Gunrock BFS on kron21

[jowens]

Our runs below result in substantially different performance numbers than those reported by the Groute authors.

• 3xK80+METIS: avg: 120.67ms, min: 114.33ms, max: 216.01ms
  https://github.com/gunrock/io/blob/master/gunrock-output/20170303/bfs_k80x3_metis_kron_g500-logn21.txt
• 3xK80+random: avg: 67.70 ms, min: 55.24 ms, max: 143.39 ms
  https://github.com/gunrock/io/blob/master/gunrock-output/20170303/bfs_k80x3_rand_kron_g500-logn21.txt
• 3xK80+idempotence+METIS+market: avg: 35.34 ms, min: 26.75 ms, max: 78.21 ms
  https://github.com/gunrock/io/blob/master/gunrock-output/20170303/bfs-idem_k80x3_metis_kron_g500-logn21.txt
• 3xK80+idempotence+random+market: avg: 22.70ms, min: 16.68ms, max: 56.84ms
  https://github.com/gunrock/io/blob/master/gunrock-output/20170303/bfs-idem_k80x3_rand_kron_g500-logn21.txt

Single-GPU runs:

• 1xK80, no idempotence, DOBFS: avg: 4.60 ms, min: 3.77 ms, max: 5.03 ms
  https://github.com/gunrock/io/blob/master/gunrock-output/20170303/dobfs_k80x1_kron_g500-logn21.txt
• 1xK80, with idempotence, DOBFS: avg: 4.53ms, min: 3.68ms, max: 4.70ms
  https://github.com/gunrock/io/blob/master/gunrock-output/20170303/dobfs-idem_k80x1_kron_g500-logn21.txt

[sgpyc]

The 3rd item is 3 x K80 + idempotence + Metis, and the 4th item is 3 x K80 + idempotence + random.

Both single GPU runs are for DOBFS.

[jowens]

Edited original issue to reflect @sgpyc's comment above.