Ethernet 1970

For all these reasons, a number of vendor shave been motivated to offer industrial communication solutions based on Ethernet interfaces with various modifications and additions in terms of protocols to improve predictability and provide realtime guarantees. The server responds, and the network conveys the response to the client. Even if the network provides temporal guarantees, it will be imposible to provide guarantees to the client if there is no temporal bound between the reception of the request at the server and the response of the server.The network guarantees will thus beuseless. (Areview of most of the transport protocols is found in [18].) Nowadays, TCP and UDP are so widely used that most of the effort is dedicated to adequate tuning of TCP parameters or to the design of realtime protocols that are compatible with TCP.At the network layer, IP is the dominant solution. This aspect will not be addressed here because most of the proposals limit the realtime traffic to a single link, thus avoiding the transit of real time messages through routers.In the data link layer, the médium access policy [medium access control (MAC)] is an integral part of the guarantee chain. Some reuse existing networkinterface cards while being incompatible with regular Ethernet nodes.A last category preserves compatibility but can provide real time guarantees only in the absence of “foreign” (unmodified Ethernet) nodes.It is worth noting that the way thecommunication software is implemented plays a nonmarginal role inthe guarantees.

VINTAGE ETHERNET

Ethernet was developed during the1970s to emerge as a product at the beginning of the following decade.The IEEE published its first 802.3 standard in 1985. The main difference is the logical link control (LLC) sublayer, whichis included in Ethernet but left to another IEEE standard (IEEE 802.2). The corresponding field carriesthe data length in IEEE 802.3.An International Organization for Standardization (ISO) recommendation [20] has guaranteed compatibility between both solutions since 1997.Nowadays, all products are complian to IEEE 802.3—but all personal computers use the Ethernet frames, while only a few protocols, such as thesimple network management protocol (SNMP), use 802.2 frames.

In this version, all the nodes on a given link share the same transmission channel. In other words, each node can hear what any other node emits. A node that wishes to transmit listens first. If it does not hear anything for a given duration (interframe gap), it starts transmitting a packet. Whilet ransmitting, the node listens to what is emitted on the channel. If it senses a collision—which can happen when two nodes start transmitting at the same time—it stops transmitting its data. If no collision is detected before the end of the transmission of the packet, the transmission is considered successful. A node that detects a collision instantly stops transmitting its packet and emits a jamming sequence. The duration of this sequence has been calculated so that all connected node sare guaranteed to detect the collision.After such a failure, the node will prepare it self to retransmit the packet.However, instead of retransmitting immediately, which would cause a second collision if all the nodes involved in the first collision adopt the same policy, the node chooses a random number in the back off interval.

It corresponds to the máximum propagation time on the link (shared medium) and thus limits the span of the network and the number of repeaters.When the network is lightly loaded, the protocol previously described exhibits a very low waiting time before sending a packet.

VINTAGE ETHERNET PERFORMANCES

It is difficult to obtain accurate theoretical results for Ethernet. This is why studies of Ethernet performance make a number of simplifying assumptions (an infinite number of nodes, constant packet length,pending traffic always) and the results are merely approximations. Some ofthe results, however, have been confirmed by practical experiments [22].An IEEE 802.3 network is able to operate at close to 100% of its capacity when packets are long and there is a small number of nodes(100% capacity corresponds to the case in which no time is lost either in collisions or retransmissions).When the packets are short (the minimum is 64 B), efficiency drops but remains much higher than the“theoretical limit” of 37%. This 37% limit, widely found in the literature,has been obtained under simplifying assumptions, which explains the discrepancy with practice.The mean transmission delay grows quasilinearly with the packet size and the number of participating nodes.The transmission delay standard deviation also increases with the number of nodes and the packet size.To summarize, when a short transmission delay and a low jitter are preferred, it is better to use small packets.Long packets are favorable to higher network utilization.

VINTAGE ETHERNET PROBLEMS

Ethernet MAC is ordinarily fair and the waiting time low when the network is moderately loaded. Suppose two nodes, A and B, have a number of packets to transmit.Node A selects 1 as a random number in the back off interval, while node B selects 0. Node B will be successful in its first retransmission.As it still has other messages, it will try immediately to transmit the next message. This transmission will collide with the emission of node A. Node A then doubles the back off interval, where as node B uses the initial one. Node B thushas a higher chance of transmitting successfully. In the worst case, node A may fail to transmit after 16 retries, even though the network is far from being overloaded. This is more of a theoretical problema than a practical one, however, because most NICs are unable to sustain a high through put, and the node will there for a quickly loose the arbitration.

In this topology (Figure 2), each node is linked through a dedicated cable to a special element called a “hub” that acts as an N port repeater. Apart from the increase intransmission speed, the working principle of these new versions remain identical to vintage Ethernet. All nodes in a network that mustcomply with realtime guarantees must be connected through switches of this kind. Due to traffic blocking and classification, it is possible to connect “pure Ethernet” nodes on the switches without altering the temporal guarantees of the isochronous traffic. P-NET on IP and TC net add a token passing mechanism on top of IEEE802.3. If the medium is free for more than a given delay, the token is implicitly passed to next node. When the counter value is equal to the node ID, the node can transmit one packet at most, including the response from the destination and possible retries.The destination node has a máximum bound in the time to respond, which is lightly less than the duration of the silence that induces

...