Patent Application
20070070933
Currently, third-generation, wide-area wireless networks based on CDMA2000 and UNITS are being deployed throughout the world. These networks provide both voice and high-speed data services. In order to reduce the cost of these services to attract more subscribers, network operators must reduce their capital and operating expenses.
Typically, such networks include base stations and radio network controllers (RNCs) that are connected by expensive point-to-point T1/E1 links. In such a point-to-point architecture a group of RNCs may be shared by a small set of base stations. During hot-spot and peak hours, this architecture is susceptible to significant call blocking. To avoid this, a network operator typically needs to appropriately add capacity to an RNC. This, however, increases capital costs.
Furthermore, in this architecture, an RNC itself may be a single point of failure. To account for this possibility, network operators typically build in redundancy which increases the cost of each RNC.
One way to reduce these costs is to replace point-to-point links with an Internet Protocol-based Radio Access Network (IP-RAN). FIGS. 1(a) and (b) depict examples of a conventional RAN and an IP-RAN, respectively.
An IP-RAN provides a number of benefits, including:
While studies have shown that it is feasible for an IP-RAN to support QoS requirements, the question of connectivity, i.e. how best to connect base stations to RNCs in an IP-RAN, had not been addressed by any research literature to our knowledge until the inventors addressed this in their research paper entitled, Connectivity and performance of IP-based CDMA Radio Access Networks (Infocom, April, 2004), the disclosure of which is incorporated in full herein as if set forth in full herein.
The inventors' research introduced techniques for optimizing the connection of base stations to RNCs in an IP-RAN. It is worthy of note that this was a hard problem to solve because, even for a simple network of 100 base stations and 10 RNCs, the number of possible connection patterns between the base stations and the RNCs is enormous (e.g., 21000).
The inventors were also the first to introduce techniques that assign an appropriate RNC to an incoming call once an optimal connection pattern was determined. These techniques ensure that call dropping (handoff calls) and call blocking (new calls) rates are minimized, with priority given to handoff calls.
One technique, known as Min-Load-1 performed significantly better than other techniques. Its performance was very close to that of more complicated schemes.
Though the inventors' earlier research introduced techniques for solving base station connectivity and RNC assignment problems for IP-RANs, there remains the problem of determining how many links to include within an IP-RAN to allow the IP-RAN to operate effectively (i.e., in a connected state) when failures occur, the so-called “resiliency” of a network.
It is therefore desirable to provide for methods and systems that provide highly resilient IP-RAN networks.
We have recognized that it is possible to estimate the number of links necessary to construct highly resilient IP-RANs. The number of additional links required over and above a single connected IP-RAN are relatively few.
In one embodiment of the invention, a resilient, IP-RAN comprises one or more base stations connected to one or more RNCs to form a network having an arc connectivity of 1 or more, wherein the IP-RAN includes a plurality of links whose number is a certain magnitude more than a number of links in a single connected IP-RAN.
In yet further embodiments of the invention, the number of links is substantially equal to 10% or 20% more than the number of single-connected links. For the 20% embodiment, the IP-RAN accommodates a failure of one of the base stations substantially effectively as a fully connected RAN.
In still yet another embodiment, when the arc connectivity is set to 2 and the number of links is substantially equal to twice the number of links in a single-connected IP-RAN, the resulting IP-RAN accommodates a single RNC failure substantially effective as a fully connected RAN.
FIGS. 6(a) and 6(b) depict graphs of rejection probabilities for RANS when a base station fails; and
FIGS. 7(a) and 7(b) depict graphs of rejection probabilities for RANS where an RNC fails.
Referring to FIGS. 1(a) or 1(b) there is shown a network architecture that includes the following components: a set of RNCs, a set of base stations, a set of communication links, that connect the base stations to the RNCs and a set of users. Note that in practice, the links may take the form of either a T1 leased line, an ATM connection or an MPLS path. Note also that many logical links may traverse the same physical link. This logical connection assures a Quality of Service necessary to ensure that CDMA soft handoffs function correctly. A user in the network can be either active or idle. A user, whether active or idle, is associated with a base station. An active user needs access to the radio resources of a base station and to the processing resources of an RNC. The links, L, may make up either a RAN (
As indicated before, the inventors' earlier research explored the question of what algorithm or technique should be used to assign a call to an RNC (i.e, to a base station responsible for the call) so that call blocking and dropping rates are minimized for a given RAN. A load-balancing technique known as Min-Load-1 was determined to be the most attractive technique.
Before we proceed further, we need to define some concepts that were used by the inventors in their earlier research. One such concept is referred to as “graph connectivity”.
A graph is said to be connected if there is at least one path between every pair of nodes. The arc connectivity of a connected graph is the minimum number of arcs whose removal from the graph disconnects it into two or more components. For example, with N RNCs (e.g., N=2) and M base stations (M>N), a mesh connectivity has M×N links and is of arc connectivity N (e.g., N=2).
