CS230: DISTRIBUTED SYSTEMS
Project Report
on
Implementation of Near Optimal Algorithm for Integrated Cellular and
Ad-Hoc Multicast (ICAM)
Prof. Nalini VenkatasubramanianProject Champion: Ngoc Do
Vimal Aiyappath Narain (80763860)Mahit Murthy (59321605)
The University of California, Irvine, CA – 92697
Index
Introduction..................................................................................................3Motivation and Related Work......................................................................4Models and Assumptions.............................................................................6The Problem.................................................................................................7Our Solution.................................................................................................8Results.........................................................................................................11Conclusion and Future Work......................................................................14
2
Introduction
Third-Generation (3G) CDMA wide-area wireless networks has experienced explosive growth in the recent times. CDMA users have crossed over 100 million in number. As the user population builds up, group communications such as on-demand video streaming, group messaging, and gaming through hand-held wireless devices have been spurring the development of multicast functions in the 3G network infrastructure. 3G standard bodies 3GPP and 3GPP2 have been actively standardizing multicast services.
Existing multicast strategy in 3G networks suffers in terms of decreased downlink channel utilization as the size of the multicast group increases. In order for the multicast receiver with the worst downlink channel condition to correctly decode data frames from 3G downlink, a conservative strategy for the 3G base station is to use the lowest data rate among all the receivers in the multicast group. Due to path loss and fast fading of the wireless medium, the likelihood that at least one receiver experiences bad downlink channel condition increases as the multicast group size increases, resulting in decreased throughput for the multicast group. This phenomenon is in stark contrast to the unicast scenarios where increasing the number of users increases the downlink channel utilization using Proportional Fairness Scheduling
One approach to increasing 3G throughput is through the use of ad hoc relays. In this model, mobile devices are assumed to have both 3G and IEEE 802.11 interfaces. The mobile receiver first discovers a proxy client with better 3G downlink channel condition. On behalf of the receiver, the proxy client receives data packets from the base station at higher data rate. The proxy then forwards the packets through the IEEE 802.11-based ad hoc network to the mobile receiver. While this model has been shown to significantly improve throughput for 3G unicast traffic, extending this model for multicast traffic is not trivial since multicast traffic can easily overload an IEEE 802.11-based ad hoc network, limiting the achievable throughput gains. Thus, in order to maximize throughput in ICAM, it is not sufficient to choose the proxy with the highest 3G data rate connection. The 3G data rate of the proxy must be balanced by the throughput achievable over the interference-prone ad hoc relay network. Therefore, it is important that the choice of the proxies and the construction of the multicast forest be performed jointly with explicit awareness of the capacity limitations of the ad hoc relay network.
A polynomial-time approximation algorithm is implemented in this project that outputs near optimal multicast relay strategy. The algorithm is based on a very general interference model and has an approximation factor of four. This is also the first multicast routing design that explicitly considers multi-hop wireless interference in ad hoc networks. The algorithm is equally applicable when the underlying wireless MAC supports broadcast or unicast, single rate or multiple rates and even when there are multiple simultaneous multicast sessions.
3
Motivation and Related Work
3G multicast is inherently limited by the worst channel rate of the multicast group. More precisely, assume that there are n clients covered by a 3G base station, and l of these clients (denoted by the set R) belongs to a multicast group. If multicast receiver v Є R has an instantaneous downlink channel rate of ri
v at time slot i, then the data rate for the multicast at time slot i is minvЄR ri
v. Due to path loss and fast wireless channel fading, the likelihood that at least one multicast receiver experiences bad downlink channel condition increases as the multicast group size increases. Therefore, increasing the number of receivers results in lower throughput for the multicast group. 3G multicast’s inefficiency motivates us to use relays to improve throughput. Specifically, for each multicast receiver v with average downlink channel rate rv, find a proxy client with higher average downlink channel rate p(v) and an ad hoc relay route from the proxy to the receiver.
