via the 32-bit FCS (drop runts or CRC errors). 2) Learn the Source MAC address, ingress port, and VLAN in the FDB (and reset its aging timer, which defaults to 300 seconds for local MACs in SR OS VPLS). 3) Filter/drop the frame if the known unicast Destination MAC maps to the same ingress port. 4) Forward out the single mapped egress port for a known unicast MAC. 5) Subject to forwarding state, filters, flood controls and multicast snooping, flood to eligible other ports for unknown unicast, unregistered multicast or broadcast (ff:ff:ff:ff:ff:ff) frames.

Inserted after the Source MAC, the 4-byte 802.1Q tag contains a 2-byte Tag Protocol Identifier (TPID = 0x8100) and a 2-byte Tag Control Information (TCI) field comprising a 3-bit Priority Code Point (PCP), a 1-bit Drop Eligible Indicator (DEI), and a 12-bit VLAN Identifier (VID). Out of 4096 values (0–4095), VIDs 1–4094 are usable VLANs, VID 0 is for 802.1p priority-tagged frames, and VID 4095 is reserved. Adding one tag increases the standard maximum frame from 1518 to 1522 bytes.

QinQ prepends an outer 4-byte Service-VLAN (S-VLAN, TPID 0x88A8 or 0x8100) tag ahead of the customer's inner 4-byte Customer-VLAN (C-VLAN, TPID 0x8100) tag, adding 8 bytes total (1526-byte max frame). On Nokia SR OS access ports, 'encap-type' can be set to 'null' (port-based SAP demultiplexing, without using a VLAN tag as the service selector; configured admission rules still apply), 'dot1q' (demultiplexes services by a single 802.1Q tag, e.g., sap 1/1/1:100), or 'qinq' (demultiplexes by outer.inner tag pairs, e.g., sap 1/1/1:100.200).

Because Ethernet lacks a TTL field, Layer 2 loops cause broadcast storms, duplicate frame delivery, and MAC address flapping in the FDB. Spanning Tree elects a single Root Bridge with the lowest 8-byte Bridge ID (2-byte Priority [default 32768] + 6-byte MAC), selects one Root Port per non-root switch (lowest root path cost), elects one Designated Port per segment, and blocks remaining redundant ports. RSTP (802.1w) uses Discarding, Learning, and Forwarding states plus Alternate/Backup port roles for faster proposal/agreement convergence when its link and neighbor conditions are satisfied; sub-second recovery is not guaranteed.

A LAG bundles multiple physical Ethernet ports into a single logical interface to increase aggregate capacity and resiliency. LACP dynamically negotiates bundle membership (at least one side must be in 'active' mode; 'passive' responds only). To prevent out-of-order packets, ordinary per-flow hashing places a conversation on one member link, so a single flow cannot exceed one member link's speed. If active links fall below a configured 'min-links' threshold, the entire LAG transitions operationally down.

For end-to-end delivery without drops or fragmentation: Access Port Physical MTU >= Service MTU + SAP encapsulation overhead, and Network Port Physical MTU >= Service MTU + MPLS transport/service labels (4 bytes per label) + Layer 2 Ethernet header (14 bytes, plus 4 bytes per VLAN tag). For example, transporting a 1514-byte untagged customer Ethernet frame inside an Epipe/VPLS (default service-mtu 1514) over a two-label MPLS SDP requires at least 1514 + 8 (two labels) + 14 (Ethernet header) = 1536 bytes on the MPLS network port. These SR OS Ethernet MTU figures exclude FCS, preamble and inter-packet idle time. The 1536-byte case assumes exactly two labels, no network VLAN tag and no pseudowire control word; each extra label, VLAN tag or control word adds its actual overhead. IP MTU limits the IP packet, excluding the link-layer header. On a routed Ethernet interface, that packet plus the applicable Ethernet encapsulation must fit the port MTU. Check the smallest path MTU and service encapsulation settings.

RFC 1918 private IPv4 blocks: 10.0.0.0/8 (10.0.0.0–10.255.255.255), 172.16.0.0/12 (172.16.0.0–172.31.255.255), and 192.168.0.0/16 (192.168.0.0–192.168.255.255). Special-purpose blocks: 127.0.0.0/8 (host loopback), 169.254.0.0/16 (APIPA link-local), 224.0.0.0/4 (224.0.0.0–239.255.255.255 multicast), 0.0.0.0/8 ('this network'; distinct from the default-route prefix 0.0.0.0/0), and 255.255.255.255/32 (limited broadcast).