The approach used by the inventors in their research was to focus on a set of balanced graphs with properties that are desirable in a homogeneous network where RNCs have the same capacity and the base stations have the same average load. Each element in this set of balanced graphs has a different number of links L. The members of this set can be enumerated by varying the number of links L from M to NM. By focusing on this set of balanced graphs, the inventors discovered that the number of possible links could be reduced from 2NM to NM. Given that there is very little known about the impact of connectivity on homogeneous networks, the inventors focused on the homogeneous network case. Heterogeneous networks are discussed briefly in subsequent sections.
Closely related to a balanced graph is an RNC accessibility graph. Such a graph was also discussed in detail in the inventors' earlier research.
A balanced graph whose corresponding RNC accessibility graph (in a full-mesh connected network) is also balanced was presented and discussed in the inventors' earlier research. For present purposes, it is sufficient to point out that the concepts of balanced graphs and an RNC accessibility graphs introduced by the inventors in their earlier research reduced the so called “state space” of possible connectivity patterns from 2NM to NM, while maximizing performance.
As indicated above, the inventors' earlier research concluded that a Min-Load-1 technique was the most useful in assigning an RNC to a base station. In arriving at this conclusion, it should be noted that the inventors compared various Min-Load, Optimal, and Min-Load-k techniques/algorithms. The details concerning these comparisons are set forth in the inventors earlier research. As the astute reader will recognize, the Min-Load-1 technique that was ultimately selected is a variant of a Min-Load-k technique.
Another discovery that came from the inventors' earlier research was the realization that it is important to keep k as small as possible because a large k incurs more hard handoffs and longer call setup times.
We now turn to a discussion of the subject matter of the present invention, namely, an estimation of the number of links needed to be included with an IP-RAN to make it sufficiently resilient against failures.
In general, there are three possible types of failures: link failures, base station failures and RNC failures. However, because the failure of a single link is, in the worst case, as serious as one base station failure when the base station the link connects to is singularly connected, we will not present the evaluation of single link failure.
In accordance with one embodiment of the invention, a highly resilient IP-RAN can be designed to absorb a single base station type of failure. Because a minimum connected balanced graph is uniform in its connectivity, we can simply randomly pick any base station to fail.
We now compare the resiliency of a RANs with arc connectivities of 1 and 2.
When the Min-Load-1 load balancing technique is used (and we assume it is), the minimum connectivity (i.e., number of links) required to allow an IP-RAN to achieve good performance is equal to the number of links in a single-connected graph plus 10 links (normalized to 10%) added in a balanced way. When a base station fails, the connectivity (i.e., number of links) needed to ensure an IP-RAN's resilience is approximately the number of links in a single-connected graph plus 20 links (i.e., 20%).
In sum, the inventors have discovered that: (a) a RAN having an arc connectivity of 2 is much more resilient to a single RNC failure than a RAN having an arc connectivity of 1; and (b) adding more links to a RAN having an arc connectivity of 2 does not improve a RAN's resilience to a single RNC failure significantly; (c) adding twice the number of links ensures resiliency.
In yet further embodiments of the invention, the same connectivities at both higher and lower loads have found that the turning point where the rejection rate drops significantly changes with load (“turning point”). For example, the turning point moves towards a RAN having an arc connectivity of 2 when the load decreases. It can be argued that a minimum connected balanced graph having an arc connectivity of 2 is the minimum connectivity required to maintain a low rejection rate after an RNC failure because any graph with a lower connectivity will be partitioned (i.e., one or more base stations are not connected to any RNC) after an RNC failure.
In sum, the inventors have discovered that in order to make a RAN resilient to RNC failures at any load, an arc connectivity of 2 is necessary.
In the approach presented so far, we have assumed that the IP-RAN is homogeneous. That is, all of the base stations have the same average load, all of the RNCs have the same capacities and all of the link costs are the same.
In a further embodiment of the invention, the techniques discussed herein and in the inventors' earlier research may be extended to a heterogeneous IP-RAN.
One approach is to map a heterogeneous IP-RAN to a constrained homogeneous one using the following strategy.
The heterogeneous RNCs/base stations may be split into homogeneous logical RNCs/base stations with capacities/loads equal to the highest common denominator of all the RNCs/base stations. In order to mimic the physical locality of the RNCs/base stations, whenever a logical base station is connected to a logical RNC in a connectivity model, additional links are added between all the corresponding logical base stations of the original heterogeneous base station to all the corresponding logical RNCs of the original heterogeneous RNCs.
It should be understood that this transformation is just one possible way of analyzing connectivity in heterogeneous networks.
In the discussion above, we have set forth some estimates of the number of links required in an IP-RAN to ensure its resiliency in the face of a base station or RNC failure. Underlying these estimates is the assumption that an RNC assignment technique, called Min-Load-1, described in the inventors' earlier research is used as a load balancing technique to assign an incoming call (i.e., new call or a handoff) to an RNC.
The estimates presented herein provide additional support for deploying IP-based RANs because they suggest that it is possible to enhance the resiliency of existing current point-to-point RANs by adding a relatively small number of additional links.
