X-Git-Url: http://git.vrable.net/?a=blobdiff_plain;f=parsetrace%2Fanalyze-tcp.py;h=83125e79a1e7c13643509fade6702f93b9270118;hb=9af190d3571bd1dbc6f58a20639863abb5b50fed;hp=47a3d7baebec85f05e72a7274b47b6fc2359d987;hpb=4a16bafc74f36e1314a722544612ad7ac6a6cd05;p=bluesky.git

diff --git a/parsetrace/analyze-tcp.py b/parsetrace/analyze-tcp.py
index 47a3d7b..83125e7 100755
--- a/parsetrace/analyze-tcp.py
+++ b/parsetrace/analyze-tcp.py
@@ -4,12 +4,15 @@
 # determine as much as possible about the performance of that connection.
 # (Specifically designed for measuring performance of fetches to Amazon S3.)
 
-import impacket, pcapy, re, sys
+import impacket, json, pcapy, re, sys
 import impacket.ImpactDecoder, impacket.ImpactPacket
 
 # Estimate of the network RTT
 RTT_EST = 0.03
 
+def dump_data(obj):
+    return json.dumps(result_list, sort_keys=True, indent=2)
+
 class Packet:
     def __init__(self, connection, ts, pkt):
         self.connection = connection
@@ -90,16 +93,23 @@ def split_trace(packets, predicate, before=True):
 
 def analyze_get(packets):
     packets = iter(packets)
-
-    # First packet is the GET request itself
     p = packets.next()
-    if not(p.direction > 0 and p.data.startswith('GET')):
-        print "Doesn't seem to be a GET request..."
-        return
 
     start_ts = p.ts
     id_out = p.id
 
+    # Check for connection establishment (SYN/SYN-ACK) and use that to estimate
+    # th network RTT.
+    if p.tcp.get_SYN():
+        p = packets.next()
+        #print "Connection establishment: RTT is", p.ts - start_ts
+        return {'syn_rtt': p.ts - start_ts}
+
+    # Otherwise, we expect the first packet to be the GET request itself
+    if not(p.direction > 0 and p.data.startswith('GET')):
+        #print "Doesn't seem to be a GET request..."
+        return
+
     # Find the first response packet containing data
     while not(p.direction < 0 and p.datalen > 0):
         p = packets.next()
@@ -107,30 +117,48 @@ def analyze_get(packets):
     resp_ts = p.ts
     id_in = p.id
     start_seq = p.seq[0]
+    tot_bytes = (p.seq[1] - start_seq) & 0xffffffff
+    spacings = []
 
-    print "Response time:", resp_ts - start_ts
+    #print "Response time:", resp_ts - start_ts
 
     # Scan through the incoming packets, looking for gaps in either the IP ID
     # field or in the timing
     last_ts = resp_ts
     for p in packets:
         gap = False
+        bytenr = (p.seq[1] - start_seq) & 0xffffffff
         if not p.direction < 0: continue
+        if p.tcp.get_FIN(): continue
+        spacings.append((p.ts - last_ts, bytenr))
         if p.id != (id_in + 1) & 0xffff:
             gap = True
-            print "Sequence number gap at", id_in
+            #print "Sequence number gap at", id_in
         if p.ts - last_ts > 2 * RTT_EST:
             gap = True
-            print "Long gap of", p.ts - last_ts
+            #print "Long gap of", p.ts - last_ts
         elif p.ts - last_ts > RTT_EST / 2:
             gap = True
-            print "Short gap of", p.ts - last_ts
+            #print "Short gap of", p.ts - last_ts
         if gap:
-            print "    [occurred after", p.seq[0] - start_seq, "bytes, time", p.ts, "sec]"
+            #print "    [occurred after", p.seq[0] - start_seq, "bytes, time", p.ts, "sec]"
+            pass
         if p.datalen not in (1448, 1460):
-            print "Short packet of", p.datalen, "bytes, brings total to", p.seq[1] - start_seq
+            #print "Short packet of", p.datalen, "bytes, brings total to", p.seq[1] - start_seq
+            pass
+        if (p.seq[0] - start_seq) & 0xffffffff != tot_bytes:
+            #print "Packet out of order; got sequence number %d, expected %d" \
+            #        % ((p.seq[0] - start_seq) & 0xffffffff, tot_bytes)
+            pass
         last_ts = p.ts
         id_in = p.id
+        tot_bytes = max(tot_bytes, bytenr)
+
+    #print "Transferred %d bytes in %s seconds, initial response after %s" % (tot_bytes, last_ts - start_ts, resp_ts - start_ts)
+    return {'bytes': tot_bytes,
+            'start_latency': resp_ts - start_ts,
+            'finish_latency': last_ts - start_ts,
+            'interpacket_times': spacings}
 
 if __name__ == '__main__':
     for f in sys.argv[1:]:
@@ -139,17 +167,22 @@ if __name__ == '__main__':
         ts = 0.0
         def request_start(p):
             return p.direction > 0 and p.datalen > 0
+        result_list = []
         for s in split_trace(conn.packets, request_start):
             s = list(s)
             if False:
                 for p in s:
-                    if p.ts - ts > 0.01:
-                        print "----"
-                    if p.ts - ts > 2 * RTT_EST:
-                        print "LONG DELAY\n----"
+                    #if p.ts - ts > 0.01:
+                        #print "----"
+                    #if p.ts - ts > 2 * RTT_EST:
+                        #print "LONG DELAY\n----"
                     ts = p.ts
-                    print p
-                    if p.direction > 0 and p.datalen > 0:
-                        print "Request:", repr(p.data)
-            analyze_get(s)
-            print "===="
+                    #print p
+                    #if p.direction > 0 and p.datalen > 0:
+                        #print "Request:", repr(p.data)
+            results = analyze_get(s)
+            if results is not None:
+                result_list.append(results)
+            #print "===="
+
+        print dump_data(result_list)