With prefix length /n, host bits h = 32 - n, block size = 2^h, and standard usable hosts = 2^h - 2 (excluding all-0s network and all-1s directed broadcast). A /27 (255.255.255.224, h=5) has block size 32 and 30 usable hosts (e.g., 192.168.1.32/27 has usable hosts .33–.62 and broadcast .63). A /30 (255.255.255.252, h=2) has 2 usable hosts. A /31 (RFC 3021) provides 2 usable addresses on point-to-point links with no directed broadcast, and a /32 (255.255.255.255) identifies a single host or system loopback address.

The exact summary is 10.24.4.0/22 (mask 255.255.252.0). In the third octet, 4 (00000100), 5 (00000101), 6 (00000110), and 7 (00000111) share the first 6 bits (000001xx), giving 16 + 6 = 22 network bits. For 2^k contiguous subnets of length /n to fold into a single exact /(n - k) summary without overlapping extra space, the lowest subnet's block number must be an exact multiple of 2^k.

An IPv6 address consists of eight 16-bit hexadecimal groups (hextets). Under RFC 5952: (1) remove leading zeros within each hextet (e.g., 00ab becomes ab, and 0000 becomes 0); (2) use a double colon '::' at most once to compress the single longest run of two or more consecutive all-zero hextets (if two runs tie in length, compress the first one; never use '::' for a single 0 hextet); and (3) write hexadecimal digits a–f in lowercase.

Key IPv6 prefixes: the currently allocated general global-unicast block 2000::/3, Unique Local fc00::/7 (commonly fd00::/8), Link-Local fe80::/10, Multicast ff00::/8, Loopback ::1/128, and Unspecified ::/128 (IPv6 has no broadcast address). To form a 64-bit Modified EUI-64 interface ID from a 48-bit MAC (e.g., 00:16:4d:11:22:33), split the MAC into two 24-bit halves, insert 'fffe' in the middle (0016:4dff:fe11:2233), and invert the U/L bit (bit 1 of the first octet: 0x00 -> 0x02), yielding 0216:4dff:fe11:2233.

The IPv4 header has a variable length (IHL field counting 4-byte words), a Header Checksum that every router must recompute after decrementing TTL, and inline fragmentation fields (Identification, DF/MF Flags, Fragment Offset). The IPv6 base header is fixed at 40 bytes, removes the header checksum (relying on L2 FCS and L4 checksums), eliminates transit-router fragmentation (only the source host fragments via an Extension Header after PMTUD), replaces TTL with Hop Limit, and replaces Protocol with Next Header.

Packet forwarding uses Longest-Prefix Match (LPM) among installed FIB prefixes; route preference is used earlier when selecting routes for the same prefix. Since 10.1.2.200 (.11001000) falls outside 10.1.2.0/25 (.0–.127), the longest matching prefix is 10.1.2.0/24 (via BGP). Assuming ordinary routing without NAT/tunneling and a packet eligible for forwarding over Ethernet, the router decrements the IPv4 TTL by 1, recalculates the IPv4 Header Checksum, and rewrites the Layer 2 Source MAC (to its egress port MAC) and Destination MAC (to the next-hop's MAC) while leaving Source and Destination IP addresses unchanged.

When forwarding to an on-link IPv4 address (or the default gateway IP for an off-subnet destination) whose MAC is not in the ARP cache, the sender transmits an Ethernet broadcast frame (Destination MAC ff:ff:ff:ff:ff:ff, EtherType 0x0806) asking who owns the target IP; the target replies with a unicast ARP Reply containing its MAC address. An RFC 5227 ARP Probe checks for duplicate IPv4 addresses before use by setting the Sender IP field to 0.0.0.0 so existing ARP caches are not polluted.

