Commit 53954671 authored by Paul E. McKenney's avatar Paul E. McKenney

rcutorture: Do better bin packing

Running the standard set of rcutorture tests on 24 CPUs results in
the following sub-optimal schedule:

	----start batch----
	 TREE07 16
	----start batch----
	 TREE08 16
	 SRCU-P 8
	----start batch----
	 TREE01 8
	 TREE02 8
	 TREE03 8
	----start batch----
	 TREE04 8
	 TREE05 8
	 TREE06 8
	----start batch----
	 SRCU-N 4
	 TINY01 1
	 TINY02 1
	 TREE09 1

If one of the eight-CPU runs were to be moved into the first batch,
the test suite would complete in four batches rather than five.

This commit therefore uses a greedy algorithm to re-order the test
entries so that the sequential batching will produce an optimal schedule
in this case:

	----start batch----
	 TREE07 16
	 SRCU-P 8
	----start batch----
	 TREE08 16
	 TREE01 8
	----start batch----
	 TREE02 8
	 TREE03 8
	 TREE04 8
	----start batch----
	 TREE05 8
	 TREE06 8
	 SRCU-N 4
	 TINY01 1
	 TINY02 1
	 TREE09 1

Please note that this is still not an optimal bin-packing algorithm,
however, it does produce optimal solutions for most common scenarios.
Signed-off-by: default avatarPaul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: default avatarJosh Triplett <josh@joshtriplett.org>
parent 0ae3f73a
......@@ -214,7 +214,42 @@ do
done
sort -k2nr $T/cfgcpu > $T/cfgcpu.sort
awk < $T/cfgcpu.sort \
awk < $T/cfgcpu.sort > $T/cfgcpu.pack -v ncpus=$cpus '
BEGIN {
njobs = 0;
}
{
cf[njobs] = $1;
cpus[njobs] = $2;
njobs++;
}
END {
alldone = 0;
batch = 0;
nc = -1;
while (nc != ncpus) {
batch++;
nc = ncpus;
for (i = 0; i < njobs; i++) {
if (done[i])
continue;
if (nc >= cpus[i] || nc == ncpus) {
done[i] = batch;
nc -= cpus[i];
if (nc <= 0)
break;
}
}
}
for (b = 1; b <= batch; b++)
for (i = 0; i < njobs; i++)
if (done[i] == b)
print cf[i], cpus[i];
}'
awk < $T/cfgcpu.pack \
-v CONFIGDIR="$CONFIGFRAG/$kversion/" \
-v KVM="$KVM" \
-v ncpus=$cpus \
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment