ietf
[Top] [All Lists]

Checksum at IP layer - is it even needed ?

2015-12-14 09:39:24
In IPv4 and IPv6, pseudo-header mandates checksum at Layer 4, and also cover IP.

In IP-FF, I plan to remove "pseudo-header" from Layer 4 checksum, for
a few reasons:
1. IPsec ESP Transport mode over NAT-PT (NAPT). (one obstacle gets
removed, and I will deal with the rest later -- IPsec-NAT looks like a
can-of-worms for now)
2. ILNP protocol will become possible. (as an extension protocol)
3. better layer separation

...and I don't have checksum at Layer 3 either. (like in IPv6)

Do we actually need chcksums at Layer 3 ?
Disadvantages:
* Slow-down of Core Routers.

Do we even need IP-layer checksums ?
* Most Data-link layer protocols cover "data"; That is Layers 3-to-7
in CRC; one notable exception is ATM. Can I assume that layer 2 CRC
protects me ?
* a single mis-routed packet shouldn't cause too much trouble.
* in IPv6, the options / extensions are not protected by checksum
either. (not at layer 3, not at layer 4 -- maybe only at layer 2)

Any advantages ?
* prevents single mis-routed packets, or a bad option from getting through.

Quick analysis of Data-link layer's checksum protection:

ATM: Header-only, 8-bit.
Frame Relay: 16 bit.
Ethernet and WiFi - 32-bit FCS. (WiFi also has re-transmission)
Token Ring - 32-bit FCS
Point-to-Point Protocol (PPP) with 16 or 32 bits CRC.
4G/LTE: 24-bit CRC, +2nd level CRC.
Firewire IEEE-1394: 32-bit CRC.

Quick result: besides ATM, everybody else seems to cover both headers
and all data.
-- 
-Alexey Eromenko "Technologov"