Add proper per-file copyright notices/licenses and top-level license.

[bluesky.git] / results / parse-sfsres.py

diff --git a/results/parse-sfsres.py b/results/parse-sfsres.py
index 96de28a..8d84ac1 100755 (executable)
--- a/results/parse-sfsres.py
+++ b/results/parse-sfsres.py
@@ -4,7 +4,7 @@
 # Parse the sfsres log file generated by SPECsfs to generate more detailed
 # latency statistics than in the sfssum summary file.
 
-import re, sys
+import re, subprocess, sys
 
 def extract_re(lines, regexp):
     if isinstance(regexp, str):
@@ -13,11 +13,39 @@ def extract_re(lines, regexp):
         m = regexp.match(l)
         if m: return m
 
-def parse_run(lines, timestamp, outfp=sys.stdout):
+OPERATIONS = ('read', 'write', 'create', 'setattr', 'lookup', 'getattr')
+STATSDATA = ('getop', 'getbyte', 'putop', 'putbyte', 'nfsincount', 'nfsinbyte', 'nfsoutcount', 'nfsoutbyte')
+COSTS = (0.01e-4, 0.15/1024**3, 0.01e-3, 0.10/1024**3, 0, 0, 0, 0)
+
+op_sum = 0
+stat_sum = [0 for _ in STATSDATA]
+
+def parse_date(datestr):
+    p = subprocess.Popen(['/bin/date', '-d', datestr, '+%s'],
+                         stdout=subprocess.PIPE)
+    d = p.stdout.read()
+    p.wait()
+    return int(d.strip())
+
+def find_stats(statsdata, timestamp):
+    if statsdata is None: return (0, [0] * len(STATSDATA))
+    for s in statsdata:
+        if s[0] > timestamp: return (s[0], s[1:])
+    return (statsdata[-1][0], statsdata[-1][1:])
+
+def parse_run(lines, timestamp, outfp=sys.stdout, statsdata=[]):
+    global stat_sum, op_sum
+
     #print timestamp
     requested_load = extract_re(lines, r"\s*Requested Load.*= (\d+)")
     load = int(requested_load.group(1))
     results = extract_re(lines, r"SFS NFS THROUGHPUT:\s*([\d.]+).*RESPONSE TIME:\s*([\d.]+) Msec/Op")
+    timestamp = extract_re(lines, r"SFS Aggregate Results.*, (.*)")
+    if timestamp is not None:
+        try:
+            timestamp = parse_date(timestamp.group(1))
+        except:
+            timestamp = None
 
     # Extract the stable of per-operation counts, response times, etc.
     regexp = re.compile(r"^(\w+)" + r"\s*([\d.]+)%?" * 9)
@@ -25,27 +53,75 @@ def parse_run(lines, timestamp, outfp=sys.stdout):
     for l in lines:
         m = regexp.match(l)
         if m:
-            table[m.group(1)] = [float(m.group(i)) for i in range(2, 11)]
+            try:
+                table[m.group(1)] = [float(m.group(i)) for i in range(2, 11)]
+            except:
+                #sys.stderr.write("Error parsing line: " + l.strip() + "\n")
+                pass
 
+    # Search for statistics on uploads/downloads in the time interval when the
+    # benchmark is running.  We have the ending time.  SPECsfs runs for the 10
+    # minutes prior, and uses the last 5 minuts of data.  Let's use the time
+    # from 6 minutes prior to 1 minute prior, to give another 5-minute period
+    # with a bit of a buffer after in case timing is slightly off.
+    (t1, s1) = find_stats(statsdata, timestamp - 6*60)
+    (t2, s2) = find_stats(statsdata, timestamp - 1*60)
+    stat_delta = map(lambda x, y: y - x, s1, s2)
+
+    outfp.write("# finish_timestamp: " + str(timestamp) + "\n")
+    outfp.write("# in %s seconds: stats are %s\n" % (t2 - t1, stat_delta))
     outfp.write("%d\t%s\t%s" % (load, results.group(1), results.group(2)))
-    for o in ('read', 'write', 'getattr'):
-        outfp.write("\t%s\t%s" % (table[o][5], table[o][6]))
+    for o in OPERATIONS:
+        val = '-'
+        try: val = table[o][5]
+        except: pass
+        outfp.write("\t%s" % (val,))
     outfp.write("\n")
 
-def parse_sfsres(fp):
+    op_sum += int(results.group(1))
+    stat_sum = map(lambda x, y: x + y, stat_sum, stat_delta)
+
+def parse_sfsres(fp, statsdata):
+    sys.stdout.write("# target_ops actual_ops latency_avg")
+    for o in OPERATIONS:
+        sys.stdout.write(" " + o)
+    sys.stdout.write("\n")
     timestamp = None
     run_data = []
     for line in fp:
         m = re.match(r"^([^*]+) \*{32,}$", line)
         if m:
             if len(run_data) > 0:
-                parse_run(run_data, timestamp)
+                parse_run(run_data, timestamp, statsdata=statsdata)
             run_data = []
             timestamp = m.group(1)
         else:
             run_data.append(line)
     if len(run_data) > 0:
-        parse_run(run_data, timestamp)
+        parse_run(run_data, timestamp, statsdata=statsdata)
+
+    print
+    print "Total SFS operations:", op_sum * 300
+    print "Statistics:"
+    cost = 0.0
+    for i in range(len(STATSDATA)):
+        cost += stat_sum[i] * COSTS[i]
+        print "%s: %s (%s)" % (STATSDATA[i], stat_sum[i], stat_sum[i] / (op_sum * 300.0))
+    print "Total cost: %s (%s per op)" % (cost, cost / (op_sum * 300.0))
+
+def parse_stats(statsfile):
+    datapoints = []
+    for line in statsfile:
+        if re.match(r"^#", line): continue
+        datapoints.append([float(x) for x in line.split()])
+    return datapoints
 
 if __name__ == '__main__':
-    parse_sfsres(sys.stdin)
+    input_sfsres = open(sys.argv[1])
+    try:
+        input_stats = open(sys.argv[2])
+        statsdata = parse_stats(input_stats)
+    except:
+        statsdata = None
+
+    parse_sfsres(input_sfsres, statsdata)