1. Measurement of 802.11 Roaming Intervals
January 2004
doc.: IEEE /xxxx
January 2004
Measurement of Roaming Intervals
Darwin Engwer
Nortel Networks
4655 Great America Pkwy
Santa Clara, CA 95054
Phone:
Fax:
Darwin Engwer, Nortel Networks
Darwin Engwer, Nortel Networks

2. January 2004
Introduction
This presentation uses animation. The reader is encouraged to view the presentation in a way that makes the animation visible (Slide Show mode).
Darwin Engwer, Nortel Networks

3. Goals of this presentation
January 2004
Goals of this presentation
define the components of an system
define roaming in an system
identify an initial set of roaming conditions for consideration/ analysis
identify the start and end points of the roaming interval
present test setups for effective roaming interval measurements
stimulate thought and discussion on handover topics ...
Darwin Engwer, Nortel Networks

4. Scope of this presentation
January 2004
Scope of this presentation
Considers existing standards:
1999 base, a, b, d, f, g and h
Does not specifically address pending standards:
802.11e, i, j and n
aspects of roaming not addressed:
criteria for triggering the roaming event
Darwin Engwer, Nortel Networks

5. Components of an 802.11 System
January 2004
Components of an System
Std defines a station (STA)
Std further defines a subset of STAs called Access Points (APs)
“STA acting as an AP”
for clarity as to the type of stations referenced herein, I further define a “STA that is NOT acting as an AP” as a Mobile Unit (MU).
Darwin Engwer, Nortel Networks

6. 802.11 System Layout AP#1 AP#2 MU January 2004 Uplink to
Infrastructure
Network
AP#1
SSID= “ACME”
AP#2
SSID= “ACME” MU
Darwin Engwer, Nortel Networks

7. January 2004
Premises
MU can only be associated with one AP at any given point in time (per clause number )
Darwin Engwer, Nortel Networks

8. January 2004
Roaming Definition
Roaming occurs when an MU changes it’s association from one AP to another within the same ESS:
i.e. the SSIDs of the two APs are identical
this is called a “BSS-transition” per clause b
uses the reassociation service per clause
Changing to an AP with a different SSID represents a change to a different network. This is a meta-case and is not discussed in this presentation. (“ESS-transition” per c)
Darwin Engwer, Nortel Networks

9. Roaming in an Ideal Network
January 2004
Roaming in an Ideal Network
Uplink to
Infrastructure
Network
AP#1
SSID= “ACME”
AP#2
SSID= “ACME” MU MU MU
Darwin Engwer, Nortel Networks

10. Core Roaming Sequence 1. MU is communicating with AP#1
January 2004
Core Roaming Sequence
1. MU is communicating with AP#1
2. MU moves to AP#2 coverage area
3. MU stops communicating with AP#1
4. MU starts communicating with AP#2
goal = measure time between events 3 and 4
Darwin Engwer, Nortel Networks

11. But, it’s more complicated than that …
January 2004
But, it’s more complicated than that …
networks are made of real components that can affect the outcome
Darwin Engwer, Nortel Networks

12. Roaming in a HUB Network
January 2004
Roaming in a HUB Network
Uplink to
Infrastructure
Network
HUB
AP#1
SSID= “ACME”
AP#2
SSID= “ACME” MU MU MU
Darwin Engwer, Nortel Networks

13. January 2004
How does roaming work …
consider the Message Sequence Chart (MSC) for roaming in a simple, hub-based network …
Darwin Engwer, Nortel Networks

14. MSC: Establish Association with AP#1
January 2004
MSC: Establish Association with AP#1 MU
AP1
AP2
Uplink
SCAN PHASE
JOIN PHASE
Auth Request
Ack
Auth Reponse
Association Request
Ack
Association Reponse
DATA
Ack
Darwin Engwer, Nortel Networks

15. MSC: Roam to AP#2 MU AP1 AP2 Uplink January 2004 DATA SCAN PHASE
Roaming event
Ack
SCAN PHASE
JOIN PHASE
Auth Request
Ack
Auth Reponse
Re-Association Request
Ack
Re-Association Reponse
802.11f MOVE-Notify
802.11f MOVE-Response t1
Ack
DATA
Darwin Engwer, Nortel Networks

