Pebellslearn by building
0%
RoadmapsMatrixCoursesSandboxSign in
IP — The Layer Underneath  ›  Lesson

Best Effort — What IP Does Not Promise

Reliability 12 minIP only tries to deliver each packet once and promises nothing about delivery, order or timing, damaged packets are silently dropped at every layer, and reliability is built above IP by the endpoints.

Best Effort — What IP Does Not Promise

When a packet carrying "kite 9 cut" vanishes somewhere between Mumbai and a player in Delhi, which part of the network apologises? None of them. IP makes a very small promise, and knowing exactly how small tells you where every other guarantee has to come from.

The rule

IP tries once and promises nothing; reliability lives above it.

What IP can do, its whole capability, is a short list:

  • give each packet a source and a destination address,
  • forward it hop by hop toward that destination,
  • split it into fragments when a link needs smaller packets (IPv4 routers may do this; in IPv6 only the sender may).

What IP does not promise: that a packet arrives, arrives only once, arrives in order, or arrives within any particular time. This is called best-effort delivery. Routers drop packets when their queues overflow, and nobody is told.

Error detection is split between the layers:

  • Each link, such as Ethernet or Wi-Fi, checks every frame with a strong CRC and drops frames that fail. Wi-Fi retries a few times before giving up; Ethernet simply drops.
  • The IPv4 header checksum covers only the IP header, not the data. The TTL changes at every hop, so each router must recompute it. IPv6 has no header checksum at all: the links below and the transport checksums above already do the checking.
  • TCP and UDP checksums cover their own header and the data.

Every one of these checks ends the same way: a damaged packet is dropped. So, seen from the two ends, corruption simply looks like loss.

That leaves the endpoints to notice and repair. TCP does it with acknowledgments and retransmission, for every byte. Manja Arena does it with its ack bits, only for the messages worth resending. Putting the guarantees at the ends, where the application knows what it needs, is known as the end-to-end principle.

The shortcut

For any failure, ask two questions: "Which layer drops it? Which layer notices?" Some layer along the way drops it; only an endpoint ever notices.

The lost "kite 9 cut"Step 1 of 4
  1. Damage

    Radio noise flips a few bits of the frame on the player's home Wi-Fi.

Drills

Answer fast — trust the rule.

quick drill

What does the IPv4 header checksum cover?

quick drill

Why must every router recompute the IPv4 header checksum?

quick drill

How does IPv6 handle header checksums?

quick drill

A router's queue overflows and it drops a Manja Arena packet. Who finds out?

quick drill

Which of these is IP's job?

Remember

  • IP tries once and promises nothing; reliability lives above it.
  • Some layer drops a damaged packet; only an endpoint notices it is gone.
  • Trap: expecting the network to report losses. Silence is what loss looks like.

Practice — level up

0/1 solved

Write your answer, then compare it with a rubric and a model answer.

ExplainWhy our own acks?

A new teammate asks why Manja Arena needs its own acknowledgments when 'the network already has checksums everywhere'. Explain what IP and the checksums do and do not guarantee.