Dynamic optical resource allocation for mobile core networks with software defined elastic optical networking

Yongli Zhao; Zhendong Chen; Jie Zhang; Xinbo Wang

doi:10.1364/OE.24.016659

1. Introduction

With the forthcoming of 5G mobile communications, the amount of IP data handled by wireless networks will increase over a factor of 100: from under 3 exabytes in 2010 to over 190 exabytes by 2018, on pace to exceed 500 exabytes by 2020. Then, the aggregate data rate will increase by roughly 1000 times from 4G to 5G [1]. Some disruptive technologies have led to fundamental changes of cellular networks [2]. To support new emerging mobile applications (e.g. high-definition video services and mobile cloud-computing services), mobile broadband communication should be realized in the whole mobile networks. However, this development means that the volume of data to be transported through the core networks will increase greatly as well. Mobile core network is the part that links the power of high-speed wireless access technologies with the power of the innovative application development enabled by the Internet [3]. 3GPP has proposed an all-IP architecture for long-term evolution (LTE) network, i.e. evolved packet core (EPC), which consists of mobility management entity (MME), service gateway (SGW) and packet data network (PDN) gateway (PGW) [4]. However, restrained by the transmission capability and complexity of the protocols, EPC will encounter its bottleneck, such as the bandwidth and elasticity required by 5G mobile networks.

In order to support various broadband services, a highly reliable and flexible EPC network with high bandwidth is necessary to be deployed. Optical network could provide high bandwidth communication links and reduce the cost of power-hungry core gateways by offloading the IP traffic into optical paths at suitable bandwidth [5]. The role of optical network in 5G is comprehensively expounded in [6], which shows the great potential to support 5G mobile communications. For example, optical access/aggregation networks based on wavelength division multiplexing (WDM) are considered as an outstanding candidate for 5G transport networks [7]. Flexible grid optical networks with the enabling technologies are introduced to provide elastic, transparent, and virtualized optical paths between the BBU pools [8]. In addition, E2E 5G platform is developed with the integration of heterogeneous wireless access and optical transport networks, distributed cloud computing, wireless sensor, and actuators networks [9], and virtualization technology is deployed for heterogeneous networks integrating optical metro networks and wireless access networks [10]. Different from previous works, this paper contributes to improve the performance of EPC networks in 5G with optical networking.

On the other hand, with the great development of software defined networks (SDN) [11], plenty works have been done in applying SDN to optical networks. Software defined optical networks (SDON) shares similarities with SDN. For example, control plane and data plane are decoupled and connected through a common open interface, most network control and management intelligence are centralized in SDN controller, and the physical hardware can be flexibly configured to support different network demands and conditions [12]. In [13], a software defined elastic optical networking (SD-EON) testbed is demonstrated with elastic lightpath provisioning, and the integrated architecture of flexible grid optical networks with IP infrastructures is proposed and verified based on SDON with end-to-end lightpath provisioning [14,15]. Considering the major functional entities of EPC specified in [16], we designed a novel architecture for EPC based on SDON in [17], where IP gateways are replaced by bandwidth-variable optical switches (BVOS), and EPC control modules and BVOS are controlled by SDN controller with extended OpenFlow protocol.

In this paper, we first build a testbed to verify the feasibility and performance of the proposed SDON-based EPC network architecture, and then propose the optical resources allocation policy in EPC networks. It is challenging to flexibly provide suitable end-to-end connection for elastic optical networks, especially under the complicated condition in mobile networks. For example, huge amount of user traffic need to be transported in mobile networks frequently due to the tidal phenomenon of traffic distribution in access networks [18]. Fulfilling the dynamic traffic requirements is critical to guarantee the performance of the networks. In this article, we design dynamic optical resource allocation algorithm to improve the performance for two different use cases of connection provision in SDON-based EPC, i.e. user’s handover and traffic overload. Load balancing policy can reduce the blocking ratio in optical networks [19]. It has been widely adopted to improve the spectrum utilization per link in elastic optical networks [20–22]. Different from previous link-based load balancing policy, we propose a node-based load balancing algorithm in this paper, i.e. Optical Switch Load Balancing (OSLB) algorithm. Simulation results show that OSLB algorithm can reduce the blocking probability and improve the spectrum resource utilization in mobile core networks.

The rest of this paper is organized as follows. Section 2 introduces functional architecture of SDON-based EPC network and procedure of lightpath establishment. Section 3 describes two EPC-lightpath (EPCL) adjustment scenarios and the proposed OSLB algorithm. In section 4, the establishment and adjustment of EPCLs handled by OSLB algorithm are demonstrated on a SDON-based EPC testbed with extended OpenFlow protocol. The OSLB algorithm is further evaluated via simulations in section 5. Section 6 concludes the paper.

2. Functional architecture and system design

An eligible and adaptable mobile core architecture should provide on-demand connections and allocate heterogeneous resources for connection requests. It should provide users with specified application functions such as assigning private tunnels to guarantee the end-to-end quality-of-service (QoS) [23], which can be realized by the software components in control plane. The function of entities in data plane (SGW and PGW) is divided into different functional blocks in this architecture: uplifting the control components as parts of SDN controller; deploying OpenFlow agent which is in charge of optical switches.

