diff --git a/var_strings.py b/var_strings.py
new file mode 100644
index 0000000..b8f8b9b
--- /dev/null
+++ b/var_strings.py
@@ -0,0 +1,111 @@
+import calendar
+import datetime
+import json
+import struct
+
+event = {"event_type": "nova.compute.run_instance.start",
+         "generated": datetime.datetime.utcnow(),
+         "request_id": "req-1234abcd5678efgh",
+         "source": "n-compute-1973",
+         "payload": {
+            "foo": 123,
+            "blah": "abc",
+            "zoo": False
+         }
+        }
+
+
+def handle_datetime(obj):
+    if isinstance(obj, datetime.datetime):
+        if obj.utcoffset() is not None:
+            obj = obj - obj.utcoffset()
+    millis = int(
+        calendar.timegm(obj.timetuple()) * 1000 +
+        obj.microsecond / 1000
+    )
+    return millis
+
+
+json_event = json.dumps(event, default=handle_datetime)
+metadata = {'request_id': event['request_id'],
+            'event_type': event['event_type'],
+            'source': event['source'],
+           }
+
+CURRENT_VERSION = 1
+
+# Version 1 SCHEMA
+# ----------------
+# i = 0x69867884  (EVNT)
+# h = version
+# i = metadata block length
+# i = raw notification block length
+# i = 0x00000000 EOR
+
+# Metadata block
+# i = number of strings
+# [i, i] = length of key, length of value
+# [*s, *s] = key, value
+
+# Raw notification block
+# *s = raw data
+
+raw_block_schema = "%ds" % len(json_event)
+raw_block = struct.pack(raw_block_schema, json_event)
+
+metadata_items = ["i"] # appended with N "%ds"'s
+metadata_values = [len(metadata) * 4]  # [n]=key, [n+1]=value
+for key, value in metadata.iteritems():
+    metadata_items.append("i")
+    metadata_items.append("i")
+    metadata_values.append(len(key))
+    metadata_values.append(len(value))
+
+for key, value in metadata.iteritems():
+    metadata_items.append("%ds" % len(key))
+    metadata_values.append(key)
+    metadata_items.append("%ds" % len(value))
+    metadata_values.append(value)
+metadata_schema = "".join(metadata_items)
+
+metadata = struct.pack(metadata_schema, *metadata_values)
+
+header_schema = "ihiii"
+header_size = struct.calcsize(header_schema)
+header = struct.pack(header_schema, 0x69867884, CURRENT_VERSION,
+                     struct.calcsize(metadata_schema),
+                     struct.calcsize(raw_block_schema), 0)
+
+with open("test.dat", "wb") as f:
+    f.write(header)
+    f.write(metadata)
+    f.write(raw_block)
+
+with open("test.dat", "rb") as f:
+    header = f.read(header_size)
+    header = struct.unpack(header_schema, header)
+    print "Marker:", hex(header[0])
+    print "Version:", header[1]
+    print "Metadata length:", header[2]
+    print "Raw notification:", header[3]
+    print "EOR:", header[4]
+
+    metadata = f.read(header[2])
+    num_strings = struct.unpack_from("i", metadata)
+    print "Num metadata strings:", num_strings
+    offset = struct.calcsize("i")
+    lengths = num_strings[0] / 2
+    lengths_schema = "i" * lengths
+    print "Lengths schema:", lengths_schema
+    key_value_sizes = struct.unpack_from(lengths_schema, metadata,
+                                         offset=offset)
+    key_value_schema_list = ["%ds" % sz for sz in key_value_sizes]
+    key_value_schema = "".join(key_value_schema_list)
+    offset += struct.calcsize(lengths_schema)
+    key_values = struct.unpack_from(key_value_schema, metadata,
+                                    offset=offset)
+    metadata_dict = dict((key_values[n], key_values[n+1])
+                         for n in range(len(key_values))[::2])
+
+    print "Key Value schema:", key_value_schema
+    print "Metadata:", metadata_dict