Professional Documents
Culture Documents
Probability
Probability
equipment can, at a
modest cost, supply TSN
services, in addition to its
best-effort services, that
Buffers allocated End-to-end latency Latency variation meet the needs of many
Figure 1. Best-effort packet service. applications that formerly
required either CBR ser-
vice or special-purpose
digital connections.
Loss probability
Probability
Probability
Buffers allocated End-to-end latency Latency variation
Probability
Probability
to “TSN flows” designated as being critical to a ant cause of packet loss is equipment failure.
real-time application. TSN networks can send multiple copies of
The essential features of TSN networks are: sequence-numbered data flows over mul-
1. Time synchronization. All network devices tiple paths, and eliminate the duplicates at
and hosts can synchronize their internal clocks to or near the destinations. There is no cycle
an accuracy between 1 µs and 10 ns. Synchroni- of failure detection and recovery, as every
zation is accomplished using some variant of the packet is duplicated and taken to or near its
IEEE 1588 Precision Time Protocol. destinations, so that one equipment failure
2. Contracts between transmitters and the net- does not cause the loss of even one packet.
work (discussed later in the section “Control and c) Flexibility. New contracts can be made and
Managemen”t): Every TSN flow is the subject of a old ones ended. As TSN flows come and go,
contract arranged between the transmitter of the the proper functioning of all TSN flows is
flow and the network. This enables TSN networks maintained at all times.
to provide: 3. Coexistence with best-effort services.
a) Bounded latency and zero congestion loss. Unless the demands of the TSN flows consume
Congestion loss, the statistical overflowing too much3 of a particular resource, such as the
of an output buffer in a network node, is the bandwidth of a particular link, TSN traffic can be
principle cause of packet loss in a best-ef- paced so that the customary best-effort Quality
fort network. By pacing the delivery of pack- of Service practices such as priority scheduling,
ets and allocating sufficient buffer space for weighted fair queuing, random early discard, etc.,
TSN flows, congestion is eliminated, and any still function in their usual manner, except that
given TSN flow can be promised a worst- the bandwidth available to these capabilities is 3 “Too much” has no fixed
case latency for delivering its packets end-to- reduced by the TSN traffic. Furthermore, data not definition. IEEE 802.1 has
end through the network. subject to a TSN contract (“non-TSN” traffic) can used 75% as a design goal
for the upper limit to the
b) Ultra-reliable packet delivery. Having elimi- use any contracted bandwidth unused by a TSN proportion of traffic that is
nated congestion loss, the next most import- flow. Deterministic.
In-order delivery except when network topol- In-order delivery except when lost packets are
In-order delivery In-order
ogy changes recovered
Response to equipment Detect failure, switch to alternate path (1:1 Detect failure, propagate new topology, alter Packet replication and elimination: no failure
failure or 1+1) local routing decisions detection or response mechanism
Primary causes of packet Random events (cosmic rays, signal/noise) Congestion: momentary overflow of output Equipment failures exceeding the number of
loss or equipment failure queue redundant paths
Packets are lost in groups whenever Packets are lost only as long as excessive equip-
Granularity of packet loss Random, relatively high probability
equipment fails or is repaired ment failures persist
Unused contracted
Lost Available to all flows, with or without contracts Available only to non-contract flows
bandwidth
Table 1. Three types of packet service.
The reader should note that item 2c above, before its intended execution time has passed.
flexibility, is the most radical change to most exist- Thus, time synchronization is required for many
ing paradigms for supporting real-time applica- applications, so it is considered a part of TSN.
tions over best-effort networks. Other alternatives However, synchronization is separable from the
to TSN (discussed later in the section “Alternatives rest of TSN, in that none of the TSN features are
to TSN”) require network simulation, prototyping, tied to any particular means for synchronizing
and/or run-time testing to determine whether a time. The Precision Time Protocol, whose root
change to the critical flows can or cannot be sup- definition is in IEEE Std 1588, is the typical means
ported. Changes can only be made to such real- for synchronizing the network’s clocks, but other
time networks when the applications are down. algorithms can be used, if they meet the accuracy
TSN networks can be built to support a dynamic requirements of the user’s application.
environment.
Zero Congestion Loss,
Use Cases for TSN
Use cases targeted by the TSN standards include:
Bounded Latency
• Professional audio and video studios. Never an Empty Buffer
• Electrical power generation and distribution. Let us assume for a moment the usual case for
• Building automation. Ethernet, that a physical link transmits packets
• Cellular radio: interconnecting the data base- serially, rather than transmitting more than one
band processing and radio frequency blocks packet simultaneously, that a packet contains the
of a cellular radio base station (fronthaul). address of its destination(s), which is used by a
• Industrial machine-to-machine: closed-cy- forwarding node to select an output port, and
cle control loops, employing measurement, that a packet is received and checked for errors
computation, and command sub-cycles. by a forwarding node before it is transmitted to
• Automotive and other vehicle applications. the next node or to the destination. Then, every
• Service provider: CBR service over best-effort packet must be stored in a buffer for some period
networking equipment. of time at each forwarding node.
In the timing model employed for TSN, each
Time Synchronization output port has some number of first-in-first-out
The natural paradigm for dedicated digital bus- (FIFO) queues associated with it for temporary
ses is, “Do what the packet says to do when you storage of packets that are to be transmitted on
receive the packet.” Timing is synchronized by that port. Each incoming packet is examined,
the clock in the controlling device in the network, and assigned to one (or more) output ports for
and the reception of the data it transmits. Trans- transmission. On each port, it is also assigned to
mission times are short and perfectly predictable. a particular queue. As each packet finishes trans-
Given that TSN uses a network, and that the mission, one or more queue selection algorithms
cost of that network depends upon the degree (discussed later in the section “Queuing Algo-
to which the timing is fixed, the natural paradigm rithms”) cooperate to select a packet from one of
for TSN is, “Do what the packet says to do at the the port’s queues to transmit next.
time the packet says to do it.” Time is then syn- There are two reasons why the rate at which
chronized separately from the data flow; the only packets enter a given queue can vary. First, any
requirement is for an upper bound on end-to- variations in the processing of the packet for for-
end latency, so that the packet can be delivered warding will cause variations in the arrival rate