In the 90s, it was (probably) faster to do linked-list / chaining as a collision resolution mechanism. But because L1 caches are so fast on today's machines (and DDR4 remains very high latency), linear probing seems to be the winner on modern machines.
It depends on the use case, for example if it's read or write heavy, the number of successful lookups and the load factor. LP is often a good choice but in some cases bucket chaining wins. Quadratic probing and Robin Hood hashing are also winning candidates, as is Cuckoo hashing.
See the flowchart at the end of this paper: Richter et al., 2005. A Seven-Dimensional Analysis of Hashing Methods and its Implications on Query Processing, https://dl.acm.org/citation.cfm?id=2850585
Comments
It depends on the use case, for example if it's read or write heavy, the number of successful lookups and the load factor. LP is often a good choice but in some cases bucket chaining wins. Quadratic probing and Robin Hood hashing are also winning candidates, as is Cuckoo hashing.
See the flowchart at the end of this paper: Richter et al., 2005. A Seven-Dimensional Analysis of Hashing Methods and its Implications on Query Processing, https://dl.acm.org/citation.cfm?id=2850585