Transcript IPPM Metric Test

Testing Standards Track Metrics
Draft-ietf-ippm-metrictest-00
Geib, Morton, Fardid, Steinmitz
IETF-78
July 2010
Geib/Morton/Fardid/Steinmitz / draft-ietf-metrictest-00
July 2010
1
Draft-ietf-ippm-metrictest-00
Tunnels: Criteria and a tested candidate
must map packets of different interfaces onto the same tunnel,
carrying an identical outer IP address/port for all packets.
 easily accessible/commodity implementation
 low operational overhead
 reliable
 low cost

Candidates after practical experience seem to be
(ordered by best match to the above criteria):
 Ethernet over L2TPv3 using port mode (RFC 4719)
 Ethernet PW using port mode
Geib/Morton/Fardid/Steinmitz / draft-ietf-metrictest-00
July 2010
2
Draft-ietf-ippm-metrictest-00.
Networking conditions and evaluation.
 Metric
implementations will be operated in real networks. Metric
compliance should be tested under live network conditions too, if possible.
 Avoid


accidental results (more positive: ensure reliable results).
Work with n singletons per evaluation interval and repeat test n times.
A rule of thumb says n = 5 or more.
 Example
C++ code for an Anderson Darling 2 sample test is included now.
Geib/Morton/Fardid/Steinmitz / draft-ietf-metrictest-00
July 2010
3
Draft-ietf-ippm-metrictest-00.
Next steps.
 Is
there community interest in setting up a test of the
proposed metric test?
 Clarify
which (parts of) metric specifications preferrably are
tested by using the Internet as “Network under Test” and
which should be tested under laboratory conditions.
Geib/Morton/Fardid/Steinmitz / draft-ietf-metrictest-00
July 2010
4
Draft-ietf-ippm-metrictest-00.
Experimental set up and experiences (1).
Metric test set up verified by a diploma thesis




Realised Ethernet over L2TPv3 tunnel in port mode and utilise Ethertype 802.1Q (0x8100)
transport to avoid local routing.
Execute Anderson Darling K-sample test for two instances of one implementation.
5 parallel measurement streams per implementation, each terminating on a different port.
One metric implementation had no GPS synch and there was undefined local load.
Conditions close to metric test specification, but not fully meeting these.
Metric Implementation 1a
CPE Routers:
Ethernet over L2TPv3
Tunnel Head,
configure port mode
10.100.64.42/24
EtherType 0x8100
VLAN ID 101
F-Eth
F-Eth
F-Eth
Metric Implementation 1b
10.100.64.32/24
EtherType 0x8100
VLAN ID 100
CPE
Internet
Router
CPE
F-Eth
CPE
Forward IP Destination
10.100.64.42/24 to SubIF
EtherType 0x8100
VLAN ID101
Forward IP Destination
10.100.64.32/24 to SubIF
EtherType 0x8100
VLAN ID100
Routing: Static
Geib/Morton/Fardid/Steinmitz / draft-ietf-metrictest-00
July 2010
5
Draft-ietf-ippm-metrictest-00.
Experimental set up and experiences (2).
„Repeating“ measurements under identical network conditions with a single implementation by measuring
with two identical implementations (five flows by using separate IP ports).
Metric
Implement.
A1
Metric
Implement.
A2
Tunnel
termination
1
Internet
Tunnel
termination
2
Ethernet over L2TPv3 / port mode Tunnel Structure
Measurement IP adresses
and ports vary. Avoid local
routing by tunneling a layer
2 port over the Internet.
Ports
IP Addresses
IEEE 802.1Q VLAN IDs vary
(2 at least)  port mode avoids
single tunnels per VLAN ID.
L2TPv3 IP: same origin and destiantion
IP address and ports for all VLAN IDs 
one path through the IP network.
VLAN ID and MAC addresses
Ports
Geib/Morton/Fardid/Steinmitz / draft-ietf-metrictest-00
IP Addresses
July 2010
6
Draft-ietf-ippm-metrictest-00.
Experimental set up and experiences (3).
Some new results
80
GPSsyn_port1
60
GPSsyn_port2
GPSsyn_port3
40
GPSsyn_port4
20
GPSsyn_port5
an
d
-20
22
gr 4
ea
te
r
20
5
18
6
16
8
93
11
2
13
0
14
9
75
56
0
37

100
19

Frequency per port
0

We’ve tested single implementation
precision by 5 parallel flows.
We’ve tested single implementation
Round trip time metric and required
adjustments of less than 20 s to pass
AD2 sample tests (see figure).
One way delay measurements of single
implementations passed ADK test for
both implememntations without any
adjustment (with one exception).
Comparing the results of the two
implementations wasn’t successful
(broken GPS in the one system
probably is one of the causes).
Frequency

Normalised Round Trip Delay [us]
Geib/Morton/Fardid/Steinmitz / draft-ietf-metrictest-00
July 2010
7