16. Important parts of the core roaming sequence:
January 2004
Important parts of the core roaming sequence:
discovery of candidate APs
roaming event triggered
reassociate request to AP#2
handover from AP#1 to AP#2 [IAPP]
was proprietary; now f defines an interoperable mechanism
ends the MU’s association with AP#1 - avoids “dangling association”
this helps AP#1differentiate between an MU that has wandered out of range - AP#1 still tries to communicate with the MU - and an MU that has roamed to another AP - AP#1 does NOT try to communicate with the MU).
AP#2 sends reassociate response
Darwin Engwer, Nortel Networks

17. Reassociate Frame recap
January 2004
Reassociate Frame recap
contains:
Capability Field
Listen Interval
Current [old] AP [MAC] address
SSID
Supported Rates
Darwin Engwer, Nortel Networks

18. there are some alternatives
January 2004
there are some alternatives
these are “technically” allowed by the standard, but not recommended …
instead of reassociating with AP#2, the MU could perform a fresh association with AP#2
in some cases this may be needed - such as when the MU has been completely out of range of the entire ESS for an extended period of time
strongly discouraged since fresh association does not properly support mobility (per clause 5.4.2)
Darwin Engwer, Nortel Networks

19. January 2004
… alternatives
prior to re-associating with AP#2 the MU could send a disassociate request to AP#1
allowed by the standard; intended to be used when changing ESS
physical aspects can make this approach troublesome when used within an ESS
MU is trying to communicate over a known to be poor link (to AP#1)
MU and/or AP#1 can get stuck retrying part of the disassociate frame exchange sequence
OK for SSID changes
unreliable and error-prone in real networks
hence, not recommended within an ESS
Darwin Engwer, Nortel Networks

20. January 2004
… alternatives
there are other techniques that are likely legal per the standard
but, for fast roaming I recommend that focus be placed on the normal, preferred case: MU reassociates with AP#2
Darwin Engwer, Nortel Networks

21. Recap - Roaming in a simple HUB Network
January 2004
Recap - Roaming in a simple HUB Network
Uplink to
Infrastructure
Network
HUB
AP#1
SSID= “ACME”
AP#2
SSID= “ACME” MU MU MU
Darwin Engwer, Nortel Networks

22. But, it’s more complicated than that …
January 2004
But, it’s more complicated than that …
a hub network doesn’t account for interceding delays or switching in a real network
Darwin Engwer, Nortel Networks

23. Roaming in a Switched Network
January 2004
Roaming in a Switched Network
Uplink to
Infrastructure
Network
Data path
Data path
SWITCH P3 P1 P2
AP#1
SSID= “ACME”
AP#2
SSID= “ACME” MU MU MU
Darwin Engwer, Nortel Networks

24. Roaming in a Switched Network
January 2004
Roaming in a Switched Network
need to consider switching delays
and, learning of MU MAC address-to-port mapping
must be addressed in the absence of uplink traffic from the MU
even when there is uplink traffic, there may be delays (e.g until the next uplink packet is sent)
solution: handled by the f Layer 2 Update (XID) packet
this packet is addressed at layer 2 as if it originated from the MU
sent just after an f ADD-Notify packet and after receipt of an f MOVE-response packet.
Darwin Engwer, Nortel Networks

25. MSC: Associating with AP#1
January 2004
MSC: Associating with AP#1 MU
AP1
Switch
AP2
Uplink
SCAN PHASE
JOIN PHASE
Auth Request
Ack
Auth Reponse
Association Request
Ack
Association Reponse
802.11f ADD-Notify
802.11f L2 Update
DATA
Ack
Darwin Engwer, Nortel Networks

26. MSC: Roaming in a Switched Network
January 2004
MSC: Roaming in a Switched Network MU
AP1
Switch
AP2
Uplink
DATA
Ack
Roaming event
SCAN PHASE
JOIN PHASE
Auth Request
Ack
Auth Reponse
Re-Association Request
Ack
802.11f MOVE-Notify
802.11f MOVE-Notify
802.11f MOVE-Resp
802.11f MOVE-Resp
802.11f L2 Update
Re-Association Response
Ack
DATA
Ack
Darwin Engwer, Nortel Networks

27. Beginning of the Roaming Interval
January 2004
Beginning of the Roaming Interval
Defined: The last point in time when all network components know and agree upon the link path [to the MU].
The relevant components are:
the MU
the AP (AP#1)
the infrastructure network (e.g. layer 2 switch)
...
Darwin Engwer, Nortel Networks

