Network Glossary
ISP technical terms explained in plain language — from engineers who operate provider networks daily.
AS — Autonomous System
A group of IP networks under a single administration, identified by an ASN. Every ISP needs at least one AS to advertise its routes via BGP.
Anycast — Anycast Routing
Multiple servers share the same IP from different locations; the network delivers each packet to the closest one. Common in DNS (8.8.8.8, 1.1.1.1) and CDNs.
BFD — Bidirectional Forwarding Detection
Protocol that detects link failures in milliseconds. Used alongside BGP, OSPF and MPLS for fast convergence. Essential in networks with tight SLA requirements.
BGP — Border Gateway Protocol
The routing protocol between Autonomous Systems — what makes the internet work between ISPs. Misconfigured BGP can take you off the internet or leak someone else's routes.
In practice, BGP decides where your ISP traffic enters and leaves the internet. The edge router keeps sessions with upstreams (IP transit), with IX.br route servers and with private peers — CDNs like Google, Meta and Netflix when there is enough volume. Each session exchanges prefixes: you announce your blocks and receive routes from the rest of the internet. The full routing table is past 1 million IPv4 prefixes, which defines how much memory and CPU the edge needs.
The mistakes we see most at ISPs: announcing prefixes without outbound filters (route leaks that can drop the upstream session), forgetting the RPKI ROA and having routes silently rejected by validating peers, and running the edge with a single upstream — when that link goes down, the whole ISP goes offline. Day to day, communities, prepend and local-preference are the tools to steer inbound and outbound traffic.
How Rasys runs this front: BGP for ISPs.
BNG — Broadband Network Gateway
Evolution of the B-RAS. Combines subscriber session termination (PPPoE, IPoE), QoS policy enforcement, CGNAT and traffic forwarding in a single control point.
B-RAS — Broadband Remote Access Server
Server that terminates subscriber PPPoE sessions and routes traffic. Huawei NE8000/NE40, Juniper MX, Cisco ASR. Single point of failure without redundancy.
CGNAT — Carrier-Grade NAT
Large-scale NAT that lets you share one public IPv4 across many subscribers. Cheaper than buying IPv4, but breaks some services (P2P, port forwarding, gaming).
CGNAT exists because IPv4 ran out: instead of one public IP per subscriber, an IP is shared by dozens of them, with the CGNAT box translating addresses and ports. Common ratios in Brazil sit around 1:32, reaching 1:64 when the network has IPv6 well deployed — the more traffic leaves natively over IPv6, the fewer IPv4 ports each subscriber consumes.
The critical part is legal: Brazilian law (Marco Civil) requires keeping connection records, and with NAT that means logging IP, port and timestamp of every translation — or using port block allocation to drastically reduce log volume. Without those records, the ISP cannot identify the subscriber when a court order arrives, and the liability is theirs. Pool sizing and allocation policy decide whether CGNAT goes unnoticed or becomes a source of tickets.
How Rasys runs this front: CGNAT for ISPs.
DNS — Domain Name System
Translates names (rasys.net) into IPs (200.x.x.x). ISPs need to run their own resolvers or point to public ones. Slow DNS is the first thing customers complain about.
ECMP — Equal-Cost Multi-Path
Distributes traffic across multiple equal-cost paths. Common in OSPF and BGP to load-balance across uplinks without additional protocols. Requires consistent per-flow hashing.
GPON — Gigabit Passive Optical Network
Passive FTTH technology (no power in the path). One OLT serves up to 128 ONUs per port via optical splitter. Dominant standard for residential fiber in Brazilian ISPs.
HSRP — Hot Standby Router Protocol
Cisco proprietary gateway redundancy protocol. One active router, one standby; automatic failover. Equivalent to VRRP (open standard) and GLBP (with load balancing).
IPv6 — Internet Protocol version 6
Next-generation IP with 128-bit addresses. Solves IPv4 scarcity. ISPs without IPv6 tend to depend more on CGNAT and may see worse experience on modern services, CDNs, and latency-sensitive applications.
For an ISP, IPv6 deployment starts at LACNIC: the typical provider block is a /32, which yields thousands of /56 prefixes — what each subscriber receives via DHCPv6 Prefix Delegation on a dual-stack PPPoE session. With a solid addressing plan, delegation stays predictable, reverse DNS works and the B-RAS delivers IPv4 (through CGNAT) and native IPv6 on the same session.
The practical payoff: the biggest traffic sources (Google, Netflix, Meta, CDNs in general) already speak IPv6, so every connection that leaves natively is one that does not consume a CGNAT port. That improves gaming and latency-sensitive applications and postpones IPv4 pool investment. The edge must follow: IPv6 BGP sessions with upstreams and IX.br, and RPKI ROAs registered for the v6 prefixes too.
How Rasys runs this front: IPv6 for ISPs.
IX — Internet Exchange Point
Location where ISPs interconnect directly without paying for IP transit. In Brazil, IX.br has PoPs in major cities. Peering at IX cuts latency and cost.
LACP — Link Aggregation Control Protocol
802.3ad protocol that bundles multiple physical links into a single logical channel. Increases bandwidth and adds redundancy. Essential on server uplinks and core switch interconnects.
MPLS — Multiprotocol Label Switching
Switches packets based on labels instead of IP destination. Used in large backbones to build L2/L3 VPNs and perform traffic engineering. Common in multi-PoP ISPs.
With MPLS, packets cross the backbone guided by labels rather than per-hop route lookups. On top of that base come the services ISPs care about: L2 VPN (VPLS and, more modern, EVPN) to deliver point-to-point or multipoint Ethernet between PoPs, and L3 VPN (BGP/MPLS) to keep corporate customer traffic separated from subscriber traffic, each in its own VRF.
