RADAR: Risk-and-Delay Aware Routing Algorithm in a Hybrid Wireless-Optical Broadband Access Network (WOBAN)

Abstract

We propose "risk-and-delay aware routing algorithm" (RADAR) for WOBAN. RADAR minimizes packet delay in the wireless front end of WOBAN and reduces packet loss for multiple failure scenarios: gateway failure, ONU failure, and OLT failure.

Full text

Page 1

RADAR: Risk-and-Delay Aware Routing Algorithm in a Hybrid
Wireless-Optical Broadband Access Network (WOBAN)

Suman Sarkar1, Hong-Hsu Yen2, Sudhir Dixit3, and Biswanath Mukherjee1
1Department of Computer Science, University of California, Davis, CA 95616
2Department of Information Management, Shih Hsin University, Taiwan
3Nokia Research Center, Burlington, MA 01083
{sarkar,mukherje}@cs.ucdavis.edu, hhyen@cs.shu.edu.tw, sudhir.dixit@nokia.com

Abstract: We propose “Risk-and-Delay Aware Routing Algorithm” (RADAR) for WOBAN.
RADAR minimizes packet delay in the wireless front end of WOBAN and reduces packet loss for
multiple failure scenarios: gateway failure, ONU failure, and OLT failure.
©2007 Optical Society of America
OCIS codes: (060.4250) Networks

1. Introduction
A hybrid wireless-optical broadband access network (WOBAN) consists of a wireless network at the front end, and it is
supported by an optical access network, viz., the passive optical network (PON) at the back end. At the back end of a
WOBAN, Optical Line Terminals (OLT) reside in a Central Office (CO) and feed to multiple Optical Network Units
(ONU). Thus, from ONU to the OLT/CO, we have a traditional fiber network; and, from ONUs, end users are
wirelessly associated. At the front end of a WOBAN, end users with wireless devices at individual homes or business
premises are scattered over a geographic area. They will associate (i.e., “connect”) with one of the wireless routers
nearby. Routers form a multi-hop wireless mesh networks and a few of them, called “gateways”, are strategically
placed. Gateways are attached to ONUs. An ONU can drive multiple gateways [1] (see Fig. 1 below).












An end user sends packet to one of its neighborhood routers. This router then injects the packet into the wireless
mesh of a WOBAN. The packet travels through the mesh, possibly in multiple hops, to find one of the gateways/ONUs
and is finally sent to the wired backbone via OLT/CO. The packet delay could be significant as the packet may travel
through several routers in the mesh before finally reaching the gateway (in the upstream direction) or to the user (in the
downstream direction). The larger the mesh of WOBAN, the higher is the expected delay. Packet loss may occur due to
multiple failure scenarios, viz., gateway failure, ONU failure, and/or OLT failure. Thus, to tackle these problems, we
propose “Risk-and-Delay Aware Routing Algorithm (RADAR)”, a proactive routing scheme (an extension to our
“Delay-Aware Routing Algorithm (DARA)” [2]). Moreover, we take a part of the city of San Francisco (called
“SFNet”, see Fig. 1) for our simulation experiments to observe how RADAR performs vs. other schemes.

2. Risk-and-Delay Aware Routing Algorithm (RADAR)
In RADAR, we model each router inside the mesh as an M/M/1 queue. Each router will advertise the wireless link
states (link-state advertisement or LSA) periodically. Based on the LSA information, we assign link weights to the
wireless links. Links with higher delay are assigned higher weights. Then, we compute the path with the minimum
average transfer delay from a router to any gateway and vice versa, and maintain a Risk List (RL) table in each router. If
a failure occurs, RL will be updated accordingly and the subsequent packets will be rerouted.
A packet’s average transfer delay, Qi along a wireless link i depends on its transmission delay, slot synchronization
delay (TDM-based operation of a wireless channel), and queuing delay (ignoring propagation delay because routers are
close to one another). Let λi denote the average packet intensity on link i, which is approximated by a Poisson
distribution. Consider that packet lengths are independent and exponentially distributed with average lengths as 1/µ, and
the effective link capacity is Ci. Then Qi is shown to be [1/(µCi) + 1/(2µCi) + 1/(µCi - λi)], where 1/(µCi): transmission
delay, 1/(2µCi): slot synchronization delay, and 1/(µCi - λi): queuing delay. Algorithm 1 shows our proposal RADAR.

