For general use, I would like to recommend Hudson's ms, which is
definitely the most popular one of all. This masterpiece is also very
efficient, qualified for most practical uses. MaCS seems also a
competent candidate, but maybe more publications are needed to justify
its practical performance.
I have used the following softwares or carefully read their
publications and documentations:
ms, by Hudson.
Type: standard coalescent-with-recombination; backward in time;
continuous time
Features: allows for simulating various events such as variable
population size, recombination, gene conversion, migration and
admixture.
Features: much faster than ms while getting similar results
to ms. Support complex demographic models including
variable population sizes, admixtures and recombination
hotspots.
Notes: I have briefly tried MaCS. It is lightening fast. For the
moment, I do not know whether MaCS delivers similar results to ms
on structured populations.
Type: SMC and SMC', approximation to the standard
coalescent; move along the sequence; continuous time
Features: much faster than ms for large-scale simulation
Limitations: constant population size or exponential growth
only. To some extent, MaCS is an updated version of FastCoal. If
you want to simulate large data set, try MaCS.
Type: approximation to the standard coalescent (see below); backward in time; discrete time
Approximation: The following sentence indicate the difference
between GENOME and ms: "In the standard approach, each
coalescent event involves exactly two sequences that coalesce to a
common ancestor but, using our approach, multiple sequences can
coalesce to a common ancestor simutaneously". I am not sure
how this may affect the accuracy of simulation, but the comparison
between ms and GENOME seems to be satisfactory.
Features: variable population size, recombination and symmetric
migration. Claim to be fast.
Notes: My experience is GENOME is not as fast as MaCS, although
the performance is highly related to the heuristic parameters in
use. In addition, this software requires recombinations to happen
between segments which may affect the accuracy of very ancient
recombinations.
Type: apparently standard coalescent-with-recombination; forward
in time
Features: from its website: "SFS_CODE performs forward
population genetic simulations under a general Wright-Fisher
island model with arbitrary demographic, selective, and mutational
effects".