Low-latency hard real-time communication over switched ethernet

Abstract

MotivationEthernet as de£ned in the IEEE 802.3 standard is the commodity network since decades, and has undergone a number of changes in its existence. It is used for hard realtime communication already, and demanding applications continue to emerge. A typical example is factory automation, where Ethernet replaces CAN for performance and cost reasons. In the context of professional audio mastering (audio-LAN) Ethernet is experimented with: multiple nodes generate samples for hundreds of instruments in parallel and send them to a central mixer node. The process is interactively controlled and delays are expected to be less than 10ms. The bandwidth requirement for such a scenario is ten to hundred megabytes a second. Another application from the audio domain is DMIDI [30], an attempt to use Ethernet LANs for MIDI control commands. Although the bandwidth demands are moderate, the delays are expected to be a few milliseconds too.CSMA/CD Ethernet Real-time approaches using the original bus-based Ethernet basically fall in three categories: token-based medium access control protocols, time slot-based protocols and statistical approaches. Time slots and token passing techniques are used by cooperating nodes for both: to avoid collisions and to obey the limit of bandwidth allocated to the participating nodes. Intuition indicates that the use of such techniques to avoid collisions limits the achievable utilization and increases the CPU load of the nodes much more than using more relaxed forms of cooperation that only control bandwidth allocation. Related research supports our intuition on the high cost for collision avoidance by node cooperation (see Section 5).Switched Ethernet is a star-based topology providing a private collision domain to each of the ports of a switch. Collisions do not occur, thus node cooperation is needed only for bandwidth control, not any more to avoid collisions. It was our starting assumption that with £ne grained traf£c shaping as only means of node cooperation, we should be able to achieve lower guaranteed delays and higher bandwidth utilization than time-slotted and tokenpassing approaches, even though Switched Ethernet does not support policing in the switches as for example in ATM switches.Although we heard rumors on the usage of Switched Ethernet in hard real-time applications that do not rely on pure time-driven technology (as for example in MARS [12]), we are not aware of any practical analysis.In this paper, we make an attempt to close that gap and to validate our assumption as stated previously. We show how commodity Switched Ethernet technology can be used for low-latency hard real-time communication, provided the right operating system support is available: In Section 2 we £rst adapt well-established scheduling theory to our needs. In Section 3 we show how the needed traf£c shaping can be practically done using the Dresden real-time operating system (DROPS) as a basis for experimentation. In Section 4 we present detailed measurements. Section 5 surveys other work in the area of rea...

Guaranteed real-time communication in packet-switched networks with FCFS queuing

“…In contrast, the delay estimations for such components are less tight in the NC analysis [11][12][13][14].…”

“…In contrast, many existing real-time analyses in the literature are only developed for simple cases, for example, deadline being equal to period [5], a single-switch network [12] [13], switches with homogeneous bit rate ports [5] [12][13][14][15] or a fixed frame size [5].…”

“…This section relates our analysis method to NC-LH [12] [13], a real-time switched Ethernet solution based on NC. The two approaches are based on the same network architecture (switched Ethernet) and queuing mechanism (FCFS), but use different traffic models and analytical schemes to calculate the worst-case end-to-end delays [16].…”

“…However, all these NC-based solutions require modification when applied to a network with standard components such as switched Ethernet, e.g., implementing traffic shapers in the source nodes [12] [13] or supporting priorities to logical real-time channels [15], which significantly increase the cost and implementation complexity. Tight end-to-end delay bounds for FCFS sink-tree networks have been derived using NC [16].…”

“…A widely accepted analytical technique, Network Calculus (NC), enables an approach for calculating the worst-case delay for FCFS queuing [6][7][8] and has been applied on packet-switched networks [9][10][11][12][13][14]. However, all these NC-based solutions require modification when applied to a network with standard components such as switched Ethernet, e.g., implementing traffic shapers in the source nodes [12] [13] or supporting priorities to logical real-time channels [15], which significantly increase the cost and implementation complexity.…”