2.1 Components of the architecture

According to the aforementioned prerequisites, this paper proposes a SDON-based EPC network architecture shown in Fig. 1, which is an evolution of current architecture of EPC shown in [4]. The most noticeable change happens in data plane, where IP gateway is replaced by optical switch. We classify optical switches into serving optical switch (SOS) and PDN optical switch (POS) according to their role in data plane. The functions of other parts of EPC such as entities and protocol are kept compatible with LTE standard to guarantee the efficiency of signaling and upgrade. The control plane is divided into three control layers as follows.

Fig. 1 SDON-based EPC architecture.

Download Full Size | PDF

EPC Connectivity Control Layer is comprised by MME, SOS controller (SOS-C) and POS controller (POS-C). It works similarly with control compounds in SDN-based EPC network [24]. MME is still in charge of access control, mobility control and user’s session setup. As optical switches in data plane become physically equivalent, the role of BVOS is defined by the two components, i.e. SOS-C and POS-C, which are control functional compounds decoupled from the optical switches. They are responsible for deciding the setup of lightpath and IP tunnel. SOS-C is in charge of the selection and control of SOS, which connects with border gateway in backhaul. POS-C is in charge of selection and control of POS, which connects with border gateway of services provider’s networks. User’s IP information is received via General Packet Radio Service (GPRS) tunneling protocol (GTP) from MME, then is transformed in the format that is used between S/POS-C and the northern interface in SDN controller.

Core Control Layer is implemented with SDN controller and is the main part of control plane, which considers both IP layer and optical layer. Connecting with S/POS-C via northbound interface of the controller, a route control and daemon (RCD) [25] translates GTP information here, then compute an IP route for each connection request. Connection provisioning and optical resource allocation algorithms are embedded into the controller core, a similar design of this function is elaborated in [11]. All the information of IP routes and lightpaths are stored in a traffic engine database (TED).

Optical Resource Control Layer is a translation layer, which controls hardware in data plane. OpenFlow agent (OFA) is allocated to take control of optical devices via hardware-specified interface and translate the configuration commands to OpenFlow protocol, and vice versa. An extended OpenFlow southbound interface is designed for the OpenFlow signaling between OFA and the controller.

Data plane is built on flexible grid optical networks, where transponders and optical switches can be sliceable, and modulation format can be adjusted dynamically. Optical switches are equipped with bandwidth-variable wavelength selected switching (BV-WSS) [26]. The optical signal can be modulated to sub-carriers with different modulation formats, such as BPSK, QPSK, 8QAM, 16QAM, and 64QAM. Combined with SDN/OpenFlow, a Spectrum-on-Demand (SoD) scheme proposed in [27] can support flexible lightpath establishment for EPC, which is referred as EPCL in this paper.

*2.2**Signaling procedure and EPCL establishment*

Provision of IP connection for mobile user is an important task of EPC networks. The destination IP address of users’ connection request in EPC networks is service providers’ address. After receiving GTP messages from the standard EPC components, the SDN controller firstly computes an IP route, then runs routing algorithm to calculate a lightpath on-demand, which passes through SOS, intermediate optical switches (IOS), and POS. The traffic in data plane is abstracted as flow entries which are presented by the packet’s characteristic (IP address or TCP/UDP port) and the actions of BVOS. Tunneling endpoint ID (TEID) [28,29] is added in OpenFlow protocol to tag each EPCL with spectrum size and optical port assigned to it. The signaling procedures to establish an EPCL have been shown in Fig. 2. Note that after SDN controller computes EPCL for a connection request, a failure message may be generated when there is no satisfactory lightpath available.

Fig. 2 Signaling procedure of EPCL establishment.
Download Full Size | PDF

3. Dynamic optical resource allocation in SDON-based EPC

In this section, we analyze how to adjust EPCL in the SDON-based EPC networks, considering the traffic dynamics in EPC. It is assumed that data rate in electrical layer can be matched with suitable optical granularity in optical domain, which has been discussed in [16] and demonstrated in [14]. Considering the traffic asymmetry of upstream and downstream traffic in EPC, we focus on illustrating the downstream traffic in this section.

3.1 EPCL adjustment

When a connection request arrives, the least cost route will be computed by the controller according to the current network state, then a unique TEID is tagged on the connection. The optical resources are allocated to establish an EPCL for this request. Specifically, the number of spectrum slices (i.e. sub-carrier in flexible grid optical networks) is first calculated according to bandwidth requested the connection, then optical switches are selected and the route along optical fibers is calculated for this EPCL, which is based on the traffic load of optical switches, the available spectrum slices in a fiber link. Note that optical switches are assumed to only serve for mobile users in a specified tracking area (TA), which is an area where a user can freely move around without notifying the MME. The data plane can be divided into different domains according to specified TAs. So when a user traverses different TAs, a handover is trigged, which leads to the adjustment of EPCL. Besides handover, EPCL adjustment can also be triggered by overload scenario, as shown in Fig. 3.

Fig. 3 Scenarios of EPCL adjustment.
Download Full Size | PDF

When a mobile user enters a new TA, its connection is moved from the source TA to the target TA, so its EPCL in data plane is required to be adjusted and goes through the TA where the user resides. Therefore, a new SOS is needed to serve the adjustment request of EPCL, and a new lightpath must be established while the former lightpath must be torn down. In this case, the source or destination node of request will be changed. As the Fig. 3(a) shows, when a user moves from TA 1 to TA 2, the new route (green dashed line) will be computed for this request and then an EPCL will be established. Different policies can be adopted to minimize the cost of the new EPCL, such as the complete, incremental, and curtailing policies proposed in [30].

