Page 30 - ITUJournal Future and evolving technologies Volume 2 (2021), Issue 1
P. 30
ITU Journal on Future and Evolving Technologies, Volume 2 (2021), Issue 1
the over‑provisioning of network resources. The general velop comprehensive centralized and distributed recon‑
idea for such an allocation scheme is to control network iguration frameworks based on irm bandwidth compu‑
access in a timely and orderly fashion such that a maxi‑ tation strategies that execute at run‑time. Further, we
mum number of streams can be effectively serviced. conduct a comprehensive performance evaluation of our
Our objective therefore is to maximize the number of ad‑ two frameworks considering common packet low QoS
mitted lows (i.e., tasks or streams) in such a dynami‑ metrics for both high‑priority ST and low‑priority BE traf‑
cally changing and volatile environment whilst keeping ic.
the TSN QoS metric guarantees. In this paper, we fo‑ The proposed approach by Nayak et al. [67] exploits the
cus on the IEEE 802.1Qbv [2] enhancements and design logicalcentralizationparadigmofSDNwithreal‑timetraf‑
a recon iguration framework taking inspiration from the ic to achieve optimal scheduling and routing. Integer Lin‑
IEEE 802.1Qcc [3] standards for managing, con iguring, ear Programming (ILP) formulations were used to solve
and recon iguring a TSN network. the combined problem of routing and scheduling time
In IEEE 802.1Qbv, a TAS time slot (corresponding to a triggered traf ic. Two main proposals for routing are
GCE and also referred to as slot time) is de ined as the given, namely ) scheduling and path‑sets routing, and )
portion of the cycle time (CT, which corresponds to the scheduling and ixed‑path routing whereby the ILP for‑
GCL); TAS time slots are allocated to high‑priority ST traf‑ mulations are used to ind near optimal low to time‑slot
ic. In our model, the switch/controller computes the TAS allocations. However, the ILP does not scale well with the
time slot for all admitted streams as follows. Essentially, number of lows, does not provide schedules at runtime
as streams get registered, we keep track of the available speeds, and does not work well with dynamic low con ig‑
remaining capacity, which we set initially to the maxi‑ uration (or recon iguration). To enhance the architecture
mum available capacity on each egress port until the load proposed by Nayak et al. [67], an augmentation is pro‑
(which depends on the ST slot size and the cycle time posed in [68] that incrementally adds time sensitive lows
is negative, i.e., oversubscribed link). Such a link over‑ to the scheduler making the proposed approach recon ig‑
subscription invokes a procedure call that increases the uration capable. Additionally, Nayak et al. [65,66] provide
slot time (by a step size of 1%, or more ine‑grained in‑ an analysis and evaluation to the problem of low‑span
crements) until the remaining load is positive. This pro‑ and routing protocol (Equal Cost Multi‑Path, and Shorted
cedure is iteratively called until all registered streams and Path) on transmission scheduling. Further routing re ine‑
the new stream are appropriately registered with a suf i‑ ments have been studied in [9,48,49,66,72].
cient ST slot time to transmit all frames during a single Focusing on in‑vehicular networks, Hackel et al. [38] have
appropriately sized CT. proposed a SDN based TSN framework that performs
Our proposed TAS con iguration/recon iguration is de‑ recon iguration using the Stream Reservation Protocol
signedforthecentralized(hybrid)modelandforthefully‑ (SRP) as a means to register and allocate resources for
distributed con iguration model. In the “hybrid” model, TSN streams. The TSN with SDN is evaluated with two
the CNC is utilized for con iguration exchanges and net‑ TSN switches and two clients (a sources and sink). In
work side management, as explained in more detail in contrast, we provide extensive evaluation for larger net‑
Section 3. In the distributed approach, the GCE slot pa‑ work topologies and sources. Using OpenFlow and open‑
rameters are con igured in a distributed manner by the PowerLink, Herlich et al. [41] have provided a proof‑of‑
switches as per the distributed algorithm/procedure ex‑ concept model that highlights the advantages of TSN with
plained in Section 4. For brevity we refer to the central‑ SDN and real‑time Ethernet protocol. While the model
ized network/distributed user model (hybrid model) also shows promising advantages in theory, only a coarse‑
as the centralized model or the centralized topology. We grained evaluation was presented that, in contrast to our
refer to the fully‑distributed (decentralized) model also evaluation, does not examine stream admission rates and
as the decentralized model or the decentralized topology. TSN QoS. Focusing on remote monitoring and telemetry,
Kobzan et al. [46] have presented a solution concept and
1.2 Related work implementation of an SDN based TSN architecture using
IEEE 802.1Qcc. However, the concept is provided with‑
We irst note that general performance evaluation strate‑ out any empirical evaluation. To the best of our knowl‑
gies for TAS have been explored in [39,50,73] and we fol‑ edge, there are no prior detailed studies on a luctuating
low these strategies in our study. Raagaard et al. [51,76] volatile source or a dynamic stream resource allocation
have presented a heuristic scheduling algorithm that re‑ and admission control policy in conjunction with a net‑
con igures TAS switches according to runtime network work recon iguration policy being executed while lows
conditions. Feasible schedules are computed and for‑ are carried in a TAS time scheduled network. We provide
warded using a con iguration agent (composed of a Cen‑ a comprehensive design and evaluation of an SDN based
tralized User Con iguration (CUC) and Centralized Net‑ TSN model that bases the speci ication on the standard‑
work Con iguration (CNC)). Raagaard et al’s model places ization given by the IEEE 802.1Q standard.
emphasis on the schedule computation complexity for ap‑ Vlk et al. [87] have proposed a simple hardware enhance‑
pearing and disappearing synthetic lows in a fog comput‑ ment of a switch along with a relaxed scheduling con‑
ing platform. Complementary to this approach, we de‑ straint that increases schedulability and throughput of
14 © International Telecommunication Union, 2021