NDP uses five ICMPv6 (Protocol 58) messages: Router Solicitation (Type 133, sent to all-routers multicast ff02::2), Router Advertisement (Type 134, multicast to ff02::1 or unicast in response to a solicitation, with applicable prefix/SLAAC options), Neighbor Solicitation (Type 135, sent to the target's solicited-node multicast address ff02::1:ffXX:XXXX formed from the low-order 24 bits of its IPv6 address for address resolution and DAD; neighbor-unreachability probes may instead be unicast), Neighbor Advertisement (Type 136), and Redirect (Type 137).

Ping measures reachability and RTT using ICMP Echo Request (Type 8, Code 0) and Echo Reply (Type 0, Code 0). Traceroute sends probes with incrementing IPv4 TTL values (1, 2, 3...) so a router that expires a probe normally returns ICMP Time Exceeded in Transit (Type 11, Code 0). PMTUD sends packets with the Don't Fragment (DF) bit set; if a link's MTU is smaller than the packet, the router drops it and returns ICMP Destination Unreachable: Fragmentation Needed and DF Set (Type 3, Code 4). Filtering or ICMP rate limits can suppress replies; absence of a traceroute response does not itself prove a forwarding failure.

A host obtains its IPv4 address, subnet mask, default gateway, and DNS server via the four-step DHCPv4 DORA exchange (Discover, Offer, Request, Acknowledge over UDP ports 67/68). When the host sends traffic to an external IP, it addresses the Ethernet frame to its default gateway's MAC while keeping the remote Destination IP. At the edge router, NAPT rewrites private (Source IP, Source Port) tuples to an externally visible source address and transport-port mapping, tracking return flows in a state translation table; uniqueness is required within the applicable mapping context, not necessarily across every destination.

When multiple routing protocols offer valid paths for the exact same destination prefix, the RTM installs the route with the lowest numeric Route Preference into the RIB and programs resolved next hops into the data-plane FIB (or multiple equal-cost next hops when ECMP is enabled). Default SR OS preferences are: Direct/Local = 0, Static = 5, OSPF Internal = 10, IS-IS Level 1 Internal = 15, IS-IS Level 2 Internal = 18, OSPF External = 150, IS-IS L1 External = 160, IS-IS L2 External = 165, and BGP (EBGP and IBGP) = 170.

A floating static route acts as a backup path by setting its preference higher than the primary dynamic routing protocol (e.g., preference 15 or 180 so OSPF at 10 is preferred while active; in MD-CLI under '/configure router "Base" static-routes route <prefix> route-type unicast next-hop <ip> preference <val>'). A summary discard ('blackhole') route drops traffic matching unused sub-ranges of an advertised aggregate prefix, preventing routing loops between a summary route and a default route.

OSPF (IP protocol 89) uses 5 packet types: (1) Hello, (2) Database Description (DBD/DD), (3) Link-State Request (LSR), (4) Link-State Update (LSU), and (5) Link-State Acknowledgment (LSAck). OSPFv2 defines eight neighbor states: Down, Attempt (used on NBMA), Init, 2-Way, ExStart, Exchange, Loading and Full. Not every neighbor traverses or remains in every state. To form an adjacency, neighbors must match on Area ID, Hello and Dead intervals, authentication type/key, stub area flags, and primary subnet mask (on broadcast links); mismatched interface MTUs can stall neighbors in ExStart/Exchange.

On broadcast multi-access links, OSPF conducts a non-preemptive election for Designated Router (DR) and Backup DR (BDR) based on highest interface priority (1–255; priority 0 = never eligible) and then highest Router ID; DROther routers remain in 2-Way state with each other. LSA types: Type 1 (Router LSA, intra-area), Type 2 (Network LSA, originated by DR), Type 3 (Summary LSA, originated by ABR between areas), Type 4 (ASBR Summary LSA), and Type 5 (AS External LSA). Nokia SR OS uses a default reference bandwidth of 100 Gbps (100,000,000 kbps), so a 10 Gbps port has cost 10 and a 1 Gbps port has cost 100.