Overload in EPC may occur for different reasons, such as temporary traffic surges due to “flash crowd” effect, and poor routing. We focus on the traffic load of optical switches in data plane where POS may face overload problem. When it happens, EPCLs should be adjusted to ease the traffic overload of optical switches. As depicted in Fig. 3 (b), the red centered POS cannot support the overload. Firstly, according to the TEID of the request, a POS with lighter traffic load will serve for this request (depicted as TEID adjustment [31]), then a new EPCL (indicated by dashed green lines) will be established.

3.2 Optical switch load balancing (OSLB) algorithm

Load balancing policy is an effective way to reduce the probability of blocking and has been widely adapted in optical networks. Different from link-based load balancing algorithm introduced in section 1, a novel node-based load balancing algorithm is designed in this section. First, the mobile core network is modeled as an undirected graph $G = (N, L)$ . $N$ denotes the set of nodes, and each node n in $N$ represents an optical switch with a TA identity, i.e. $T A_{n}$ . $T L_{n}$ is defined as the traffic load of optical switch n (defined in Eq. (1)). $L$ denotes the set of bidirectional fiber links, and $L_{n}$ denotes the adjacent links of the optical switch $n$ . $S$ is the set of spectrum slices on each link. It is assumed that each fiber link is denoted as $l$ , and has capacity denoted as $c_{l}$ . The throughput of an optical switch n is denoted as $C_{n}$ . Connection request is directional [32] and denoted as $r (s, d, b)$ . $s$ and $d$ denote the source node and destination node which are the corresponding SOS and POS. $b$ denotes bandwidth requirement. $b_{g u a r d}$ denotes the guard-band for the connection. In this paper, $b_{g u a r d}$ is same for each connection. The set of all connection requests is denoted by $R$ . The set of bandwidth requirements is denoted by $B$ .

T L_{n} = \sum_{l_{i} \in L_{n}} \sum_{r_{i} \in R} (b_{r_{i}, l_{j}} + 2 \times b_{g u a r d}) \forall r_{i}, l_{j}

where $r_{i}$ denotes the ith connection request, and $l_{j}$ denotes the jth link. $b_{r_{i}, l_{j}}$ denotes the bandwidth on link $l_{j}$ occupied by connection $r_{i}$ .
Combined with handover and load balancing of optical switches, the adjustment can be generalized into three cases:
1) A request $r$ needs to adjust the SOS caused by handover, i.e. $s \to s^{'}$ , and POS does not need adjustment;
2) A request $r$ needs to adjust the POS caused by overload, i.e. $d \to d^{'}$ , and SOS does not need any adjustment;
3) A request $r$ needs to adjust the SOS caused by handover, i.e. $s \to s^{'}$ , and the POS needs to be adjusted simultaneously by overload.
Three cases are shown in Fig. 4. When handover case occurs, a new route for $r$ is needed as shown in Fig. 4(a). The source of adjusted connection request $r^{'}$ should be changed from $s$ in tracking area $T A$ to $s^{'}$ in new a tracking area $T A^{'}$ . When overload case happens, the optical switches should be selected considering the distance in terms of the number of hops, for example two purple arrowed dash lines as shown in Fig. 4(b), except for traffic load. In this case, minimizing the average hop is equivalent to maximizing the total network throughput under balanced flows through the adjusted lightpaths.