Locations Archit-
Ecture
Multi-layer
deployment
Flat
deployment
Flat with
Intermesh
OC3ingress
in gateways
Compa-
tibility
WiFi,
WiMax
WiFi,
WiMax
WiFi
Config-
uration
Multi-
radio
Single
radio
Multi-
radio
Multi-
radio
Op.
range
2.4, 5
GHz
2.4
GHz
2.4
GHz
2.4, 5
GHz
Player
Athens,
GA
Chaska,
MN
Isla
Vista, CA
Tempe,
AZ
San Francisco, CA and Philadelphia, PA: currently
being deployed.
Belair
Tropos,
Pronto
Firetide
WiFi,
WiMax
Strix,
NeoReach
Earthlink,
Google
Fig. 1: SFNet: a part of San Francisco WOBAN. Table 1: Part of a survey on wireless mesh of WOBAN.
a1444_1.pdf


OThM4.pdf

Page 2

3. Analysis of RADAR
3.1 Risk Awareness: To reduce packet loss, each router maintains a “Risk List (RL)” to keep track of failures. An RL in
each router contains six fields, viz., path number (PN), Primary Gateway Group (PGG), Secondary Gateway Group
(SGG), Tertiary Gateway Group (TGG), path status (PS) (“live” or “stale”), and corresponding path delay (PD). The
primary gateway for a router is the gateway with the minimum delay path, Min(Σi∈PkWi), where Pk∈F and ∀k ∈ [1, K].
PGG contains paths with the primary gateway and the gateways connected to the same ONU as with the primary
gateway. SGG contains paths with gateways that are connected to different ONUs but the same OLT as with the PGG.
TGG contains paths with gateways that are connected to a different OLT (and consequently a different ONU).
Consider that we have a number of gateways at the edges of the front end of a WOBAN. Also consider that, at the
optical back end, we have different OLTs and each supports multiple ONUs. Each gateway will be assigned a gateway
id, viz., Cu
which, in turn, is connected to the w-th OLT (denoted by A) at the back end. For example, gateway id C1
assigned to the 1st gateway associated with the 16th PON group (or ONU) of the 2nd OLT.
Now, if a router finds five minimum-weight paths for a packet satisfying its delay requirement in the mesh, then F
= 5. Consider that the minimum-weight path (or alternatively the minimum-delay path) chooses gateway C1
Therefore, PGG in the RL in that router will contain the path with gateway C1
paths with gateways Cu
gateways Cu
paths with gateways Cu
C1
router, then its RL will be as shown in Table 2 (with A < B < C < D < E).

PN PGG SGG TGG PS PD
1 C1
Live A
2 C3
Live B
3 C1
Live C
4 C3
Live D
5 C2
Live E
5
Algorithm 1 Risk-and-Delay Aware Routing Algorithm (RADAR)
1. Link-State Advertisement (LSA): For each link i, advertise periodically current packet intensity (λi),
effective link capacity (Ci), and time stamp (tn).
2. Link-Weight Assignment: Assign weight of each link i as Wi = Qi = (1/(µCi) + 1/(2µCi)+ 1/(µCi - λi)).
3. Path Computation: Compute K minimum-weight paths (K > 1), (Σi∈PkWi), from a user to a gateway or
vice versa, where Pk is the k-th path for k∈[1, K]. Derive a set of paths, F (called “feasible paths”), from K
minimum-weight paths that satisfy the delay requirement of the packet in the mesh.
4. Risk List Update: Maintain a Risk List (RL) table in each router based on F. Update RL at next LSA.
5. Path Selection: Among paths in RL, choose a “live” path. (See Section 3 for details.)
6. Admission Control: Admit a new packet in the mesh only if its delay requirement (Treq) satisfies the
minimum delay among the “live” paths, i.e., MinPk∈F(Σi∈PkWi) < Treq. Else reject the packet.
BvAw, where Cu
BvAw stands for the u-th gateway (denoted by C), associated with the v-th ONU (denoted by B),
B16A2 will be
B1A1.
B1A1 (primary gateway) and any other
B1A1 (means gateways that are connected to ONU B1 of OLT A1). SGG will contain paths with
BvA1, v ≠ 1, (means gateways that are connected to ONUs of OLT A1, except ONU B1). TGG will contain
BvAw, w ≠ 1, (means gateways that are connected to ONUs of any OLT, except OLT A1). Now, if
B3A1, C1
B1A1, C3
B16A1, C3
B1A1, and C2
B5A2 are the gateways of five minimum-weight paths (in ascending delay) for that






