Pkg.add("Metis")

On systems without a pre-packaged Metis library, adding this package will download, configure and install metis-5.1.0 in the directory

Pkg.dir("Metis", "deps")

Configuration requires Cmake version 2.8 or later. At present the build is only available on Linux/Unix. Contributions of build stanzas for Windows or OS X are welcome.

The names of the available Julia functions are those from the Metis API

• nodeND(al) : recursively bisect the undirected graph implied by the adjacency list, al, and return the fill-reducing permutation that results
• nodeND(m) : recursively bisect the undirected graph of the nonzero structure of the symmetric sparse matrix, m, and return the fill-reducing permutation
• vertexSep(al): compute a vertex separator for the adjacency list, al, of an undirected graph
• vertexSep(m): compute a vertex separator for the undirected graph of the nonzero structure of the symmetric sparse matrix, m
• partGraphKway(al, nparts::Integer): partition a graph given as an adjacency list, al, into nparts
• partGraphRecursive(al, nparts::Integer): partition a graph given as an adjacency list, al, into nparts

The function Metis.testgraph can be used to read one of the sample graphs available in the graphs directory of metis-5.1.0. These graphs correspond to 2D and 3D finite element meshes.

4elt : a smaller sample graph (15606 vertices, 45878 edges) copter2 : a medium size sample graph (55476 vertices, 352238 edges) mdual : a larger sample graph (258569 vertices, 513132 edges)

using Graphs, Metis
copter2 = Metis.testgraph("copter2");
perm, iperm = nodeND(copter2)
sizes, part = vertexSep(copter2)
objval, part = partGraphKway(copter2, 6)
counts = zeros(Int, 6);
for p in part counts[p] += 1 end
println(counts)