Fig. 4 Cases of EPCL adjustment, (a) caused by handover, (b) caused by overload, (c) caused by both handover and overload where find two possible EPCL for adjustment.
Download Full Size | PDF
During the procedure of EPCL establishment, spectrum contiguity and continuity (SCC) constraints in elastic optical networks must be satisfied [26]. Considering the guard-band [33], the bandwidth of all requests is under the constraint
$\sum_{r_{i} \in R, l_{j} \in r_{i}} (b_{r_{i}, l_{j}} + 2 \times b_{g u a r d}) \leq c_{l_{j}} \forall r_{i}, l_{j}$
All the requests in each adjacent links of the optical switch $n$ is considered. As the load balancing is based on the traffic load of nodes, the overload occurs when combined with handover. EPCL adjustment should be under the constraints:
$T L_{n} > C_{n} \forall n \in N$ $\sum_{r_{i}^{'} \in R, l_{j} \in r_{i}} (b_{r_{i}^{'}, l_{j}} + 2 \times b_{g u a r d}) \leq c_{l_{j}} \forall r_{i}^{'}, l_{j}$ $T L_{n^{'}}^{} < C_{n^{'}} \forall n^{'} \in N_{}$
It is assumed that $h_{n}$ denotes the number of lightpaths passing through node $n$ , then the weight $w$ of a node $n$ can be computed as:
$w_{n} = \frac{T L_{n}}{C_{n}} \times h_{n}$
After adapting this policy, a shortest-path algorithm is used to compute a new route for the adjusted EPCL. In this paper, we do not consider how to use the established lightpaths. Based on the analysis above, OSLB algorithm is designed as follows. Some parameters used in the algorithm are defined first. $T L_{P O S}$ is the traffic load of POS. $C_{P O S}$ denotes the throughput of POS. $N_{P O S}$ is the set of POSs.
Algorithm: Optical Switch Load Balancing (OSLB) Algorithm
Initialization:
Construct the Graph according $G = (N, L)$ to initialize network state;
When $r (s, d, b)$ arrives, do
1. Path = Dijkstra ( $s$ , $d$ );
2. if $P A T H = N U L L$ , then
3. block ;
4. else
5. if $T L_{P O S} > C_{P O S}$ after adding,r then
6. for all $n \in N_{P O S}$ , do
7. get $n$ with the minimum; $T L$
8. $d \to d^{'} (n)$ ;
9. end for;
10. Path = Dijkstra ( $s$ , $d^{'}$ );
11. if $P A T H = N U L L$ , then
12. block r;
13. else
14. if enough slices for b and following SCC, then
15. establish EPCL between $n_{s o s}$ and;
16. else
17. block ;
18. end if;
19. end if;
20. else
21. if enough slices forand following SCC, then
22. establish EPCL between and ;
23. else
24. block ;
25. end if;
26. end if;
27. end if;
28. update $w_{n}$ ;
When $r (s, d, b)$ handovers as $r^{'} (s^{'}, d, b)$ , do
1. repeat step 1 to 28 with $r \to r^{'}$ .
As described above, lines 1 to 28 establish the lightpath for connection request . Line 1 runs the shortest-path algorithm, and the time complexity is ${| N |}^{2}$ . Lines 6 to 19 establish EPCL with changing POS, and lines 21 to 25 establish EPCL without changing POS. The worst time complexity occurs from lines 6 to 19, which is $| N_{P O S} | + {| N |}^{2} + | L | \times | S |$ . Line 28 updates the weight of each node, and the time complexity is $| N |$ . When handover case occurs, the similar procedure will be conducted with the source node changing from $s$ to $s^{'}$ . So the time complexity is same with step 1 to 28. Thus, the total time complexity of OSLB algorithm is $O ({| N |}^{2} + | L | \times | S |)$ , which is polynomial.
4. Protocol extension and experimental demonstration
Protocol extension solution for SDON-based EPC network architecture is given in this section, and then we validate the extended protocol and latency performance of establishment and adjustment of EPCLs on the testbed.
4.1 Protocol extension
To realize the SDON-based EPC network, OpenFlow protocol is needed with two extensions: (1) GTP message from the LTE network should be sent to the controller; (2) messages of configuring optical components should be added. Several solutions of OFP extensions have been given in [34,35]. In our implementation, the PACKET_IN message is extended with new fields for GTP. As a UDP based protocol, GTP tunnel is established via both layer 3 and layer 4 information, including TEID, GTP source and destination IP addresses. We mainly reserved IP address and UDP ports of GTP which could be directly translated as configuration of optical switches in data plane. Therefore, GTP tunnel is physically realized by lightpath with the same TEID. The actions of optical switches in the data plane are conducted according to the GTP message type (e.g. we defined two types: connection-establish and connection-adjust in this demonstration).
The dynamic status of lightpaths is managed by the controller and does not need to be synchronized via OFP. The match field used in the FLOW_MOD message is extended by adding new match type of input and output optical ports. To support packet switching in EPC, a flow entry containing match fields is used for “match and forwarding” when a packet arrives. For SDON, match is used to record the configuration of an optical switch, which realizes the action field depicted in section 2. We follow the works of [14] where the FLOW_MOD extensions is presented. In the data plane, lightpaths are assigned with the same datapath ID (DPID) according to TEID, and all the optical switches in the data path execute configuration actions correspondingly. To measure the latency of EPCL establishment, we also add an EPCL_ESTABLISH_ACK message which contains the result of EPCL establishment or adjustment. It is generated by optical switch’s OFA and sent to the SDN controller.
4.2 Demonstration setup and results
Figure 5 depicts the main components of the experiment testbed. In the control plane, virtual machines with Fedora V11.0 are running on IBM servers, where we realize the standard EPC components via open-source nwEPC, OpenDayLight controller (ver. Hydrogen) [36] and Open vSwitch (OVS) used as OpenFlow Agent (OFA). Both nwEPC and OVS are deployed in the same virtual machine. We use nwEPC to emulate MME with GTP control interface and GTP message is conveyed by internal API which is realized via socket interface. In EPC connectivity layer, we extended OpenFlow-Plugin of OpenDayLight to realize S/POS-C. PACKET_IN message is resolved here, then sent to OpenDayLight controller via JavaScript Object Notation (JSON) API. In core control layer, we mainly extend OpenDayLight controller with RCD to maintain the state of connection requests (e.g. duration and handover). In optical resource control layer, the OFAs are modified for all optical nodes with unified optical extension of the OFP described in section 4.1. The global state of network in this demonstration can be viewed via management system.