The traffic in future cellular networks is expected to be unbalanced. In a cellular system, a Mobile Host (MH) can use only the data channels of the base transceiver station (BTS) located in the same cell, which is a subset of the data channels available in the system. No access to data channels in other cells by the MH limits the channel efficiency and consequently the system capacity. Specifically, some cells may be heavily congested (called hot spots), while the other cells may still have enough available data channels (DCH). Even though the traffic load does not reach the maximum capacity of the entire system, a significant number of calls may be blocked and dropped due to localized congestion. Since the locations of hot spots vary from time to time it is difficult to provide the guarantee of sufficient resources in each cell in a cost-effective way. ICAR attempts to solve this issue of balancing traffic loads between cells by using ad hoc relaying stations (ARS) to relay traffic from one cell to another dynamically based on need. ARS is a specialized hardware that can communicate with both MHs and Base Station using different channels. The mobile node hence needs an additional radio interface to communicate with ARS.
In UCAN, a mobile client has both 3G cellular link and IEEE 802.11-based peer-to-peer links. The 3G base station forwards packets for destination clients with poor channel quality to proxy clients with better channel quality. The proxy clients then use an ad-hoc network composed of other mobile clients and IEEE 802.11 wireless links to forward the packets to the appropriate destinations, thereby improving cell throughput. The 3G base station scheduling algorithm is refined so that the throughput gains of active clients are distributed proportional to their average channel rate, thereby maintaining fairness. With the UCAN architecture in place, novel greedy and on-demand protocols for proxy discovery and ad-hoc routing that explicitly leverage the existence of the 3G infrastructure to reduce complexity and improve reliability are proposed.
Multicast over ad hoc networks has been intensively studied in recent years. The proposed multicast routing protocols can be classified into two categories. One category
4
is tree-based, including Reservation-Based Multicast (RBM), Lightweight Adaptive Multicast (LAM), Ad hoc Multicast Routing Protocol (AMRoute), Ad hoc Multicast Routing protocol utilizing Increasing id-numberS (AMIS), and multicast extension of Ad hoc On-demand Distance Vector (MAODV). They all build a shared or core-based tree to deliver multicast data, but differ in detailed mechanisms for tree construction, maintenance, and adaptation to the network topological dynamics. The other category is mesh-based, including Core-Assisted Mesh Protocol (CAMP), and On-demand Multicast Routing Protocol (ODMRP). They enhance the connectivity by building a mesh with multiple forwarding paths and, therefore, improve the resilience as the network topology changes.
LAM is a multicast protocol which is built upon the temporally ordered routing algorithm (TORA). The protocol termed Lightweight Adaptive Multicast routing algorithm is designed for use in a Mobile Ad Hoc network and conceptually can be thought of as an integration of the Core Based Tree multicast routing protocol and TORA. The integration with TORA increases reaction efficiency as the new protocol can benefit from TORA's mechanisms while reacting to topological changes. LAM does not require timer based messaging during its execution and hence does not incur additional group membership and stable topology period overhead.
The Core-Assisted Mesh Protocol (CAMP) is introduced for multicast routing in ad-hoc networks. CAMP generalizes the notion of core-based trees introduced for internet multicasting into multicast meshes that have much richer connectivity than trees. CAMP consists of the maintenance of multicast meshes and loop-free packet forwarding over such meshes. Within the multicast mesh of a group, packets from any source in the group are forwarded along the reverse shortest path to the source, just as in traditional multicast protocols based on source-based trees. CAMP guarantees that, within a finite time, every receiver of a multicast group has a reverse shortest path to each source of the multicast group. Multicast packets for a group are forwarded along the shortest paths from sources to receivers defined within the group’s mesh. CAMP uses cores only to limit the traffic needed for a router to join a multicast group; the failure of cores does not stop packet forwarding or the process of maintaining the multicast meshes.
5
Models and Assumptions
Interference Model
We assume a general proximity-based interference model for the IEEE 802.11-based multi-hop wireless network. In this model, the transmission of a node u does not cause interference to the transmission of a node x, if their distance is greater than RI, where RI is the maximal interference range. We assume that RI is qxRt, where Rt is the transmission range of each node and q >= 1.
