TS: re-add --resume

4 files changed, 11 insertions, 12 deletions

diff --git a/TS/dimes-hetsim-hbm.sh b/TS/dimes-hetsim-hbm.sh
index 192a705..0b57ab2 100755
--- a/TS/dimes-hetsim-hbm.sh
+++ b/TS/dimes-hetsim-hbm.sh
@@ -5,7 +5,7 @@ make -B NUMA=1
 make inputs/randomlist10M.txt
 
 mkdir -p log/$(hostname)
-fn=log/$(hostname)/$(date +%Y%m%d)
+fn=log/$(hostname)/dimes-hetsim-hbm
 
 # Upstream DPU version uses ts_size = 33554432 elements and query_length = 256 with int32_t data
 # Upstream CPU version uses inputs/randomlist33M.txt with 33618177 elements and query_length = 256 with double
@@ -22,7 +22,7 @@ export -f run_benchmark
 
 echo "single-node execution (1/2)" >&2
 
-parallel -j1 --eta --joblog ${fn}.1.joblog --header : \
+parallel -j1 --eta --joblog ${fn}.1.joblog --resume --header : \
 	run_benchmark nr_threads={nr_threads} ram={ram} cpu={cpu} \
 	::: i $(seq 0 5) \
 	::: nr_threads 1 2 4 8 12 16 \
@@ -31,13 +31,11 @@ parallel -j1 --eta --joblog ${fn}.1.joblog --header : \
 
 echo "multi-node execution (2/2)" >&2
 
-parallel -j1 --eta --joblog ${fn}.2.joblog --header : \
+parallel -j1 --eta --joblog ${fn}.2.joblog --resume --header : \
 	run_benchmark nr_threads={nr_threads} ram={ram} cpu={cpu} \
 	::: i $(seq 0 20) \
 	::: nr_threads 32 48 64 96 128 \
 	::: cpu -1 \
 	::: ram $(seq 0 15)
 
-) > ${fn}.txt
-
-xz -f -v -9 -M 800M ${fn}.txt
+) >> ${fn}.txt
diff --git a/TS/dimes-hetsim-nmc.sh b/TS/dimes-hetsim-nmc.sh
index bf8eafe..14ed35c 100755
--- a/TS/dimes-hetsim-nmc.sh
+++ b/TS/dimes-hetsim-nmc.sh
@@ -5,7 +5,7 @@ fn=log/$(hostname)/$(date +%Y%m%d)
 
 # Upstream DPU version uses ts_size = 33554432 elements and query_length = 256 with int32_t data
 # Upstream CPU version uses inputs/randomlist33M.txt with 33618177 elements and query_length = 256 with double
-# This benchmark uses int32 and 33554432 elements for both.
+# This benchmark uses int32 and 2²⁵ == 33554432 elements for both.
 
 run_benchmark_nmc() {
 	local "$@"
diff --git a/TS/run-paper-strong-full.sh b/TS/run-paper-strong-full.sh
index bb87a93..5b7656d 100755
--- a/TS/run-paper-strong-full.sh
+++ b/TS/run-paper-strong-full.sh
@@ -15,12 +15,13 @@ echo "Started at $(date)"
 echo "Revision $(git describe --always)"
 
 # >2048 is not part of upstream
+# 12 tasklets are not part of upstream (code does not work with 16…)
 for nr_dpus in 2543 2304 256 512 1024 2048; do
-	for nr_tasklets in 1 2 4 8 16; do
+	for nr_tasklets in 1 2 4 8 12 16; do
 		echo
 		# upstream code did not respect $BL in the makefile and used 256B (BL=8) instead.
-		# BL=10 appears to be slightly faster.
-		if make -B NR_DPUS=${nr_dpus} NR_TASKLETS=${nr_tasklets} BL=10; then
+		# This appears to be faster than BL=10.
+		if make -B NR_DPUS=${nr_dpus} NR_TASKLETS=${nr_tasklets} BL=8; then
 			timeout --foreground -k 1m 30m bin/ts_host -w 0 -e 100 -n 33554432 || true
 		fi
 	done
diff --git a/TS/run-paper-weak.sh b/TS/run-paper-weak.sh
index f97d2bd..64892f4 100755
--- a/TS/run-paper-weak.sh
+++ b/TS/run-paper-weak.sh
@@ -15,12 +15,12 @@ echo "Started at $(date)"
 echo "Revision $(git describe --always)"
 
 # 256 and 512 are not part of upstream
-for nr_dpus in 512 256 1 4 16 64; do
+for nr_dpus in 1 4 16 64; do
 	for nr_tasklets in 1 2 4 8 16; do
 		echo
 		# upstream code did not respect $BL in the makefile and used 256B (BL=8) instead.
 		# BL=10 appears to be slightly faster.
-		if make -B NR_DPUS=${nr_dpus} NR_TASKLETS=${nr_tasklets} BL=10; then
+		if make -B NR_DPUS=${nr_dpus} NR_TASKLETS=${nr_tasklets} BL=8; then
 			i=$(( nr_dpus * 524288 ))
 			timeout --foreground -k 1m 30m bin/ts_host -w 0 -e 100 -n $i || true
 		fi