Fig. 5 (a) Experimental demonstration deployment diagram, (b) and (c) OFP message trace of EPCL establishment and adjustment.
Download Full Size | PDF
Data plane is equipped with 4 commercial 40 wavelength ROADMs with 10GbE add and drop ports to form a four-node mesh network. Some modulation formats, such as BPSK, RZ-DQPSK, and DP-QPSK are supported by each node. Each ROADM node is equipped with an Open Switch acting as an agent, which communicates with the OpenDaylight controller through extended OpenFlow protocol. We selected two nodes as SOSs and other two nodes as POSs. The deployment of these nodes is shown in Fig. 5(a).
Because there are only two SOS nodes and POS nodes in the testbed, and MME is also emulated by an open-source software, i.e. nwEPC, so we cannot conduct batch tests of OSLB algorithm. The procedure and latency performance of EPCL establishment and adjustment are mainly tested and measured with SDON-based EPC testbed in this section. The large-scale validation of OSLB algorithm will be conducted through simulation in section 5. First, a connection request from MME is invoked by nwEPC via GTP message, and OFA agent of MME generates EPC_PACKET_IN message to convey a bandwidth request with TEID. The SDN controller analyzes the request and queries the TEID to compute an IP route, which is realized according to the GTP message type and IP address in the message. OSLB algorithm has been deployed in SDN controller. Then, an EPCL is assigned for the connection according to OSLB algorithm and the controller generates FLOW_MOD messages to establish the lightpath via southbound interface. Then we emulate the overload of one POS to demonstrate EPCL adjustment. We manually configure high traffic load of one link (the yellow one in the management system in Fig. 5(a)), and a new EPCL is established via OSLB algorithm. The procedures of establishment and adjustment of EPCL can be found in the OFP message trace shown in Fig. 5(b) and 5(c). We can see that the adjusted EPCL has the same TEID with the original one shown in Fig. 5(c).
The establishment latency varies with the hardware response time. As shown in Fig. 5(b) and 5(c), the interval between EPC_PACKET_IN message and the first EPCL_FLOW_MOD is the latency of control plane, which is about 273.7ms. The latency of EPCL establishment is the interval between EPC_PACKET_IN and the last EPCL_ESTABLISH_ACK, which is about 316.3ms. The latency of EPCL adjustment is the interval between EPC_PACKET_IN and the last EPCL_ESTABLISH_ACK in different GTP message types, which is about 322.7ms.
From the protocol validation and latency performance of EPCL establishment and adjustment on the testbed, we can see that SDON-based EPC architecture can effectively handle the handover and overload cases in mobile core networks. Especially, the latency of EPCL establishment and adjustment are both about 300ms, which can satisfy the requirement of Non-GBR end-to-end delay [16]. On the other hand, we can easily see that huge bandwidth lightpaths can be provided with SDON-based EPC networks.
5. Simulation results
To further evaluate the performance of the proposed OSLB algorithm, a 24-node USNET topology is employed in this study. As shown in Fig. 6, we classify the nodes in the topology with different colors as different roles: SOSs in green, POSs in yellow and IOSs in white. To simulate the scenario of handover, we divided this topology into three TAs. If a source node of a connection changes to another node in a different TA, handover is to occur. All nodes are considered as bandwidth-variable optical switches but without wavelength-conversion capability. It is assumed that there are enough transponders equipped at each node. Blocking happens only when there are not enough spectrum slices. The maximum transmission capacity of a transponder is set to be 100Gbps. Each link in this simulation is bidirectional (with two unidirectional fibers) and each fiber has total 128 spectrum slices with granularity of 12.5GHz. The guard band is assumed to be 25GHz which is two spectrum slices. The data-rate-to-bandwidth ratio, which depends on the modulation format adopted by each spectrum slot is first assumed to be 2 bits/s/Hz. Lightpaths are established with SOS as a source node and POS as a destination node. There are three types of connection requests: 10Gbps, 40Gbps, and 100 Gbps, with their proportion being 10:4:1. While we set duration for each request and set the timing when the handover happens. The connection request is assumed to follow a Poisson process and the handover timing of a request follow a normal distribution in the duration (the SOS of the request is decided randomly in the neighbor TAs).

Fig. 6 24-node USNET divided into three TAs.
Download Full Size | PDF
We compare different performance metrics of OSLB algorithm with two optical resource allocation algorithms, i.e. MinHops and Link-based Load Balancing (LLB). MinHops algorithm minimizes the newly-established lightpaths and minimizes the average hops of lightpath in the network by reducing the number of hops per lightpath [19]. LLB algorithm can reduce the traffic load per fiber link [21]. To implement LLB in optical network, the new connection should be routed through lightly-loaded fiber links, then the load-balanced routing can achieve lower blocking ratio in optical networks.
To evaluate the performance of different algorithms, three important metrics are employed in the simulation: bandwidth blocking ratio (BBR), traffic load distribution (TLD) and resource utilization ratio (RUR). Once a connection request blocks whether before or after handover, we assume this request is blocked.
Bandwidth blocking ratio (BBR) represents the percentage of blocked traffic over the total traffic of all traffic requests.
Traffic load distribution (TLD). Traffic load ratio (TLR) of a node is first defined as the ratio between the traffic and throughput of a node when the state of network is stable (i.e. when and after 80000 requests have arrived with total 100000 requests in this simulation). Traffic load and capacity of optical switch are defined in section 3. Let $α_{n}$ denote the TLR of node $n$ , which is defined as:
$α_{n} = \frac{2 \times \sum_{l_{i} \in L_{n}} \sum_{r_{i} \in R} (b_{r_{i}, l_{j}} + 2 \times b_{g u a r d})}{C_{n}} \forall r_{i}, l_{j}, n \in N$
TLD is defined as the number of nodes with the same range of TLR. As load balancing policies can achieve balanced traffic load by allocating the suitable link resources, the global traffic distribution can be represented by TLD.
Resource utilization ratio (RUR) represents how efficiently connections are routed. It can be computed as the average of carried traffic divided by the total link capacity in the network (i.e. maximum number of allocated spectrum slices is 128 in this simulation) over the entire simulation. Let $β$ denotes the RUR, then $β$ can be computed as:
$β = (\sum_{R} b_{r}) / (\sum_{L} c_{l} \times 128) \forall r \in R, l \in L$ where $b_{r}$ is the connection bandwidth of a request $r$ , $c_{l}$ represents the capacity of a spectrum slice in link $l$ . In the following parts of this section, different performance metrics of three algorithms are shown.
(1) Bandwidth blocking ratio vs. traffic load: Fig. 7 shows the bandwidth blocking ratio under various traffic loads. We observe that MinHops achieves the highest BBR. The reason is that although MinHops adopts the shortest path algorithm to establish more lightpaths, more connection requests will be blocked for no spectrum available in the routes.

