Page 32 - ITUJournal Future and evolving technologies Volume 2 (2021), Issue 1

ITU Journal on Future and Evolving Technologies, Volume 2 (2021), Issue 1

ing resource reservation, scheduling, and other types of fore, the Resource Allocation Protocol, IEEE 802.1Qdd
con iguration via a remote management protocol, such (RAP) [15], has been proposed to apply a distributed re‑
as NETCONF [25] or RESTCONF [12]; hence, 802.1Qcc is source reservation that can exchange TSN features.
compatible with the IETF YANG/NETCONF data modeling
language. 3. HYBRID MODEL DESIGN AND FRAME‑
The IEEE 802.1Qcc standard speci ies three models for WORK CONSIDERATIONS
con iguring the Time‑Aware Shaper (TAS) gating sched‑
ules (GCL/GCE timing): a fully‑centralized model, a cen‑ This section presents our design methodology and
tralized network/distributed user model (hybrid model), main signaling framework for the centralized net‑
and a fully‑distributed con iguration model. The central‑ work/distributed user model (hybrid model). Our main
ized model greatly eases control and con iguration mes‑ goals behind designing the CNC are given by the following
sages sent across the network and can precisely con igure constraints. Additionally, the CNC can be logically or
TAS schedules due to having the complete knowledge of physically connected to the data plane with in‑band or
the network and the full capabilities of each bridge. How‑ out‑of‑band management links. With in‑band commu‑
ever the centralized model suffers from common disad‑ nication under the hybrid model, only one switch is
vantages, such as a single‑point of failure, relatively large physically connected to the CNC; thus, signaling packets
capital/operational (CapEx/OpEx) expenditures (as the between the switches and CNC affect data traf ic similar
centralized control may be super luous in a small‑scale to the distributed approach, but the CNC still functions
network [15]), and adding unnecessary complexity to a as the centralized con iguration. For the hybrid model
small‑scale network. evaluations in this study, we consider out‑of‑band com‑
Compared to the centralized network/distributed user munication, i.e., all switches are physically attached to
model (hybrid model), the fully centralized model does the CNC.
not add any bene its for the recon iguration approach 1. Our focus is mainly on stream based network adapta‑
towards enhancing the resource allocation and QoS nor tion. By this technique, luctuating streams (already
does it allow better deterministic forwarding. The main registered streams and new incoming streams) and
usage for the CUC is to take into account the application’s their requirements can be accommodated by the net‑
complex timing and computation requirements for indus‑ work dynamically based on a single remote proce‑
trial applications which is out of scope for our evaluation. dure call to the CNC.
Rather, our focus is on the recon iguration for proper re‑
source allocation. Therefore, we focus on the centralized 2. We identify and execute low requirements by pop‑
network/distributed user model (hybrid model) form of ulating the registration table. The control plane re‑
the centralized model in this study. source orchestration is purely carried out by moni‑
A fully‑distributed con iguration model (e.g., SRP over toring existing lows which have been satis ied.
MRP or RAP over LRP) may be attractive for some net‑ 3. We conduct resource allocation based on the stream
works. The fully‑distributed con iguration model avoids network resource utilization.
the added complexity and single point of failure of a cen‑
Our main assumption to accurately apply admission
tralized management entity. Moreover, Chen et al. [15]
control and, consequently, recon iguration, is that each
have argued that the centralized con iguration models
source must de ine a low in terms of total resources
can be an over‑design for real‑time applications with re‑
needed (governed by the bandwidth requirements) and
laxed latency requirements (order of magnitude of mil‑
liseconds). Chen et al. have also argued that the dis‑ the total time needed for the resource to be used (which
tributed model is more scalable. (However, studies of the in our traf ic model is termed the resource utilization
fully distributed model with RAP over LRP targeted typi‑ time). Essentially, the CNC uses this information (which
cally applications with relatively relaxed latency require‑ is tagged in the Ethernet frame header) to determine
ments.) whether a stream ( low) is admitted or rejected.
In the absence of a Centralized Network Con iguration
(CNC) node, the TSN Task Group (TG) speci ies the IEEE 3.1 Core components
802.1CS (Link‑Local Registration Protocol, LRP) [29] Our design is split into two layers, Control Plane and Data
standard for registration and distribution of application Plane, following the decoupling SDN paradigm, thereby
con iguration parameters over point‑to‑point links tar‑ inheriting the bene its of the orthogonality of the two
geting newly published TSN features. A legacy protocol, planes, as shown in Fig. 1.
such as the Stream Reservation Protocol (SRP) [1] which
is primarily used for Audio‑Video Bridging (AVB) applica‑ 3.1.1 Con iguration module
tions, is intended to serve as the main resource reserva‑
tion and admission control protocol. However, extending The con iguration module is the main component that
and porting the SRP to be utilized for bridges that support interacts with the registered lows and network com‑
TAS will not suf ice since bandwidth reservation cannot ponents. It includes the global stream registration ta‑
directly apply TAS’s time slot reservation natively. There‑ ble which records all approved streams transmitting in

16 © International Telecommunication Union, 2021