|
|
@@ -151,11 +151,11 @@ comments) to divide your system into two partitions as separate as
|
|
|
possible: separate NUMA nodes if you have them, otherwise separate CPUs
|
|
|
if you have them, otherwise separate cores. If each of your two
|
|
|
partitions has _n_ cores, ensure that the elements of `CORESLIST` do not
|
|
|
-exceed _n_ (but of course you will not be able to replicate those data
|
|
|
-points in that case). The paper uses values of _n_ up to 32 cores in
|
|
|
-each partition, so 64 cores in total; as above, P2 can reuse the cores
|
|
|
-of P0, since P2's main work is done after P0 and P1's main work has
|
|
|
-finished.
|
|
|
+exceed _n_ (of course you will not be able to replicate those data
|
|
|
+points in that case, but the trend should still be apparent). The paper
|
|
|
+uses values of _n_ up to 32 cores in each partition, so 64 cores in
|
|
|
+total; as above, P2 can reuse the cores of P0, since P2's main work is
|
|
|
+done after P0 and P1's main work has finished.
|
|
|
|
|
|
So the steps are:
|
|
|
|