Fig. 7 Bandwidth blocking ratio.
Download Full Size | PDF
We also observe that OSLB achieves lower BBR than LLB by 3%. This is because LLB always tries to route an EPCL under high traffic load with increasing the hops. OSLB achieves load balancing only according to adjacent links and the shortest path algorithm is still adopted to adjust EPCL. Therefore, less spectrum resource of links would be occupied in OSLB algorithm. However, as the traffic load increases, more requests would arrive in one time, LLB cannot allocate spectrum resource for requests. We can observe that as the traffic load increase, BBR achieved by LLB and OSLB are converging.
(2) Traffic load distributions for POSs: In the simulation, we count TLD for POSs. Figure 8 shows the statistics of number of POSs with different TLR ranges. Simulation have been conducted for twenty time to compute the average value of the number of POSs. As no POS in the simulation has TLR below 20%, we do not show this zone. We observe that more POSs in OSLB and LLB have TLR concentrated in the range from 40% to 80%. OSLB achieves a TLD with similar number in TLR ranges from 40% to 60% and from 60% to 80%. And OSLB and LLB achieve less TLD than MinHops in the TLR range from 80% to 100%. The TLD results verify that load balancing policies achieve better performance than MinHops, and OSLB can achieve the best performance among the three algorithms.

Fig. 8 Traffic load distribution.
Download Full Size | PDF
(3) Resource utilization ratio vs. average hops per EPCL: Fig. 9 shows the average hops of EPCL under different algorithms. The average hops of EPCL under three algorithms are slightly different as the increase of traffic load. This is because the configuration of the topology in the simulation [Fig. 6] constrains the number of hops of successfully established EPCL. As we expect, MinHops achieves the minimum hops per EPCL, because no balancing strategy is adopted in this case, and each EPCL is established according to the shortest path algorithm. While the number of hops of LLB algorithm is the biggest, because it will try to find lightpath with more hops, which may introduce more 3-hop EPCL in this case. OSLB achieves a performance between two other algorithms, which is with average hops of 1.55. Both OSLB and LLB show an increasing tendency, this is because as the traffic load increases, more spectrum resource in different links is needed to realize the balancing policy in this simulation.

Fig. 9 Average hops of EPCL under different policies.
Download Full Size | PDF
RUR is tightly related with BBR. Figure 10 shows the RUR. We observe that OSLB achieves the highest RUR due to the low BBR. And we can see that LLB achieves higher RUR than LLB. Because although MinHops takes fewer hops, the BBR of LLB is lower than MinHops. Therefore, there is a tradeoff between BBR and hops of EPCL in this simulation. As the increase of traffic load, OSLB can achieve more 10% in RUR than MinHops and more 6% than LLB when traffic load is larger than 500 Erlang.

