ARP and RARP
LESSON 7

(September 27, 1995)
Daniel Z. Tabor Jr.
New Jersey Institute of Technology

? How does a host or gateway map an IP address to the correct physical address when it needs to send a packet over a physical network ?
Devise a low-level software that hides physical addresses and allows higher-level programs to work only with internet addresses.
Mapping of high-level to low-level addresses is the address resolution problem.

IP address A	conceptual	B IP address
Physical address A	physical	B Physical address

Mapping must be computationally efficient (simply mask all portions of the address excluding the host-id portion)
When address mapping can only be done via an address table (X.25 to IP), hashing is used to speed up lookup.
Problem with representation of 48-bit Ethernet addresses within a 32-bit IP address and allowing new machines to be added without recompilation.
Avoid maintaining a static table of mappings by using the ARP (Address Resolution Protocol).

ARP - is a low-level protocol used to bind addresses dynamically.
ARP allows a host to find a physical address of a target host on the same physical network, given only it’s IP address.
ARP hides the underlying network physical addressing. It can be thought of as part of physical network system and not the internet protocols.

ARP broadcasts special packets with the destination’s IP address to ALL hosts.
The destination host (only) will respond with it’s physical address.
When the response is received, the sender uses the physical address of destination host to send all packets.

Broadcasting is expensive on network resources, so an ARP cache of recently acquired IP-to-Physical address bindings is kept.
Other Broadcast Improvements:
- Include the sender’s IP-to-Physical address binding along with the request, to the destination. This reduces future traffic.
- During each broadcast, ALL machines can find out the senders physical address and record it locally in it’s ARP cache.
- When a new machine comes on-line, it immediately broadcasts it’s IP-to-Physical address binding to all nodes.

There are two main functional parts of the address resolution protocol:
- Determine the destination’s physical address before sending a packet.
- Answer requests that arrive for it’s own Physical-to-IP address binding.
Because of lost/duplicate packets, ARP must handle this to avoid many re-broadcasts.
Bindings in ARP cache (actual cache table) must be removed after a fixed period of time to ensure validity.
When a packet is received, the sender’s IP address is stripped and the local table is updated (ARP cache), then the rest of the packet is processed.
Two types of incoming packets:
- Those to be processed (correct destination).
- Stray broadcast packets (can be dropped after updating the ARP cache).
Application programs may request the destination address many times before the binding is complete. This must be handled, by discarding enqueued requests, when the correct binding returns.

ARP sets the field "TYPE" for the ID of a frame</u>.
ARP packets DO NOT have a fixed format header, so they can be used with arbitrary physical addresses and arbitrary protocol addresses.
The lengths of physical addresses may vary up to 48-bits.

Hardware Type: (16-bits) - the type of interface the sender seeks an answer for.
Protocol Type: (16-bits) - the high-level software address type provided.
HLEN: (8-bits) - length of arbitrary physical address.
PLEN: (8-bits) - length of arbitrary protocol address.
OPERATION: (16-bits) - the specific type of operation requested.
- ARP.request (1)
- ARP.response (2)
SENDER HA: (6-octets) - the sender’s actual hardware address, scalable up to six bytes.
SENDER IP: (4-octets) - the sender’s IP address, always 32-bits.
TARGET HA: (6-octets) - the destination node’s hardware address, scalable up to six bytes.
TARGET IP: (4-octets) - the destination node’s IP address, always 32-bits.

Machines without secondary-storage (diskless workstations) that cannot store their IP addresses need to find out by broadcasting their physical address and requesting a binding to the appropriate IP address.
Terminals and nodes cannot connect to their fileservers without using source and destination IP addresses.
IP addresses should not be bound into a machines bootstrap image. Not doing so allows for flexibility and convenient changes to configuration information.
A connection is needed at start-up to allow the diskless workstation to begin functioning (sends request to a server and waits for the response).
To provide the connection, workstations include a minimal set of TCP/IP protocols in firmware.
Two advantages of using a unique physical address:
- Physical address is available from the network interface hardware and does not have to be bound into the operating system.
- The machine ID depends on the network (not it’s CPU) so all machines will be given uniform and unique IDs.

RARP functionality supports multiple physical network types.
All machines receive RARP requests but only those authorized to supply RARP services can respond. Machines supplying RARP services are called RARP servers.
RARP is only used on LANs with a low probability of failure since bootstrapping requires quick responses.

Primary servers respond while backup servers simply record the time of the incoming requests.
If the primary server is down, backup server will see reoccurring requests and take over. This take over is done by using a random delay to see if other backup servers respond first.

RARP requests and responses use the same frame format as ARP, however the OPERATION field value differs.
OPERATION: (16-bits) - the specific type of operation requested.
- RARP.request (3)
- RARP.response (4)