Go memory allocation tracing

make_trace.sh

#!/bin/dash

ltrace -x runtime.mallocgc -f -ttt -o malloc.trace $@
python parse.py malloc.trace

parse.py

import fileinput
import re
import sys

total_mallocs = 0
cold_misses = 0
false_shares = 0

cache_line_size = 64
cache = {}

core_map = {}

def update(core, time, size, ptr):
	global total_mallocs
	global cold_misses
	global false_shares
	line = ptr / cache_line_size
	if line not in cache:
		cold_misses += 1
	elif cache[line] != core:
		false_shares += 1

	total_mallocs += 1

	core_map[core] = core
	cache[line] = core

regex = re.compile(r'([^ ]*) ([^ ]*) runtime.mallocgc\(([^,]*),.*= (.*)')
# Note: If not using ltrace -o filename use the following regex.
# regex = re.compile(r'\[pid ([^ ]*)\] ([^ ]*) runtime.mallocgc\(([^,]*),.*= (.*)')

for line in fileinput.input(sys.argv[1:]):
	m = regex.match(line.strip())
	if m is not None:
		g = m.groups()
		update(int(g[0]), float(g[1]), int(g[2]), int(g[3],16))

print "Malloc trace of %d mallocs over %d cores" % (total_mallocs, len(core_map))
print "\tCache line shared %d (%.2f%%) times" % (false_shares,
					       float(false_shares) /
					       float(total_mallocs) * 100)
print "\tCache line cold missed %d (%.2f%%) times" % (cold_misses,
					       float(cold_misses) /
					       float(total_mallocs) * 100)

results

# uiktest
Malloc trace of 105753 mallocs over 6 cores
	Cache line shared 6673 (6.31%) times
	Cache line cold missed 10885 (10.29%) times

# go install -v std
Malloc trace of 307308 mallocs over 2 cores
	Cache line shared 0 (0.00%) times
	Cache line cold missed 25260 (8.22%) times

# go install github.com/petar/GoLLRB/llrb
Malloc trace of 15059 mallocs over 5 cores
	Cache line shared 184 (1.22%) times
	Cache line cold missed 4684 (31.10%) times