Full PON Virtulisation Supporting Multi-Tenancy Beyond 5G

Abstract

In this paper, we introduce a virtualization technique to enable fully customizable resource sharing for Passive Optical Networks. We provide a summary of the concept, economic challenges and implementation.

Full text

Full PON Virtulisation Supporting
Multi-Tenancy Beyond 5G
Nima Afraz, Frank Slyne and Marco Ruffini
CONNECT Center, Trinity College Dublin, Ireland
{nafraz, fslyne, marco.ruffini}@tcd.ie
Abstract: In this paper, we introduce a virtualization technique to enable fully customiz-
able resource sharing for Passive Optical Networks. We provide a summary of the concept,
economic challenges and implementation. © 2019 The Author(s)
OCIS codes: 060.4250 Networks; 060.4256 Networks, network optimization.
1. Introduction
The current scene on the broadband/mobile operators’ market is an oligopoly, where novelty is limited by the
market development policies of a hand-full of operators. The cost of entering this market is unaffordably high for
smaller service providers who could bring considerable revenue to the access market by introducing new services.
Sharing the last mile of access networks, which is the most CAPEX demanding part, can dramatically reduce the
required initial investment and facilitate market entrance for new operators. However, the current sharing methods,
especially in fixed access networks, operate at too a high-level (e.g., virtual unbundled local access (VULA)) where
they are not capable of providing enough control over the service provided to the customers [1]. Other proposals
exists for a low-level access, which typically translate in assigning a dedicated wavelength to a second operator.
However, besides being inefficient, they are currently hindered by the fact that multi-wavelength PON (e.g., NG-
PON2) has not been deployed anywhere due to its high cost. Therefore, we propose a new sharing technique for
Passive Optical Networks (PONs) which meets the above-mentioned methods in the halfway by providing frame-
level scheduling control for the operators while being more affordable and easier to attract new entrants [2]. In the
next section, we will elaborate on the importance of providing scheduling control for the operators. Then we will
slightly touch on the economic market challenges in multi-tenant fixed access sharing environments and finally
share some details about the test-bed implementation of the proposed sharing method.
2. The Sharing Model
Due to their worldwide deployment and efficiency of resource usage, PON networks are considered a strong
candidate for providing networks’ services to 5G networks and beyond [3]. However, the capability of the current
PON networks to support new services with stringent and diverse requirements such as latency has been a center
of focus for a large and growing body of literature.
1
Merging Engine
Splitter
ONU1
Virtual
BWMap
PHY BWMap
VNO1
… …
vDBA1
VNONVNOM
vDBAmvDBAM
ONUN
…
Virtual
BWMap
Virtual
BWMap
Fig. 1. Multi-Tenant PON Layout with vDBA
OLT
Final BwMap
InP (Auctioneer)
Merging
Capacity
Auction
Excess Demand
VNOSeller1
Excess
Supply
VNOSeller2
Excess
Supply
VNOBuyer1
ONU
smaller
ONU
ONU
Fig. 2. Multi-Tenant PON Sharing Market
However in the upstream, conventional PON scheduling schemes, called Dynamic Bandwidth Allocation
(DBA), are incapable of supporting low latency as it operates extensive and time-consuming signaling between
the OLT and the ONUs. Our proposed virtualisation of the DBA (vDBA), recently standardised in the BroadBand
Forum (BBF) TR-402 and depicted in Figure 1, allows different Virtual Network Operators (VNOs) to implement
their flavour of the DBA, providing them with the required flexibility to control the upstream scheduling, paving
the way for adoption of PONs as primary transport network solution for new bandwidth-intensive services (5G,
Virtual reality etc.). In [4] we have addressed the possibility of providing the required control to the VNOs by ded-
icating virtual and programmable instances of the DBA algorithms. Therefore, each VNO will operate a portion
of the network and their bandwidth allocation decision (referred to as bandwidth map (BwMap)) is aggregated
by the merging engine in the final BwMap is issued. The technical details of this work are further discussed in
subsection 3.

One issue that arises once the scheduling is passed to VNOs is that without proper incentives, VNOs have little
motivation in sharing any unused capacity with other VNOs, making the overall PON inefficient. Indeed, once
VNOs can fully control their slice scheduling their best strategy is to always pretend they need all the contracted
capacity, which might otherwise be re-directed to a competing VNO. To address this challenge, we have pro-
posed [5] monetization of the excess PON slices where using an auction mechanism the VNOs can trade their
excess capacity in return for monetary compensation (depicted in Figure 2). Through rigorous theoretical proofs
and market simulations in [6] we validate that the proposed market will incentivize the VNOs to trade their excess
resources in a setting where any attempt to manipulate the market will lead to worst or the same outcome. Nonethe-
less, the previous work assumes an open access architecture where a fully trusted central authority (Infrastructure
Provider (InP)) is in charge of operating the market (bookkeeping, conducting the auction, settlements, etc.). This
assumption may not be valid considering the new network ownership models. Hence, the natural progression of
this work is to study distributed means of consensus (e.g., Blockchain-based Smart Contracts) which do not rely
on a central entity.
3. Implementation
Our vDBA concept was demonstrated in [7] on test bed incorporating a physical PON, a set of emulated ONUs, a
traffic generator and a multi-access edge computing node. The physical PON was implemented on Xilinx VC709
FPGA boards, operating at symmetric 10Gb/s line rate.
Fig. 3. vDBA edge Compute stack
The PON hardware and software virtualisation architecture is shown in Fig. 3. The Multi-access Edge Com-
puting (MEC) node hosted the Merging Engine (ME) and the vDBA functions for the Virtual Network Operators
(VNOs), implemented as Virtual Network Functions. The Merging Engine is the element that blends together all
virtual bandwidth maps from the different VNOs so as to generate one physical bandwidth Map allocation) and
the SDN control plane. Due to the real-time critical nature of receiving and transmitting status report messages
from the ONUs (called DBRUs) and Bandwidth Map data, our latest implementation [8], [9] made advanced use
of the Data Plane Development Kit (DPDK) toolkit to optimize the packet transfer through the physical host to
and from the Virtual Network Functions.
References
1. M. Ruffini, Multidimensional convergence in future 5g networks, JLT, 35 (3), Feb. 2017.
2. A. Elrasad et al., Virtual dynamic bandwidth allocation enabling true PON multi-tenancy, OFC 2017, paper
M3I.3.
3. J. S. Wey et al., Passive optical networks for 5g transport: Technology and standards, JLT, July 2018.
4. A. Elrasad et al., Frame level sharing for DBA virtualization in multi-tenant PONs, ONDM 2017.
5. N. Afraz et al., DBA capacity auctions to enhance resource sharing across virtual network operators inmulti-
tenant PONs. OFC 2018, paper Th1B.3.6.
6. , A sharing platform for multi-tenant pons, JLT, 36 (23) Dec 2018.
7. F. Slyne et al., Demonstration of real time VNF implementation of OLT with virtual DBA for sliceable
multi-tenant PONs, OFC 2018, papaer Tu3D.4.
8. F. Slyne et al., Experimental demonstration of DPDK optimised VNF implementation of virtual DBA in a
multi-tenant PON ECOC 2018.
9. M. Ruffini et al., Moving the network to the cloud: The cloud central office revolution and its implications
for the optical layer. JLT 37 (7), April 2019.