```
julia> A = rand(1000,1000); A = A + A'; A = sparse(A);
julia> e = ones(1000);
julia> @time A * e
0.054203 seconds (17.26 k allocations: 990.259 KiB)
julia> @time A * e;
0.001510 seconds (5 allocations: 8.094 KiB)
julia> @time A * e;
0.001486 seconds (5 allocations: 8.094 KiB)
julia> B = Symmetric(A);
julia> @time B * e;
0.220572 seconds (70.48 k allocations: 3.445 MiB)
julia> @time B * e;
0.062014 seconds (5 allocations: 8.094 KiB)
julia> @time B * e;
0.063677 seconds (5 allocations: 8.094 KiB)
```

Why is the product with `B`

so much slower?

With the fairly large symmetric sparse matrix https://www.cise.ufl.edu/research/sparse/MM/Schenk_AFE/af_shell8.tar.gz, MatrixDepot downloads a `Symmetric{...}`

matrix and a matrix-vector product never returns (I’m using Julia 0.5.2). By contrast, downloading the file directly and using MatrixMarket, the matrix-vector product is almost instantaneous (the matrix isn’t cast as `Symmetric{...}`

).

I’m not able to test Julia 0.6 because loading the matrix using MatrixMarket never returns (see https://github.com/JuliaSparse/MatrixMarket.jl/issues/27). The same happens with MatrixDepot, which presumably relies on MatrixMarket.jl.

Thanks,