Fig. 10 Resource utilization ratio.
Download Full Size | PDF
6. Conclusion
To meet the challenge of traffic increase and requirement of dynamic control in 5G communications, we design a SDON-based EPC architecture for mobile core networks. Based on the novel architecture, OSLB algorithm is proposed, which makes optical resource allocation policy adaptable to the environment of mobile networks where handover and overload are two notable issues. To verify this architecture, we extend PACKET_IN message of OpenFlow protocol for EPC request and build a SDON-based EPC testbed. Then, we study the optical resource allocation for users’ dynamic connection requests with different policies. Simulation results show that OSLB can reduce the blocking probability of connections and improve the resource utilization ratio in mobile core networks compared with two other algorithms, i.e. LLB and MinHops.
Funding
Parts of this work has appeared in the proceeding of 2015 European Conference on Optical Communication (ECOC), Valencia, Spain, in September, 2015. This work has been supported in part by National Natural Science Foundation of China (NSFC) project (61571058, 61271189), and Open Fund of State Key Laboratory of Information Photonics and Optical Communication (BUPT, IPOC2014ZZ03).
References and links
1. J. G. Andrews, S. Buzzi, W. Choi, S. V. Hanly, A. Lozano, A. C. K. Soong, and J. C. Zhang, “What will 5G be?” IEEE J. Sel. Areas Comm. 32(6), 1065–1082 (2014). [CrossRef]
2. F. Boccardi, R. W. Heath Jr, A. Lozano, T. L. Marzetta, and P. Popovski, “Five disruptive technology directions for 5G,” IEEE Commun. Mag. 52(2), 74–80 (2014). [CrossRef]
3. S. Chen and J. Zhao, “The requirements, challenges, and technologies for 5G of terrestrial mobile telecommunication,” IEEE Commun. Mag. 52(5), 36–43 (2014). [CrossRef]
4. 3GPP Tech. Spec. 23.401, “General packet radio service (GPRS) enhancements for evolved universal terrestrial radio access network (E-UTRAN) access,” v8.4.1.
5. T. Tanaka, A. Hirano, and M. Jinno, “Impact of multi-flow transponder on equipment requirement in IP over elastic optical networks,” in Proceedings of ECOC2014, London, UK, Sept. 2013, paper We.1.E.3.
6. B. Skubic, G. Bottari, A. Rostami, F. Cavaliere, and P. Ohlen, “Rethinking optical transport to pave the way for 5G and the networked society (Invited Paper),” J. Lightwave Technol. 33(5), 1084–1091 (2015). [CrossRef]
7. F. Musumeci, C. Bellanzon, N. Carapellese, M. Tornatore, A. Pattavina, and S. Gosselin, “Optimal BBU placement for 5G C-RAN deployment over WDM aggregation networks,” J. Lightwave Technol. 34(8), 1963–1970 (2016). [CrossRef]
8. J. Zhang, Y. Ji, J. Zhang, R. Gu, Y. Zhao, S. Liu, K. Xu, M. Song, H. Li, and X. Wang, “Baseband unit cloud interconnection enabled by flexible grid optical networks with software defined elasticity,” IEEE Commun. Mag. 53(9), 90–98 (2015). [CrossRef]
9. R. Muñoz, J. Mangues-Bafalluy, R. Vilalta, C. Verikoukis, J. Alonso-Zarate, N. Bartzoudis, A. Georgiadis, M. Payaró, A. Perez-Neira, R. Casellas, R. Martínez, J. Nunez-Martínez, M. Esteso, D. Pubill, O. Font-Bach, P. Henarejos, J. Serra, and F. Vazquez-Gallego, “The CTTC 5G end-to-end experimental platform: integrating heterogeneous wireless/optical networks, distributed cloud, and IoT devices,” IEEE Veh. Technol. Mag. 11(1), 50–63 (2016). [CrossRef]
10. A. Tzanakaki, M. P. Anastasopoulos, G. S. Zervas, B. R. Rofoee, R. Nejabati, and D. Simeonidou, “Virtualization of heterogeneous wireless-optical network and IT infrastructures in support of cloud and mobile cloud services,” IEEE Commun. Mag. 51(8), 155–161 (2013). [CrossRef]
11. “Software-defined networking: The new norm for networks,” ONF white paper, Apr. 13, 2012.
12. P. N. Ji, “Software defined optical network,” Optical Communications and Networks (ICOCN),201211th International Conference on, Pattaya, Thailand, Nov.2012. [CrossRef]
13. J. Zhang, J. Zhang, Y. Zhao, H. Yang, X. Yu, L. Wang, and X. Fu, “Experimental demonstration of OpenFlow-based control plane for elastic lightpath provisioning in Flexi-Grid optical networks,” Opt. Express 21(2), 1364–1373 (2013). [CrossRef] [PubMed]
14. Y. Yu, J. Zhang, Y. Zhao, Y. Lin, J. Han, H. Zheng, Y. Cui, M. Xiao, H. Li, Y. Peng, Y. Ji, and H. Yang, “Field demonstration of multi-domain software-defined transport networking with multi-controller collaboration for data center interconnection [Invited],” J. Opt. Commun. Netw. 7(2), A301–A308 (2015). [CrossRef]
15. H. Chen, J. Zhang, Y. Zhao, J. Deng, W. Wang, R. He, X. Yu, Y. Ji, H. Zheng, Y. Lin, and H. Yang, “Experimental demonstration of datacenter resources integrated provisioning over multi-domain software defined optical networks (Invited Paper),” J. Lightwave Technol. 33(8), 1515–1521 (2015). [CrossRef]
16. 3GPP TS 23.203, “Technical specification group services and system aspects; policy and charging control architecture (Release10),” v10.6.0.
17. Z. Chen, Y. Zhao, H. Yang, G. Zhang, and J. Zhang, “Optical resource allocation for dynamic traffic in mobile core networks based on software defined optical networks,” in Proceedings of ECOC2015, Valencia, Spain, Sep. 2015, paper 0849.
18. J. Ding, Y. Li, and D. Jin, “Characterizing the phenomenon of traffic tide for large-scale mobile cellular data networks,” 2015 IEEE Conference on Computer Communications Workshops (INFOCOM WKSHPS), Hong Kong, China, Apr. 2015, pp.45–46. [CrossRef]
19. L. Ruan, H. Luo, and C. Liu, “A dynamic routing algorithm with load balancing heuristics for restorable connections in WDM networks,” IEEE J. Sel. Areas Comm. 22(9), 1823–1829 (2004). [CrossRef]
20. S. Zhang, C. Martel, and B. Mukherjee, “Dynamic traffic grooming in elastic optical networks,” IEEE J. Sel. Areas Comm. 31(1), 4–12 (2013). [CrossRef]
21. B. Chen, J. Zhang, Y. Zhao, C. Lv, W. Zhang, S. Huang, X. Zhang, and W. Gu, “Multi-link failure restoration with dynamic load balancing in spectrum-elastic optical path networks,” Opt. Fiber Technol. 18(1), 21–28 (2012). [CrossRef]
22. H. Lin, S. Wang, and M. Hung, “Finding routing paths for alternate routing in all-optical WDM networks,” J. Lightwave Technol. 26(11), 1432–1444 (2008). [CrossRef]
23. T. Taleb, M. Corici, C. Parada, A. Jamakovic, S. Ruffino, G. Karagiannis, and T. Magedanz, “EASE: EPC as a service to ease mobile core network deployment over Cloud,” IEEE Netw. 29(2), 78–88 (2015). [CrossRef]
24. J. Kempf, B. Johansson, S. Pettersson, H. Luning, and T. Nilsson, “Moving the evolved packet core to the cloud,” IEEE 8th International Conf. on Wireless and Mobile Computing, Networking and Communications (WiMob), Barcelona, Spain, Oct. 2012, pp. 784–791. [CrossRef]
25. J. Salim, H. Khosravi, A. Kleen, and A. Kuznetsov, “Linux Netlink as an IP Services Protocol,” RFC5349.
26. M. Jinno, H. Takara, B. Kozicki, Y. Tsukishima, Y. Sone, and S. Matsuoka, “Spectrum-efficient and scalable elastic optical path network: architecture, benefits, and enabling technologies,” IEEE Commun. Mag. 47(11), 66–73 (2009). [CrossRef]
27. M. Klinkowski, M. Ruiz, L. Velasco, D. Careglio, V. Lopez, and J. Comellas, “Elastic spectrum allocation for time-carying traffic in flex-grid optical networks,” IEEE J. Sel. Areas Comm. 31(1), 26–38 (2013). [CrossRef]
28. 3GPP Tech. Spec. 29.060, “General packet radio service (GPRS); GPRS tunneling protocol (GTP) across the Gn and Gp interface,” v11.6.0.
29. 3GPP Tech. Spec. 32.752, “Evolved packet core (EPC) network resource model (NRM) integration reference point (IRP); information service (IS) (Release 9),” v9.3.0.
30. I. Szczesniak, A. Jajszczyk, and A. Pach, “Mobile routing in elastic optical networks,” IEEE/CIC International Conference on Communications in China (ICCC), Shanghai, China, Oct.2014, pp. 107–111. [CrossRef]
31. E. Dahlman, S. Parkvall, and J. Skold, 4G-LTE/LTE-advanced for mobile broadband (Academic, 2011).
32. F. Giust, L. Cominardi, and C. J. Bernardos, “Distributed mobility management for future 5G networks: overview and analysis of existing approaches,” IEEE Commun. Mag. 53(1), 142–149 (2015). [CrossRef]
33. Y. Wang, X. Cao, Q. Hu, and Y. Pan, “Towards elastic and fine-granular bandwidth allocation in spectrum-sliced optical networks,” J. Opt. Commun. Netw. 4(11), 906–917 (2012). [CrossRef]
34. L. Liu, T. Tsuritani, I. Morita, H. Guo, and J. Wu, “Experimental validation and performance evaluation of OpenFlow-based wavelength path control in transparent optical networks,” Opt. Express 19(27), 26578–26593 (2011). [CrossRef] [PubMed]
35. M. Channegowda, R. Nejabati, and D. Simeonidou, “Software-defined optical networks technology and infrastructure: enabling software-defined optical network operations [Invited],” J. Opt. Commun. Netw. 5(10), A274–A282 (2013). [CrossRef]
36. OpenDaylight Project, “A Linux Foundation Collaborative Project,” http://www.opendaylight.org.