Table 2: Risk List (RL) in a router. Table 3: Update Risk List for Gateway failure.
PN PGG SGG TGG PS PD
1 C1

Stale
A
2 C3
Live B
3 C1
Live C
4 C3

Stale
D
5 C2
Live E

for C1
through “live” PGG, SGG, or
TGG paths. Table 4 considers
an ONU failure. If ONU B1
fails, all PGG paths will be
“stale”; packets will be rerouted
through SGG and TGG paths.





Table 4: Update Risk List for ONU failure. Table 5: Update Risk List for OLT failure.
Table 5 considers an OLT failure. If OLT A1 fails, all PGG and SGG paths will be “stale”; but the packets could still be
rerouted through “live” TGG path with C2
back end hierarchical failure scenarios.
B1A1
B3A1
B16A1
B1A1
B5A2
PN
1
2
3
4
PGG
C1


C3
SGG

C3
C1

TGG




C2
PS
Stale
Live
Live
Live
Live
PD
A
B
C
D
E
B1A1
B3A1
B16A1
B1A1
B5A2
B1A1
B3A1
B16A1
B1A1
B5A2
PN
1
2
3
4
5
PGG
C1


C3

SGG

C3
C1


TGG




C2
PS
Stale
Stale
Stale
Stale
Live
PD
A
B
C
D
E
B1A1
B3A1
B16A1
B1A1
B5A2
Table 3 shows how RL will be
updated if a gateway failure
occurs. If primary gateway
C1
with that gateway will be
“stale”, and packets, destined
B1A1, will be rerouted
B1A1 fails, then all the paths
B5A2. Therefore, RADAR can provide protection for multiple front end and
3.2 Self-healing: If all links adjacent to gateway ids C1
“live”, then routers can infer that either both gateways C1
failed. Then packets will be rerouted through SGG and TGG paths. If all links adjacent to gateway ids Cu
then routers can infer that either all these gateways Cu
have failed simultaneously, or OLT A1 has failed. Then TGG paths should take care of the packets. Therefore, we
observe that a router does not always need to recompute a new set of K minimum-weight paths even if a failure occurs.
B1A1 and C3
B1A1 and C3
B1A1 go down, and that of C3
B1A1 have failed simultaneously or ONU B1 has
B3A1 and C1
B16A1 are
BvA1 go down,
BvA1 have failed simultaneously or their corresponding ONUs Bv
a1444_1.pdf


OThM4.pdf

Page 3

Router will recompute paths only if all its previously computed paths fall under PGG and the ONUs/OLT fail, or all
paths fall under PGG and SGG and the corresponding OLT fails. After path-recomputation, packets will be admitted in
WOBAN with degraded service (alternatively, increased delay). We call this mechanism “self-healing”.
3.3 Delay Awareness: In LSA (see Algorithm 1), each router/gateway will periodically advertise its link conditions, and
link weights are assigned in such a manner that links with more delay get higher weights and vice versa. Then we
compute the K minimum-weight paths, which are the K minimum-delay paths as well. We choose a path whose delay is
below a threshold. So, RADAR works on finding a packet’s delay-optimized path in the front end mesh of a WOBAN.

