@@ -61,7 +61,7 @@ home rank after the relevant tasks complete.
61
61
62
62
Each of four ranks has three shared blocks. The memory constrains dictate that
63
63
at most four unique shared_id values can coexist on each rank. Under these
64
- memory constraints, it is possible to balance the load (time) well . There
64
+ memory constraints, it is possible to perfectly balance the load (time). There
65
65
is more than one way to do so. The communication cost to migrate a task off-rank
66
66
is extremely low, but the cost to communicate back the result should be
67
67
significant enough to discourage migrating shared_ids to other ranks without it
@@ -73,9 +73,9 @@ rank-averaged load. The sum of the loads for the task corresponding to one of
73
73
its shared_id values is more than the rank-averaged load, so the tasks for that
74
74
shared_id will need to be split across two ranks to achieve good balance. The
75
75
tasks for the other shared_ids across all ranks do not need to be split across
76
- multiple ranks to balance the load (time).
76
+ multiple ranks to perfectly balance the load (time).
77
77
78
- Below is one solution with a well balanced load and decent communication.
78
+ Below is one solution with a perfectly balanced load and decent communication.
79
79
I have not evaluated whether it is optimal.
80
80
81
81
Rank 0:
@@ -97,4 +97,4 @@ Rank 2:
97
97
Rank 3:
98
98
[3,0,0],[3,0,1],[3,0,2] (home)
99
99
[3,1,0],[3,1,1],[3,1,2] (home)
100
- [3,2,0],[3,2,1],[3,2,2] (home)
100
+ [3,2,0],[3,2,1],[3,2,2] (home)
0 commit comments