28. Beginning of the Roaming Interval
January 2004
Beginning of the Roaming Interval
Is it:
the start of the Scan process? (i.e. MU sends probe request)
No, MU could have scanned in advance
No, MU could be doing passive scanning
the start of the Join process?
No, does not include any lost service due to possible scanning
the reassociation request?
No, again may not include any lost service due to possible scanning
From the MU’s perspective the last data frame received from AP#1 marks a definitive point in time when service via AP#1 was known to be good.
Darwin Engwer, Nortel Networks

29. End of the Roaming Interval
January 2004
End of the Roaming Interval
Defined: The point in time when all network components know and agree upon the new link path [to the MU].
The relevant components are:
the MU
the old AP (AP#1)
the new AP (AP#2)
the infrastructure network (e.g. layer 2 switch)
From the MU’s perspective the first data frame received from AP#2 marks a definitive point in time when service via AP#2 is known to be good.
Darwin Engwer, Nortel Networks

30. January 2004
Roaming Interval
Defined: the end of service from AP#1 and the start of service from AP#2
Beginning: last data frame successfully received from AP#1
End: first data frame successfully received from AP#2
Darwin Engwer, Nortel Networks

31. Determining the Roaming Interval
January 2004
Determining the Roaming Interval MU
AP1
Switch
AP2
Uplink
DATA
Ack
Roaming event
SCAN PHASE
JOIN PHASE
Auth Request
Ack
Auth Reponse
Roaming
Interval
Re-Association Request
Ack
802.11f MOVE-Notify
802.11f MOVE-Notify
802.11f MOVE-Resp
802.11f MOVE-Resp
802.11f L2 Update
Re-Association Response
Ack
DATA
Ack
Darwin Engwer, Nortel Networks

32. Determining the Roaming Interval
January 2004
Determining the Roaming Interval
the specifics of what happens during the roaming interval will vary from implementation to implementation
under some conditions additional packet exchanges may be required (e.g )
but, it is enough to just say that roaming happens, and the path of the data service changes from AP#1 to AP#2 and the time required to do all that is the Roaming Interval value in which we are interested ...
Darwin Engwer, Nortel Networks

33. Determining the Roaming Interval
January 2004
Determining the Roaming Interval MU
AP1
Switch
AP2
Uplink
DATA
Ack
ROAMING
SCAN PHASE
JOIN PHASE
Auth Request
Ack
Auth Reponse
Roaming
Interval
Re-Association Request
Ack
802.11f MOVE-Notify
802.11f MOVE-Notify
802.11f L2 Update
802.11f MOVE-Resp
802.11f MOVE-Resp
Re-Association Response
Ack
DATA
Ack
Darwin Engwer, Nortel Networks

34. Measuring the Roaming Interval
January 2004
Measuring the Roaming Interval
Now that the extent of the roaming interval is known, how can a test setup be configured in order to measure it?
Darwin Engwer, Nortel Networks

35. Measurement by Powering Down AP#1
January 2004
Measurement by Powering Down AP#1
Switch P1 P2 P3
AP#1
SSID= “ACME”
AP#2
SSID= “ACME” MU
Sniffer
Darwin Engwer, Nortel Networks

36. Powering Down AP#1 - Test Sequence
January 2004
Powering Down AP#1 - Test Sequence
power up sniffer
power up AP#1
power up the MU
wait until the MU establishes association with AP#1
power up AP#2
wait until the MU detects AP#2 (client site survey software?)
start sniffer capture
power down AP#1
wait until MU reassociates with AP#2 and receives data from AP#2
stop sniffer capture
save sniffer trace
Darwin Engwer, Nortel Networks

37. Issues with Powering Down AP#1
January 2004
Issues with Powering Down AP#1
not representative of the real world:
not representative of real world RF conditions at roaming event
NOT the more usual roaming AP-to-AP with some amount of coverage overlap
effects (and capabilities) of the interceding network are not considered
no interceding delays or switching effects in the network
deals with AP failure (or completely out of range) case rather than real seamless AP-to-AP roaming scenario
MU is not comparing 2 candidate APs and choosing one
it is just choosing the only AP available
power down of AP#1 means the MU state machine is dealing with the case of a poor connection due to a complete end to the stream of beacons from AP#1
APs cannot communicate with each other during roaming event
Darwin Engwer, Nortel Networks

