Dynamic Latency GuaranteeChina MobileBeijing100053Chinaliupengyjy@chinamobile.comChina MobileBeijing100053Chinagengliang@chinamobile.com
Networking
Deterministic Networking Working GroupLatency; Dynamic; DeterministicAiming at the deterministic demand for network latency in future
vertical industry applications, this document analyzes the existing
latency control methods for data transmission, points out the possible
shortcomings, and proposes some directions for optimizing the latency
control method. .The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
"SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this
document are to be interpreted as described in RFC 2119.With the vigorous development of 5G and industrial Internet, network
technology has become more and more important to support new types of
services, such as AR/VR, V2X, industrial motion control, etc., which
have stringent requirements for latency and stability. In order to meet
the requirements of the above applications, new network technologies
such as time-sensitive network TSN, deterministic network DETNET, etc.,
have proposed corresponding technical means to provide network bearers
with deterministic latency and packet loss rate and guarantee the user's
business experience.TSN includes a set of standards developed by the IEEE 802.1 Working
Group's. The TSN task group inherited from the previous Audio/Video
Bridging working group and has expanded its applications to in-vehicle,
industrial, and mobile networks. Deterministic network (DETNET) is based
on the mechanism of TSN. The difference is that TSN is applied to the
data link layer and below. DETNET is committed to applying the method to
the IP layer to provide more reliable and stable network
transmission.This document will present some problems when applying TSN in DETNET,
and try to propose reference methods to solve the corresponding
problems.Based on time synchronization, TSN has a range of bounded latency
technologies.IEEE 802.1Qav Forwarding and Queuing Enhancements for
Time-Sensitive Streams inherited from the AVB, including priority
mapping algorithms and Credit-based Traffic Shaping algorithms. The
priority mapping algorithms is to mapping the priority to 'traffic
class', which represents whether the stream is time sensitive or not.
Credit-based Traffic Shaping algorithms provide the method to allocate
bandwidth of different streams.In IEEE 802.1Qbv, the gate control list is created according to the
actual stream and timescale. It contains the transmission sequence of
all streams, and controls whether the data stream of each priority is
sent at the current time or not. All streams will be transmitted
strictly according to the current list. More Than This, IEEE 802.1Qbv
also defines the guard band mechanism and spares part of the time to
guarantee the transmission of high priority data frames at the
beginning of the next time slice.In the preemption mechanism, high-priority frames can interrupt the
transmission of low-priority data frames unless low-priority data
frames can no longer be fragmented. This standard fully guarantees the
transmission delay of the highest priority data frame, and also
reduces the guard band in IEEE 802.1Qbv to 127 bytes. The frame
preemption mechanism changes the transmission rules of the ethernet
frame and is used in conjunction with the IEEE 802.3Qbr .In addition to these, there are also other standards to guarantee
the sequence of receiving data streams, which are fine-grained traffic
scheduling technology and the key technologies of TSN in bounded
latency.DETNET refers to the bounded latency mechanism of TSN, so it needs to
pay attention to some problems in the bounded latency mechanism. There
are several standards refers to bounded latency. Users can decide
whether to use a specific standard or not, which depends on the
requirments of network and business. Some TSN testbeds have been
established these years whose basic concept is realizing 802.1Qbv to
ensure the deterministic transmission of time sensitive stream. Though
it realized ignoring the interfere of background stream, the testbed was
too simple. In fact, networking is complicated. There will be more than
two kind of streams being transmitted. So it is not that easily to apply
those mechanisms on real networks.Because of the complicated of real networks, there may be some
situations that the preemptable data frame transmission delay is too
large or cannot be transmitted. Thoes might occur when both
Enhancements for Scheduled Traffic and Frame Preemption are
enabled.Except for the highest priority, the others may be preempted by the
time slice to wait for transmission. In the actual scenario, the
preemptable data frame is not necessarily a completely non-time
sensitive frame, so it also need to guarantee the transmission of some
preemptable frame. However, Under the current mechanism, there may be
multiple preemption to cause a very large transmission delay or no
transmission of preemptable frame, depending on the size of the
express frame and the period of the timescale. In an actual scenario,
a data frame with a Secondary high priority may also be a
time-sensitive. If it cannot be transmitted or the transmission delay
is large, the service cannot be operated.For example, there are currently two queues are transmited
following the gate control list which assuming is the following table.