Dynamic optical resource allocation for mobile core networks with software defined elastic optical networking

Abstract

1. Introduction

2. Functional architecture and system design

2.1 Components of the architecture

*2.2**Signaling procedure and EPCL establishment*

3. Dynamic optical resource allocation in SDON-based EPC

3.1 EPCL adjustment

3.2 Optical switch load balancing (OSLB) algorithm

4. Protocol extension and experimental demonstration

4.1 Protocol extension

4.2 Demonstration setup and results

5. Simulation results

6. Conclusion

Funding

References and links

Cited By

Figures (10)

Equations (8)

Optics Express

Algorithm: Optical Switch Load Balancing (OSLB) Algorithm
	Initialization:
		Construct the Graph according $G = (N, L)$ to initialize network state;
	When $r (s, d, b)$ arrives, do
		1.				Path = Dijkstra ( $s$ , $d$ );
		2.				if $P A T H = N U L L$ , then
		3.				block ;
		4.				else
		5.				if $T L_{P O S} > C_{P O S}$ after adding,r then
		6.									for all $n \in N_{P O S}$ , do
		7.									get $n$ with the minimum; $T L$
		8.									$d \to d^{'} (n)$ ;
		9.					end for;
		10.					Path = Dijkstra ( $s$ , $d^{'}$ );
		11.					if $P A T H = N U L L$ , then
		12.									block r;
		13.					else
		14.								if enough slices for b and following SCC, then
		15.										establish *EPCL* between $n_{s o s}$ and;
		16.									else
		17.									block ;
		18.							end if;
		19.					end if;
		20.				else
		21.					if enough slices forand following SCC, then
		22.				establish *EPCL* between and ;
		23.					else
		24.				block ;
		25.						end if;
		26.					end if;
		27.			end if;
		28.			update $w_{n}$ ;
When $r (s, d, b)$ handovers as $r^{'} (s^{'}, d, b)$ , do
			1.	repeat step 1 to 28 with $r \to r^{'}$ .

Abstract

1. Introduction

2. Functional architecture and system design

2.1 Components of the architecture

2.2Signaling procedure and EPCL establishment

3. Dynamic optical resource allocation in SDON-based EPC

3.1 EPCL adjustment

3.2 Optical switch load balancing (OSLB) algorithm

4. Protocol extension and experimental demonstration

4.1 Protocol extension

4.2 Demonstration setup and results

5. Simulation results

6. Conclusion

Funding

References and links

Cited By

Figures (10)

Equations (8)

Optics Express

*2.2**Signaling procedure and EPCL establishment*