Hop limit for Proxy Discovery
We assume that the hop distance between a proxy and any of its receivers is upper-bounded by a small number h. There are several reasons for h being small. First, due to interference, wireless channel error, and lack of scheduling, IEEE 802.11 ad hoc network throughput decreases very fast as the number of hops increases. Thus, if the ad hoc throughput decreases to the minimal HDR downlink channel rate in less than h number of hops, the IEEE 802.11 ad hoc network would become the bottleneck, contradicting the need for relays to increase 3G multicast traffic. Second, paths of length exceeding a certain number of hops are not desirable because of the increased probability of route breakage due to mobility, latency, overhead of proxy discovery, and routing update overhead over the HDR uplink. Our simulation study shows that, for a 500m radius cell, using a proxy beyond a h=4 hop neighborhood of the multicast receiver does not further increase throughput.
Minimal Separation and Location
The near optimal algorithm makes two more assumptions. We assume a minimal separation of distance sRt between any pair of transmitters where s > 0. This assumption is natural since the two transmitters can not be co-located in space. We also assume the base station knows the location of each node. Clearly, this is not an issue when the relays are fixed as part of the infrastructure. For the case where relays are mobile nodes, location information has to be obtained in other ways such as through the base station’s estimate using signal strength and angle of arrival as part of the E911 service, or through GPS and other localization mechanisms.
6
The Problem
An ad hoc network is represented as a graph G = (V, E) where V is the set of (n) nodes and E the set of (m) links. If (u, v) Є E, then node u and v are at most Rt apart. The set R ⊆ V is the set of receivers of a given 3G multicast group. The average 3G downlink rate of node v Є V is rv >= 0. A receiver v Є R may receive data either directly from the 3G base station— at rate rv, or through a proxy. The ad hoc relay sub-network for a given multicast group is graph G’ = (V’, E’), where (u, v) Є E iff u, v Є V’ and (u, v) Є E.
Relay sub-network G’ is composed of a (forest) collection of directed trees F spanning node set V’. Along each tree T Є F, data (received from the 3G base station) is multicasted from the root (proxy) to all the receivers in T. Receivers in R - R3G receive data at the rate of max{rv, ra
v}, where rav is the rate at which data is received through the
ad hoc relay sub-network G’. For a given G’, we denote max{rv, rav} as rv(G’). We denote
by I = (V, A) the interference graph for the ad hoc network. Thus, two ad hoc nodes u and v interfere with each other iff (u, v) Є A. (u, v) ∉ A if u and v are at least qRt apart for some fixed constant q. Given G’ = (V’, E’), let k(G’) denote the minimum number of colors required to color the nodes in V’ such that two nodes u, v Є V’ have the same color iff (u, v) ∉ A. Therefore, the best multicast rate that can be achieved in G’ is at most f=k(G’). To be precise, only non-leaf nodes need to be colored since the leaf nodes (receivers) do not participate in transmissions. Our results, although applicable to this more precise model, are more involved and hence for ease of presentation we will use the model where all leaf receivers, except R3G, are colored.
We make the assumption that the best proxy for receiver v is no more than h hops away for some small value h, e.g., h = 4. For receiver v Є R - R3G, let p(v) be the rate of the 3G proxy for v. Then, ra
v = min{f/k(G’); p(v)}. Hence, the multicast rate for v in G’ is
rv(G’) = max{rv, min{f/k(G’), p(v)}}
Denote r(G’) = minvЄR rv(G’). The ICAM problem is to compute G0 such that the multicast rate r(G’) for its associated ad hoc sub-network G’ is maximized. The ICAM problem is NP-hard.
