4.2 Complementing the load balancing algorithm with guard bands
4.2.1 Handover prioritization and load balancing
We will first do some general considerations on what kind of a handover prioritiza- tion scheme could be used in Mobile WiMAX. We won’t design any detailed scheme, but just present the basic framework for it.
As discussed in part 3.2.1 the handover prioritization scheme in Mobile WiMAX should be distributed and local21, so that it complies with the WiMAX Forum net- work architecture and enhances scalability. Dedicated Resource Reservation in the next cell, let alone in an entire shadow cluster, is expensive and doesn’t fit well to Mobile WiMAX at least in the early stages.
Due to the flexible nature of Mobile WiMAX, dynamic guard band adaptation based on mobility22and traffic intensity in the neighboring BSs23is a natural choice as a basis for handover prioritization. Since efficient Resource Utilization is a cru- cial issue in Mobile WiMAX we don’t want the guard band to be too conservative.
Therefore a scheme that uses some kind of an initial prediction for the guard band and then reactively adapts it, based on how QoS guarantees, such as handover drop- ping rate, are fulfilled could be good for Mobile WiMAX. Such an approach would also be very simple.
What is especially interesting to us, in terms of load balancing, is how large the guard band is and how much it will vary, since load balancing can also be triggered in relations to the guard band.
4.2.1.2 Triggering load balancing in relations to the handover guard band
In the schemes discussed above load balancing is triggered in relations to high Re- source Utilization (Case 1 in Figure 4.6). It is however possible, when new flows are initiated with a rapid rate, that all resources become reserved before it is shown in the Resource Utilization measurements. This is bad because load balancing won’t be triggered to release resources and admission control will unnecessarily start to block calls.
21No BS to BS signaling except for the Spare Capacity Report.
22Rescue handovers conductedtothe BS.
23Although the SCR won’t give any information on the reserved resources, the Resource Utiliza- tion can be still used as a general indicator.
Figure 4.6: Resource Utilization and Resource Reservation based triggering.
Hence it will be beneficial to be able to trigger load balancing also in relations to the guard band for handovers24 (Case 2 in Figure 4.6). An example method to set the Resource Reservation based triggering threshold is discussed next. Ifλres is the average arrival rate of the new slot reservations and ts is the average holding time of a slot, we can use Little’s formula to calculate the number of reserved slots when the system is balanced
N =λrests (4.3)
We can use this to compute an estimation of a threshold for triggering load balancing in relations to current Resource Reservation
TR,est=G−(N −G)λres
λrel (4.4)
whereλrel indicates the rate at which the load balancing scheme can release slots.
As can be seen the higher N and the lower λrel are the earlier load balancing will be triggered. Since measurements can be inaccurate, we want to trigger load bal- ancing at latest when Resource Reservation reaches G and hence the final triggering threshold will be
TR= min(TR,est, G) (4.5)
So, the idea is to trigger load balancing before G is reached, but not too early to avoid unnecessary handovers25. The value of λrel depends on the duration of the discovery process used to identify overlapping cells and the handover mechanisms used. Since the handover guard band G might also vary, threshold setting can be a challenging task. The threshold could be further reactively tuned in relations to
24If no handover guard bands are used load balancing can be triggered in relations to maximum capacity or other possible guard bands (e.g. reserved for MCS changes or MAC headers).
25Some fluctuation can happen also on the flow arrival level.
a maximum call blocking rate valuebmax indicating the case where handovers were triggered too late and unnecessary handover rate valuehmax indicating when han- dovers were triggered too early.
When load balancing is triggered based on Resource Reservation the logic from the basic load balancing algorithm will not apply since it is based on Resource Uti- lization. Deciding which and how many MSs to handover and to which TBS is tricky with the current network architecture since not much Resource Reservation information is communicated between the BSs26. The per QoS profile Spare Ca- pacity Report could be utilized to some degree to determine which MSs to handover.
In any case directed handovers could be initiated with HO req messages and the admission control of the TBS could respond according to its Resource Reservation situation. The handover type could be differentiated from the regular (Resource Utilization based) directed handover by using the additional bits in the HO req handover type field. The arriving Resource Reservation based directed handovers could be treated as new calls in the TBS up until a certain point27. This way the flow arrival burden experienced by one BS would be distributed to the other BSs of the system.
4.2.1.3 Network directed retry and roaming
What about when increasing Resource Reservation has triggered load balancing and all possible load balancing directed handovers have been conducted? As discussed in part 3.1.2.2, in such a case, network directed retry and network directed roaming are potential methods to balance the load of incoming new flows. The basic idea is presented in Figure 4.7.
Figure 4.7: Network directed retry and network directed roaming.
26Again if a single Resource Reservation based triggering threshold is used, if the load difference between the BSs is clear and if Resource Reservation does not fluctuate that much, the highest priority connections (and most delay sensitive) should be handed over first to the less congested cell.
27A hysteresis approach could be used here also.
To make directed retry and network directed roaming work in Mobile WiMAX a few modifications to the initial network entry procedures should be made. When blocking occurs in a BS, a DSA RSP message could be sent to the MS initiating the service flow with an indication that directed retry or network directed roaming could be conducted. After that a similar discovery process to find out if the MS is in an overlapping area as desribed in the basic algorithm (see Figure 4.2), could be carried out resulting in a directed handover if the MS is residing in an overlapping area. Network directed roaming would be conducted as a last resort for the MS that is not in the overlapping area by communicating the location of the closest lightly loaded BS28. This would however require co-operation with application level protocols.