Unlike OSPF (which runs over IP protocol 89), IS-IS encapsulates its control PDUs (IIH, LSP, CSNP, PSNP) directly inside Layer 2 Data Link frames. Each router is identified by an ISO Network Entity Title (NET) comprising a variable Area ID (1–13 bytes), a 6-byte System ID, and a 1-byte NSEL selector set to 0x00. In IS-IS, an entire router belongs to an area: Level 1 routes intra-area, Level 2 forms the contiguous inter-area backbone, and L1/L2 routers act like ABRs. On broadcast LANs, Designated Intermediate System (DIS) election is preemptive, has no backup DIS, and priority 0 remains eligible. The current 2026 4A0-100 course explicitly teaches OSPF; Nokia’s announced March 9, 2027 refresh replaces it with IS-IS. This comparison is background, not evidence that the future outline already governs the current exam.

BGP (TCP 179) progresses through Idle -> Connect -> Active -> OpenSent -> OpenConfirm -> Established using OPEN, UPDATE, NOTIFICATION, and KEEPALIVE messages. Well-Known Mandatory attributes (required when advertising classic IPv4-unicast NLRI, not in withdrawal-only UPDATEs) are ORIGIN, AS_PATH, and NEXT_HOP; LOCAL_PREF is Well-Known Discretionary; MED is Optional Non-Transitive. For routes with a resolvable NEXT_HOP, an introductory comparison of major preferences is: (1) highest LOCAL_PREF (default 100, AS-wide), (2) shortest AS_PATH, (3) lowest ORIGIN (IGP < EGP < Incomplete), (4) lowest MED where comparison is applicable (normally routes from the same neighboring AS), (5) EBGP over IBGP, (6) lowest IGP metric to NEXT_HOP, and (7) lowest BGP Router ID / peer IP. This is not the complete SR OS best-path algorithm: local-origin, policy and further tie-breakers can affect selection. Multiprotocol address families also carry reachability differently.

Because AS_PATH is only modified across EBGP boundaries, a standard non-reflector IBGP speaker does not re-advertise a route learned from one IBGP peer to another IBGP peer (IBGP split horizon), requiring N(N-1)/2 logical TCP sessions for a full mesh of N routers (e.g., 10 routers = 45 sessions). A Route Reflector (RR) relaxes split horizon by reflecting client-learned routes to clients and non-clients, and non-client-learned routes to clients, using ORIGINATOR_ID and CLUSTER_LIST attributes to prevent loops. Because IBGP does not alter NEXT_HOP by default, edge routers commonly apply 'next-hop-self' so internal peers can resolve the route via the IGP.

Inserted between the Layer 2 header and Layer 3 payload, the 32-bit (4-byte) MPLS header contains: a 20-bit Label value (0 to 1,048,575), a 3-bit Traffic Class (TC / EXP) field for QoS, a 1-bit Bottom-of-Stack (S) bit (set to 1 on the innermost label), and an 8-bit TTL field. Along a unidirectional Label Switched Path (LSP), the ingress Label Edge Router (iLER) performs a Push operation, intermediate Label Switch Routers (LSRs) perform a Swap operation using their LFIB, and label removal occurs at egress or one hop earlier through PHP. Additional labels and operations depend on the service.

Reserved MPLS labels 0–15 include: 0 = IPv4 Explicit Null, 1 = Router Alert, 2 = IPv6 Explicit Null, and 3 = Implicit Null. Implicit Null (Label 3) is a control-plane signaling value advertised by the egress LER to its upstream neighbor via LDP or RSVP-TE; it never appears inside a data-plane MPLS shim header on the wire. Instead, it instructs the penultimate LSR to pop the outer transport label one hop early (PHP), saving the egress PE from performing two label lookups when an inner VPN service label is present.

Link LDP sends multicast Hello messages to 224.0.0.2 on UDP port 646 (Targeted LDP sends unicast UDP 646 Hellos to remote PEs). The router with the higher IPv4 LDP transport address (usually the system IP) initiates the active TCP port 646 connection to exchange Initialization, KeepAlive, Address, and Label Mapping messages. Nokia SR OS supports Downstream Unsolicited distribution, Ordered Control, and Liberal Label Retention (keeping all neighbor bindings in the LIB while installing only the active IGP next-hop's binding into the LFIB), with export policy governing additional advertised prefixes; distinguish default system-address transport FECs from service labels and explicitly exported routes.