It makes sense when the ISP has multiple PoPs and needs per-customer traffic separation, VLAN transport between cities or protected paths — fast reroute restores traffic in tens of milliseconds when a link fails. For a small single-PoP network the complexity rarely pays off; for a ring backbone with dedicated customers, it is the standard tool.
How Rasys works with this: MPLS for ISPs.
MSS — Maximum Segment Size
Maximum TCP segment size without fragmentation. Must be adjusted when CGNAT, tunneling or PPPoE is in play (MTU below 1500). Wrong MSS silently breaks HTTPS connections.
MTU — Maximum Transmission Unit
Maximum packet size on a given link. Standard Ethernet: 1500 bytes. PPPoE drops it to 1492. MPLS and tunnels reduce it further. Incorrect MTU causes intermittent connection drops.
OLT — Optical Line Terminal
Central device of a GPON/EPON FTTH network. Connects the ISP backbone to the ONUs (subscriber side). Huawei MA5800, ZTE C300, FiberHome AN5516 are common.
The OLT is where the subscriber optical network meets the backbone: each PON port serves dozens of ONUs behind passive splitters (1:64 and 1:128 are common topologies), and profile configuration — line profile and service profile in Huawei terms — defines bandwidth, VLANs and the service each subscriber gets. Every vendor has its own provisioning logic, and running a mixed plant means knowing the differences between Huawei, ZTE, FiberHome and Datacom.
In troubleshooting, optical levels are the ruler: ONU RX outside the expected range points to a dirty connector, macrobend or splitter issue; LOS on several ONUs of the same splitter points to a break in the shared segment. Vendor migration and profile standardization are the projects that show up most when an ISP grows by acquisition.
How Rasys runs this front: Multi-vendor OLT GPON.
OSPF — Open Shortest Path First
Interior gateway protocol based on link-state. Calculates shortest paths inside the ISP's AS. Faster than RIP, simpler than IS-IS.
Inside the AS, OSPF has a specific job: distributing loopbacks and internal links so iBGP has a reachable next-hop at every PoP. Each router announces its link states, all of them compute the full topology and the cost of every path — which is why convergence is fast and predictable, as long as the area design matches the network size.
In operation: BFD coupled to OSPF drops the adjacency in milliseconds when a link fails instead of waiting for timers; OSPFv3 covers IPv6; and redistributing between OSPF and BGP without filters is the classic recipe for loops or a bloated internal table. An adjacency that will not come up is usually mismatched MTU, wrong area or diverging authentication.
How Rasys works with this: OSPF for ISPs.
Peering — BGP Peering
Direct route exchange between two ASes without paying transit. Can be bilateral (direct agreement) or multilateral (via IX). Cuts bandwidth cost and improves latency.
PPPoE — Point-to-Point Protocol over Ethernet
Authentication protocol still dominant in Brazilian ISPs. Client authenticates via login/password over a PPP session encapsulated in Ethernet. The B-RAS terminates the session.
The flow: the subscriber ONU or router starts PPPoE discovery, the B-RAS answers, the PPP session comes up and authentication goes to the management system RADIUS, which returns contracted speed, IP pool and other attributes. Since the PPPoE header takes 8 bytes, session MTU drops to 1492 — which is why bad MSS adjustment causes intermittent stalls on HTTPS sites.
Concentrator sizing is what separates a quiet operation from a bad night: simultaneous sessions, throughput per slot and redundancy to avoid a single point of failure. In smaller networks a MikroTik CCR2116 with broad IPv6 handles up to roughly 800-1,200 PPPoE subscribers well; above that the conversation moves to Huawei NE or Juniper MX. Debugging stuck sessions and intermittent authentication is routine work for us.
How Rasys runs this front: PPPoE, RADIUS and B-RAS/BNG for ISPs.
QinQ — IEEE 802.1ad (Double VLAN)
Encapsulates a VLAN inside another VLAN (S-TAG + C-TAG). Isolates customer traffic on metro Ethernet without exhausting the VLAN space (4096 IDs). Used in ISP backbones.
RADIUS — Remote Authentication Dial-In User Service
Central authentication, authorization and accounting (AAA) system that validates PPPoE login, enforces contracted speed and tracks online time. FreeRADIUS is the most common base.
At an ISP, RADIUS usually comes embedded in the management system (MK-Auth, IXC, SGP, Voalle) and talks to the B-RAS at three moments: Access-Request when the subscriber authenticates, response attributes that define speed, pool and IPv6 addressing, and accounting that records the session. Vendor-specific attributes vary — what a Huawei concentrator understands is not what a MikroTik expects.
When authentication fails intermittently, the problem can be anywhere along the chain: B-RAS, management network, the RADIUS server or the ERP registration. End-to-end debugging — capturing what the concentrator sends and what RADIUS answers — is what shortens that diagnosis. Another sensitive point: the management network carrying this traffic needs hardening, because subscriber credentials travel through it.
How Rasys runs this front: PPPoE, RADIUS and B-RAS/BNG for ISPs.
RPKI — Resource Public Key Infrastructure
BGP route certification system. Validates that whoever announces a prefix has authorization from the IP block owner. Reduces risk of route hijacking and BGP leaks.
VRRP — Virtual Router Redundancy Protocol
Open standard (RFC 5798) for gateway redundancy. A virtual router shares IP and MAC between devices; the active master handles traffic, the backup takes over on failure. Open alternative to HSRP.
Want to dig deeper?
The Rasys team has over 10 years of experience operating ISP networks in Brazil. If any of these terms is a headache in your operation, talk to us — initial conversation, no commitment.