38. Analysis of Power Down AP#1 Measurement Setup
January 2004
Analysis of Power Down AP#1 Measurement Setup
roaming time measurement: power down setup is inadequate bcus:
1. not representative of real world RF conditions at roaming event
2. not representative of infrastructure conditions at roaming event
Desired setup is for the signal from AP#1 to fade rather than stop, so that at the roaming event AP#1 is still powered up and connected to the infrastructure network.
Also, need to ensure reassociate vs. fresh association
Darwin Engwer, Nortel Networks

39. Measurement Using a “Cone of Silence”
January 2004
Measurement Using a “Cone of Silence”
Switch P1 P2 P3
AP#1
SSID= “ACME”
AP#2
SSID= “ACME” MU
Sniffer
Darwin Engwer, Nortel Networks

40. “Cone of Silence” - Test Sequence
January 2004
“Cone of Silence” - Test Sequence
power up sniffer
power up AP#1
power up the MU
wait until the MU establishes association with AP#1
power up AP#2
wait until the MU detects AP#2 (client site survey software?)
start sniffer capture
place “Cone of Silence” over AP#1
wait until MU reassociates with AP#2 and receives data from AP#2
stop sniffer capture
save sniffer trace
Darwin Engwer, Nortel Networks

41. “Cone of Silence” Issues
January 2004
“Cone of Silence” Issues
better but not as definitive as we would like
Murphy’s Law: when we need APs to be poor they are amazingly good
Darwin Engwer, Nortel Networks

42. Measurement Using an RF Attenuator on AP#1
January 2004
Measurement Using an RF Attenuator on AP#1
Switch P1 P2 P3
Note: APs are on different channels to ensure worst case roaming interval.
AP#2
SSID= “ACME”
CH= 11
AP#1
SSID= “ACME”
CH= 1
0-30 db
3 db
Sniffer MU
Darwin Engwer, Nortel Networks

43. RF Attenuator - Test Sequence
January 2004
RF Attenuator - Test Sequence
power up sniffer
power up AP#1; set attenuator to minimum attenuation
power up the MU
wait until the MU establishes association with AP#1
power up AP#2
wait until the MU detects AP#2 (client site survey software?)
start sniffer capture
increase attenuation on AP#1 radio/ antenna(s)
wait until MU reassociates with AP#2 and receives data from AP#2
stop sniffer capture
save sniffer trace
Darwin Engwer, Nortel Networks

44. Attenuator Test Setup Notes
January 2004
Attenuator Test Setup Notes
Attenuator test setup provides a convenient, reproducible, definitive way of creating real world conditions that force a roaming event
can be done using external antennas on the AP with an adjustable RF signal attenuator inline between the two
ensure that the attenuator is of a continuously variable type, NOT a discrete step type
note that the rate at which the attenuator is adjusted can affect the measurement result (therefore, must define and record the adjustment rate)
could also be done using Tx power adjustment on AP#1
ensure that the RF output level can be adjusted low enough to force the roaming event
may require a fixed attenuator combined with the Tx power adjustment
Darwin Engwer, Nortel Networks

45. Recap - limits of the Roaming Interval
January 2004
Recap - limits of the Roaming Interval MU
AP1
Switch
AP2
Uplink
DATA
Ack
ROAMING
SCAN PHASE
JOIN PHASE
Auth Request
Ack
Auth Reponse
Roaming
Interval
Re-Association Request
Ack
802.11f MOVE-Notify
802.11f MOVE-Notify
802.11f MOVE-Resp
802.11f MOVE-Resp
802.11f L2 Update
Re-Association Response
Ack
DATA
Ack
Darwin Engwer, Nortel Networks

46. Example Measurement Sniffer Trace
January 2004
Example Measurement Sniffer Trace
MU leaves channel to authenticate with AP#2
MU leaves channel to reassociate with AP#2
First packet from AP#2
ACK to the last data frame exchanged with AP#1
30 ms
Sniffer did not capture the ACKs from the MU, but ACKs were sent.
Note: In this example the two APs are on the same channel, which is not the worst case scenario.
Darwin Engwer, Nortel Networks

47. Measuring the Handover Interval
January 2004
Measuring the Handover Interval
Now know when to start and end measuring
Now know the test setup
How do we actually make the measurement?
Need a method to detect and determine the end of service from AP#1 and the start of service from AP#2
...
Darwin Engwer, Nortel Networks

