Adds scripts for running/parsing evaluation.

Author: Jinglei Ren

parse_result.py

@@ -0,0 +1,60 @@
+#!/usr/bin/python
+
+import sys
+import os
+import numpy
+
+# Main
+
+if len(sys.argv) != 2:
+ print "Usage: %s FileName" % sys.argv[0]
+ sys.exit(-1)
+
+path = sys.argv[1]
+
+lines = [line.strip().split('\t') for line in open(path)]
+
+results = { }
+
+db_index = set() # db name
+tn_index = set() # thread number
+
+for line in lines:
+ if line[0][0] == '#':
+ continue
+ db_name = line[0]
+ db_index.add(db_name)
+
+ dir_name, workload = os.path.split(line[1])
+ workload = os.path.splitext(workload)[0]
+
+ num_threads = int(line[2])
+ tn_index.add(num_threads)
+
+ throughput = float(line[3])
+ if not results.has_key(workload):
+ results[workload] = { }
+ if not results[workload].has_key(db_name):
+ results[workload][db_name] = { }
+ if not results[workload][db_name].has_key(num_threads):
+ results[workload][db_name][num_threads] = [ ]
+ results[workload][db_name][num_threads].append(throughput)
+
+db_index = sorted(db_index)
+tn_index = sorted(tn_index)
+
+for wl in sorted(results.keys()):
+ out_file = open(os.path.join(dir_name, wl + ".result"), 'w+')
+ # Prints header
+ line = "#"
+ for db in db_index:
+ line += '\t' + db
+ out_file.write(line + '\n')
+ # Prints results
+ for tn in tn_index:
+ line = str(tn)
+ for db in db_index:
+ data = results[wl][db][tn]
+ line += '\t' + str(numpy.median(numpy.array(data)))
+ out_file.write(line + '\n')
+

run.sh

@@ -0,0 +1,33 @@
+#/bin/bash
+
+repeat_num=3
+db_names=(
+ "lock_stl"
+ "lock_slib"
+ "tbb_rand"
+ "tbb_scan"
+ "svm_stl"
+ "svm_slib"
+)
+
+trap 'kill $(jobs -p)' SIGINT
+
+if [ $# -ne 1 ]; then
+ echo "Usage: $0 [dir of workload specs]"
+ exit 1
+fi
+
+workload_dir=$1
+
+for file_name in $workload_dir/workload*.spec; do
+ for ((tn=1; tn<=8; tn=tn*2)); do
+ for db_name in ${db_names[@]}; do
+ for ((i=1; i<=repeat_num; ++i)); do
+ echo "Running $db_name with $tn threads for $file_name"
+ ./ycsbc -db $db_name -threads $tn -P $file_name 2>>ycsbc.output &
+ wait
+ done
+ done
+ done
+done
+

workloads/workloada renamed to workloads/workloada.spec

@@ -6,8 +6,8 @@
 # Default data size: 1 KB records (10 fields, 100 bytes each, plus key)
 # Request distribution: zipfian
 
-recordcount=1000
-operationcount=1000
+recordcount=100000
+operationcount=1000000
 workload=com.yahoo.ycsb.workloads.CoreWorkload
 
 readallfields=true

workloads/workloadb renamed to workloads/workloadb.spec

@@ -12,7 +12,7 @@
 # key, and then request a number of records; this works fine even for hashed insertion.
 
 recordcount=100000
-operationcount=10000
+operationcount=50000
 workload=com.yahoo.ycsb.workloads.CoreWorkload
 
 readallfields=true