1. First level

2. Second and third levels

3. Notes

4. References

The Luleå algorithm of computer science, designed by {{harvtxt|Degermark|Brodnik|Carlsson|Pink|1997}}, is a technique for storing and searching internet routing tables efficiently. It is named after the Luleå University of Technology, the home institute of the technique's authors. The name of the algorithm does not appear in the original paper describing it, but was used in a message from Craig Partridge to the Internet Engineering Task Force describing that paper prior to its publication.^[1]

The key task to be performed in internet routing is to match a given IPv4 address (viewed as a sequence of 32 bits) to the longest prefix of the address for which routing information is available. This prefix matching problem may be solved by a trie, but trie structures use a significant amount of space (a node for each bit of each address) and searching them requires traversing a sequence of nodes with length proportional to the number of bits in the address. The Luleå algorithm shortcuts this process by storing only the nodes at three levels of the trie structure, rather than storing the entire trie.

The main advantage of the Luleå algorithm for the routing task is that it uses very little memory, averaging 4–5 bytes per entry for large routing tables. This small memory footprint often allows the entire data structure to fit into the routing processor's cache, speeding operations. However, it has the disadvantage that it cannot be modified easily: small changes to the routing table may require most or all of the data structure to be reconstructed.

The Luleå algorithm is patented in the United States ({{harvnb|Degermark|Brodnik|Carlsson|Pink|2001}}).

First level

The first level of the data structure consists of

• A bit vector consisting of 2¹⁶ = 65,536 bits, with one entry for each 16-bit prefix of an IPv4 address. A bit in this table is set to one if there is routing information associated with that prefix or with a longer sequence beginning with that prefix, or if the given prefix is the first one associated with routing information at some higher level of the trie; otherwise it is set to zero.
• An array of 16-bit words for each nonzero bit in the bit vector. Each datum either supplies an index that points to the second-level data structure object for the corresponding prefix, or supplies the routing information for that prefix directly.
• An array of "base indexes", one for each consecutive subsequence of 64 bits in the bit vector, pointing to the first datum associated with a nonzero bit in that subsequence.
• An array of "code words", one for each consecutive subsequence of 16 bits in the bit vector. Each code word is 16 bits, and consists of a 10-bit "value" and a 6-bit "offset". The sum of the offset and the associated base index gives a pointer to the first datum associated with a nonzero bit in the given 16-bit subsequence. The 10-bit value gives an index into a "maptable" from which the precise position of the appropriate datum can be found.

To look up the datum for a given address x in the first level of the data structure, the Luleå algorithm computes three values:

1. the base index at the position in the base index array indexed by the first 10 bits of x
2. the offset at the position in the code word array indexed by the first 12 bits of x
3. the value in maptable[y][z], where y is the maptable index from the code word array and z is bits 13–16 of x

The sum of these three values provides the index to use for x in the array of items.

Second and third levels

The second and third levels of the data structure are structured similarly to each other; in each of these levels the Luleå algorithm must perform prefix matching on 8-bit quantities (bits 17–24 and 25–32 of the address, respectively). The data structure is structured in "chunks", each of which allows performing this prefix matching task on some subsequence of the address space; the data items from the first level data structure point to these chunks.

If there are few enough different pieces of routing information associated with a chunk, the chunk just stores the list of these routes, and searches through them by a single step of binary search followed by a sequential search. Otherwise, an indexing technique analogous to that of the first level is applied.

Notes

References

• {{citation

• {{citation

• {{citation

• Carlton Complex Fire
• Carlton (County Durham)
• Carlton County (Minnesota)
• Carlton Cricket Club (Barbados)
• Carlton Cycles
• Carlton Day Reed Jr.
• Carlton D. Everhart II
• Carlton Draught Stakes
• Carlton Dry
• Carlton (Falmouth, Virginia)
• Carlton Faulkner
• Carlton Football Club salary cap breach
• Carlton Fredericks
• Carlton Freeman
• Carlton Gardens, Melbourne
• Carlton Gebbia
• Carlton (Georgia)
• Carlton Gilchrist
• Carlton + Godard = Cinema
• Carlton (Hambleton)
• Carlton Haney
• Carlton Hayes Hospital
• Carlton H. Colwell
• Carlton Hill (disambiguation)
• Carlton Hill, New Jersey