In a basic LDP deployment, transport follows IGP-resolved paths; policies, tunneling and repair mechanisms can alter that behavior. By contrast, RSVP-TE supports traffic-engineered LSPs computed by the ingress LER using Constrained Shortest Path First (CSPF) against the Traffic Engineering Database (checking bandwidth reservations, hop limits, and admin-groups/SRLGs). The ingress LER sends an RSVP PATH message downstream containing a LABEL_REQUEST object, and the egress LER returns an RSVP RESV message upstream carrying the allocated MPLS label. RSVP-TE also supports local Fast Reroute (FRR), often designed for sub-50ms restoration but not guaranteeing that measured result, via one-to-one detour LSPs or facility bypass tunnels.

A Customer object records administrative ownership. A Service Access Point (SAP) is the local customer demarcation point defined by a physical port (or LAG) plus encapsulation identifier (e.g., sap 1/1/2:100). A Service Distribution Point (SDP) is a unidirectional logical transport tunnel (using LDP, RSVP-TE, or GRE) to a far-end PE's system IP. An SDP binding (spoke-sdp or mesh-sdp) attaches a specific service to an SDP using a 32-bit Virtual Circuit Identifier (VC-ID); while local Service IDs and SDP IDs only have local significance, the VC-ID must match on both PEs.

In a basic two-label MPLS VPN example, when a customer frame enters a SAP on the ingress PE, the PE pushes an inner 20-bit Service Label (signaled end-to-end between PEs via Targeted LDP for Epipe/VPLS or MP-BGP for VPRN, with Bottom-of-Stack S = 1) and an outer 20-bit Transport Label (signaled hop-by-hop via Link LDP or RSVP-TE, with S = 0). Core P routers inspect and swap only the outer Transport Label without examining the inner label or customer payload. At the egress PE, the outer label is popped (or removed one hop early via PHP), and the inner Service Label identifies the exact target VPLS, Epipe SAP, or VPRN VRF. Each label value is 20 bits inside a 32-bit shim entry; two complete entries add eight bytes, not five. Real stacks may contain additional labels.

An Epipe is a point-to-point Layer 2 Ethernet Virtual Leased Line (VPWS) that connects either two local SAPs on the same router or two remote SAPs across an MPLS network using a spoke-sdp pseudowire signaled by Targeted LDP. Unlike VPLS, an Epipe performs no MAC address learning, maintains no Forwarding Database (FDB), and runs no Spanning Tree; subject to admission, filtering and forwarding state, a valid frame entering one endpoint (SAP or spoke-sdp) is cross-connected directly out the opposite endpoint.

A VPLS emulates a multipoint Layer 2 Ethernet switch across an IP/MPLS backbone. Each PE maintains an independent per-service FDB, learning source MAC addresses from frames arriving on local SAPs (associating the MAC with the SAP) and remote SDPs (associating the MAC with the spoke-sdp or mesh-sdp). Unknown unicast, multicast, and broadcast frames are flooded to eligible other SAPs and SDPs, excluding the ingress and respecting configured flood/filter controls in the VPLS, subject to the mesh-sdp split-horizon rule: a frame received on a mesh-sdp is never forwarded out another mesh-sdp in the same split-horizon group, in a conventional mesh-SDP VPLS design requiring a full logical mesh of pseudowires between participating PEs.

IES provides customer Layer 3 IP interfaces (with SAPs) directly inside the router's global 'Base' routing instance for dedicated Internet access. VPRN (RFC 4364 Layer 3 BGP/MPLS IP VPN) creates an isolated per-customer Virtual Routing and Forwarding (VRF) table on each PE. A 64-bit Route Distinguisher (RD) is prepended to an IPv4 prefix to create a globally unique 96-bit VPN-IPv4 address family route so overlapping customer subnets remain distinct in MP-BGP. A 64-bit Route Target (RT) extended community controls which VRFs import and export those routes, while MP-BGP advertises the inner VPN service label and 'auto-bind-tunnel' resolves the next hop to an MPLS transport tunnel.

An autonomous system presents a coherent routing policy to other networks; BGP exchanges reachability between autonomous systems. Transit typically purchases connectivity to the wider Internet through an upstream network. Peering exchanges the participating networks’ own and customer traffic under their agreement, rather than automatically providing unrestricted transit. An Internet exchange point can provide a shared place to interconnect for peering. Peering may be paid or settlement-free, and joining an exchange alone does not establish every desired route.