7
Our Solution
The basic idea behind our polynomial-time approximation algorithm is to leverage the property that for each multicast receiver the best proxy can be at most h-hops away. We perform localized proxy search and construct the optimal multicast trees for the receivers within certain locality. All local optimal multicast trees are then combined to form the global multicast forest. The whole algorithm can be described as follows:
It first divides the coverage area of a 3G base station into a two-dimensional grid of (2h + q + Є)Rt X (2h + q + Є)Rt cells, where Є > 0. It makes the algorithm polynomial-time by choosing the above mentioned size for the grid cells as it restricts the number of receivers per cell by placing an upper bound on the distance within which the proxy for each receiver in the cell exists. It then computes a solution for each cell of the grid that contains at least one receiver. In other words, it computes the optimal solution for ICAM when restricted only to the receivers RC ∉ R in a cell C Є Γ. Finally, the algorithm merges these solutions for all cells to compute a feasible solution to the original instance of the problem. A description of the scenario is as shown below:
8
(2h+q+ ε )Rt
(2h+q+ ε )Rt
Let C be a cell with at least one receiver (|RC| > 0) and VC ⊆ V be the set of all nodes that are at most h hops from at least one receiver in RC. Note that in any optimal solution the set of proxies for the receivers in RC and any intermediate relay nodes must be in VC. The algorithm computes the optimal solution for a cell C as follows:
It enumerates all subsets V’C of nodes in VC. For a given subset V’C, let G’C be its associated ad hoc relay sub-network. The algorithm computes the minimum number of colors k(G’C) needed to color the vertexes of G’C based on the interference graph I.
9
BaseStation
Mobile Receiver which directly receives from the base station
Mobile node selected as Proxy
Mobile node which receives from the proxy
The base station covers the whole area
The algorithm then computes the best proxy p(v) in G’C for every receiver v Є V’C. Note that all this information is sufficient to compute rv(G’C) for every receiver v Є V’C. The rv(G’C) of all receivers v Є RC that are not in V’C is rv(G’C) = rv. Taking the minimum of rv(G’C) for all receivers v Є RC, the algorithm is able to compute the multicast rate r(G’C) for the ad hoc relay sub-network for the receivers in RC. The algorithm then selects the subset V’C ⊆ VC, whose associated ad hoc relay sub-network has the highest rate, to generate the optimal multicast strategy for the receivers RC in cell C. We will show later that all this can be done in constant time. Finally it outputs the union of the sub-networks computed for each grid cell, i.e., V0 = ∪ CЄΓ V’C ⊆ V as the solution for the original problem instance. The set of receivers R3G that receive directly from 3G base station is R ∩ (V – V0).
10
ResultsWe implemented the Near Optimal algorithm for ICAM. We were not able to configure the 3G simulation settings on Qualnet. For that reason, now we are using Wi-Fi interfaces for both. For Wi-Fi, nodes have to be within transmission range of each other. We simulated with different setups and compared our algorithm with the original 3G algorithm. We were able to achieve up to 600% improvement in the throughput with the node setup having a total of 30 nodes being served by the base station. Following are the graphical representation of our results:
The messages are used for sending coordinates by the mobile nodes to the base station and by the base station to inform the nodes about how they are going to receive their data and if they are going to transmit their received data to anyone else.
11
The setup time increases exponentially with the increase in the number of nodes. Hence, this algorithm does not scale very well and it does not work really well when the scenario is highly mobile. It is because nodes continuously keep moving and hence they keep leaving one multicast group and joining other multicast groups and the base station has to recalculate the whole algorithm again and again. The greedy algorithm outperforms the near optimal algorithm in such a setting. But, in static scenarios, our algorithm works very well.
12
The above graph shows the comparison of performance with the original 3G. It can be clearly seen that the near optimal algorithm outperforms the original 3G rates. We are able to achieve about 600% improvement in the 3G downlink rate when compared to the original 3G downlink rates.
13
Conclusions and Future Work
In this project, we implemented the ICAM architecture for improving 3G multicast throughput using ad-hoc relays. The near-optimal algorithm can be extended to multi-rate ad-hoc networks and to networks with multiple multicast groups. 3G multicast throughput is limited by the receiver with the worst channel rate. By finding proxies for receivers with poor channel quality and relaying multicast packets through an IEEE 802.11-based ad hoc network, the throughput of multicast sessions can be significantly improved. We implement a near-optimal algorithm that runs in polynomial time. Through extensive simulations, we showed that the near-optimal algorithm improves the average 3G multicast throughput by up to 840 percent and outperforms the greedy heuristics up to 92 percent in static scenarios, while the greedy algorithm achieves throughput gains of as much as 410 percent in relatively high mobility scenarios.
14