4. Performance Study
We compare how RADAR performs vs. Min-Hop Routing Algorithm (MHRA, where links weight are unity), Shortest-
Path Routing Algorithm (SPRA, where link weights are inversely proportional to link capacities), and Predictive-
Throughput Routing Algorithm (PTRA, a popular algorithm used in front end of a WOBAN where packets choose path
with higher estimated throughput). PTRA is similar to Tropos’s (www.tropos.com) Predictive Wireless Routing
Protocol or PWRP. We consider a part of San Francisco city for our performance study (SFNet, see Fig. 1). SFNet is
approximately a one square-mile area in downtown San Francisco with an estimated population of around 15,000
residents. In SFNet, we distributed 25 wireless routers. We designated five of these 25 routers as gateways to the optical
back end of WOBAN and placed them at the edges of SFNet. We generated packets in a Poisson distribution. The
packet lengths are independent and exponentially distributed. We assumed a wireless router’s capacity to be 11 Mbps.

0
10
20
30
40
0 204060 80100
Simulation time (%)
Packet loss (%)
MHRA
SPRA
PTRA
RADAR

0
5
10
15
20
25
02040 60 80 100
Simulation time (%)
Packet loss (%)
MHRA
SPRA
PTRA
RADAR
Fig. 2: Packet loss for gateway failure in SFNet.
Figure 2 shows that, for a wireless gateway failure, RADAR (packet loss < 1%) performs much better than MHRA,
A (packet loss ~20%), and PTRA (packet loss ~5%). For an ONU failure (see Fig. 3), packet loss in RADAR is 1-
2%, whereas for MHRA and SPRA, it is ~35%, and for PTRA, it is around 10%. For OLT failure (see Fig. 4), though
packet loss for RADAR is increased, it still performs better than other schemes MHRA, SPRA, and PTRA (loss ~45%).
Also note that PTRA performs worse in OLT failure; this is because, PTRA senses the wireless channel and finds if any
link fails due to either gateway failure or ONU failure in close geographical proximity. So, PTRA can reroute packets to
other gateways in close proximity. OLTs are often far apart and gateways attached to different OLTs are also spaced
out. So, PTRA often does not have any information regarding these gateways. RADAR, on the other hand, keeps track
of all the gateways in its RL table and performs well.
Figure 5 shows that RADAR also achieves lowe
and PTRA). Therefore, RADAR achieves its goal of delay-optimization as well packet-loss-minimization due to failure.

SPR
r systemwide average packet delay in SFNet (vs. MHRA, SPRA,
Fig. 3: Packet loss for ONU failure in SFNet.
0
20
40
60
020 406080100
Simulation time (%)
Packet loss (%)
MHRA
SPRA
PTRA
RADAR
0.1
1
10
00.20.40.60.81
Load (normalized)
Delay in ms
MHRA
SPRA
PTRA
RADAR

5. Summary
We developed “Risk-and-Delay Aware Routing Algorithm (RADAR)” for WOBAN. Our performance studies show
that RADAR minimizes the average packet delay in the front end mesh of WOBAN and reduces the packet loss for
multiple failure scenarios, viz., gateway failure, ONU failure, and OLT failure.
Fig. 4: Packet loss for OLT failure in SFNet. Fig. 5: Average packet delay in SFNet.
B. Mukherjee, and S. Dixit, “Optimum Placement of Multiple Optical Network Units (ONUs) in Optical-Wireless
Hybrid Access Networks,” Proc., OFC 2006, Anaheim, CA, March 2006.
S. Sarkar, H. Yen, S. Dixit, and B. Mukherjee, “DARA: Delay-Aware[2] Routing Algorithm in a Hybrid Wireless-Optical
Broadband Access Network (WOBAN),” submitted.
References:
[1] S. Sarkar,
a1444_1.pdf


OThM4.pdf