48. Measuring the Handover Interval
January 2004
Measuring the Handover Interval
connect a packet generator (e.g. Chariot) to the switch
set the packet generator to send out [short] packets on a continuous basis
Packet type = directed L2 packets with a destination address equal to the MU’s MAC address
set a fixed interval between packets
include a sequence number in every packet
downlink packet generator ensures worst case scenario for roaming interval
i.e. showcases the worst case effects through the switch
Darwin Engwer, Nortel Networks

49. Measurement Using a Packet Generator
January 2004
Measurement Using a Packet Generator
Traffic
Generator
Data path
Data path P3 P1
Switch P2
Note: APs are on different channels to ensure worst case roaming interval.
AP#2
SSID= “ACME”
CH= 11
AP#1
SSID= “ACME”
CH= 1
0-30 db
3 db
Sniffer MU
Darwin Engwer, Nortel Networks

50. Measuring the Handover Interval
January 2004
Measuring the Handover Interval
Assume generator sends packets every ‘n’ msec, then
by measuring the Roaming Interval (as defined earlier),
time from last packet thru AP#1 to first packet thru AP#2
we can determine the Roaming Interval with an accuracy of +/- ‘n’ msec.
Or, restated, an absolute accuracy of 2x ‘n’ msec.
(per Nyquist’s theory)
The sequence numbers in the packets will also indicate if any packets (and how many) were lost during the roaming interval.
Both the roaming interval and the number of packets lost are critical pieces of information needed to evaluate the effect of certain applications (e.g. VOIP).
Darwin Engwer, Nortel Networks

51. Summary defined the components of an 802.11 system
January 2004
Summary
defined the components of an system
defined core roaming in an system
identified an initial set of roaming conditions for consideration/ analysis
identified the start and end points of the roaming interval
end of AP#1 data service, beginning of AP#2 data service
presented test setups for effective roaming interval measurements
Best = 1 MU, 2 APs, switch, packet generator, RF attenuator and sniffer
Darwin Engwer, Nortel Networks

52. Other Topics to Be Explored
January 2004
Other Topics to Be Explored
other cases to possibly consider:
inter-subnet layer 3 handover measurement - [need to describe setup too]
handover from one PHY type to another - within the same AP, e.g. 11b->11a, 11a->11b
must consider single MAC and multi MAC cases
MU client software considerations (e.g. higher level authentications, and so on)
Darwin Engwer, Nortel Networks

53. Credits Haixiang He - artistic design and animation
January 2004
Credits
Haixiang He - artistic design and animation
Bob O’Hara - technical review and sample sniffer trace
Darwin Engwer, Nortel Networks

54. January 2004
References
Std
Std F-2003
i-tgi-4-way-handshake-timings.ppt - July 2003, Nick Petroni
Netwave Roaming Specification , Engwer, et. al.
Darwin Engwer, Nortel Networks

55. January 2004
The End
Darwin Engwer, Nortel Networks

56. January 2004
Backup Slides
Darwin Engwer, Nortel Networks

57. Roaming in Multi-PHY APs
January 2004
Roaming in Multi-PHY APs DS
11a PHY
MAC
11b PHY
MAC
11g PHY
MAC
0x…1234
0x…1235
0x…1236
5 GHz
Radio
2.4 GHz
Radio
Darwin Engwer, Nortel Networks

58. Roaming in Multi-PHY APs
January 2004
Roaming in Multi-PHY APs DS
MAC
0x…1234
11a PHY
11b PHY
11g PHY
5 GHz
2.4 GHz Radio
Darwin Engwer, Nortel Networks

59. 802.11 STA MU AP* Definitions MU = Mobile Unit (Mobile STA)
January 2004
Definitions
STA MU
AP*
MU = Mobile Unit
(Mobile STA)
* STA operating as an AP.
Darwin Engwer, Nortel Networks

60. 802.11 STA AP MU Definitions January 2004
Darwin Engwer, Nortel Networks

61. 802.11 STA AP Definitions MMU UMU January 2004 (adhoc/ IBSS)
MMU = Managed MU UMU = Unmanaged MU (adhoc)
Darwin Engwer, Nortel Networks

62. January 2004
The End - Really
Darwin Engwer, Nortel Networks