In the table, T00, T01, T02... represent the order of each time slice
and switching. The "01" and "10" in the right represent whether the
two queue can be transmitted in the current period. Assuming that 0
indicates that the gate is closed and the corresponding queue cannot
be transmitted. while 1 indicates the gate is open and corresponding
queue can be transferred. Then the following two cases may occur:Case 1, The preemptable frame is interrupted many times before the
transmission is completed, which causes a high transmission delay of
the preemptable frame.Case 2, the preemptable frame cannot be transmitted after once
being interrupted.Deterministic network includes deterministic latency and
deterministic packet loss. We need to think how to apply the bounded
latency mechanism effectively.Before using the bounded latency mechanism, network manager needs
to know enough about the network and applications. For example, which
kind of stream is time sensitive? How about the frame's transceiver
frequency of thoes stream? How much bandwidth does it need? ... When
you have a clear understanding of the real-time state of the network,
you can configure a delay-limited algorithm for the network.However, the transmission state of the network is not invariable.
Some transfer table might make corresponding adjustments according to
the current network situation. So the parameters that have been
configured before should also be changed. More than this, the bounded
latency mechanism also need a feedback system to receive current
network status and adjust/reconfigure the network.The implementation of the mechanism to guarantee latency requires
sophisticated calculation, including timescale and gate control tist .
When the stream in the network becomes diverse, it will consume a lot of
computing resources to schedule each stream. Therefore, a single
transmission rule may not be able to meet the problem of multiple
streams' transmission. Worst of all, the gate control list is not
properly calculated, the network may not transmit or failure.Dynamic latency guarantee is a way of thinking based on the latency
guarantee of the whole network. that is, to dynamically adjust the
priority through the current network condition and the transmission of
data stream.In the transmission process, the priority of data is based on the
"Traffic Class" in IEEE802.1 Qav. that is, the priority of data frames
is converted into traffic class according to the mapping table. If the
data frame is preempted once, the corresponding traffic class is
increased according to certain functional rules.Functional rules can be defined as needed, for example, by
assuming:T = Time of preemptedM = Lifting coefficientand F(tm)=Increased traffic classThat is to say, with the increase of preemption times, the
preemptable frames will gradually increase their priority (the
corresponding traffic class). When it is greater than or equal to the
Traffic Class of express frames, the preemptable frames could complete
the transmission.The lifting factor M can be either a constant or a variable varying
with T, depending on the network requirements of specific business
application scenarios, which will not been discussed in detail in this
document.One of the reasons for this situation is that the prediction or
mastery of the transmission of frames in the network is not accurate,
so a feedback system is needed to tell the network to centrally
configure the system. So it could help to optimize the gate control
list to avoid the frequent occurring of this problems. The most basic
case is that once there are multiple preemption occured, the switch
need to report it to the Centralized Configuration System. It
represent that there might be some unjustified configurations need to
be reconfiguration. For example, distribute more bandwidth to the
corresponding traffic class.It should be noted that all devices in the network share the same
gate control list. However, due to the difference in time of the
transmission path, it is necessary to keep all devices in the network
"asynchronous" to execute the gate control list. For example, when the
data frame is received by the device A, it is queued to be transmited
first in the currently divided time slice. When the frame is received
by the device B, the time t1 has elapsed. So the gate control list of
device B needs to perform the time difference of t1 with the A device,
which can ensure that this frame arrives at every device with a
first-transmiting in current time slice.This draft described the existing mechanism of bounded latency and
point out some problems when using them. It also proposed some reference
methods to solve them. In the process of network evolution, there might
also be more problems need to be noticed and disscuss. TBD.TBD.Forwarding and Queuing Enhancements for Time- Sensitive
Streams (IEEE 802.1Qav)Enhancements for Scheduled TrafficCyclic Queuing and ForwardingFrame Preemption