- Frame-mode
bearer service,
or frame relay.
- A WAN protocol originally designed for ISDN.
- Packet-switched networks.
- A genesis in the early 60s; developed in the early 70s; used in
essentially the same format today.
- An effective technology for long-distance data communications.
- Frame relay, ATM, and the IP protocols.
- Packet switched networks offer
- flexibility, resource sharing, robustness, responsiveness
- but require essentially global information, which is impossible to know
and expensive to approximate well.
- Circuit switching works well for voice (high utilization) but not so well
for data.
- Bursty data underutilizes the circuit.
- The circuit rate-couples the two end-points.
- A packet or protocol data unit PDU.
- Contains control information in the header and possibly trailer.
- Has a maximum size of 1 to 2k bytes.
- Larger payload sizes are fragmented into appropriately sized chunks.
- Control information helps the receiver reassemble the fragments.
- Packet transmission.
- Each network node has some in-links and some out-links (which may be
the same).
- Packets come in and are sent along on next-hop based on packet's
control information and routing information stored at each node.
- Control information includes the ultimate destination, and may include
intermediate hops (source routing).
- Link congestion may buffer packets at a node.
- Different packets in the same source-destination transmission may take
different routes through the network.
- Advantages to packet switching.
- Node-to-node links can be shared among different source-destination
transmissions.
- Looser rate coupling between end-points.
- Graceful degradation under load.
- Some packets can be elevated over others (in the queue or on the
link).
- Disadvantages of packet switching.
- Extra buffering, routing, and queuing delays at each node.
- Packet arrivals can be highly variable (jitter).
- Maintaining routing information at each node increases overhead.
- Extra processing overhead at each node (hop-count, fragmentation).
- Handling packet trains.
- A packet train is a sequence of packets going from source to
destination.
- Caused by, for example, a payload much larger than the MTU.
- Can we exploit the redundant routing information implied by a packet
chain?
- Datagram services say no.
- Each packet is handled independently of the others in the chain (of
all other packets).
- Virtual circuits say yes.
- The routing information is predefined through the network (call
set-up).
- The chain follows the circuit, using a circuit identifier in the
control data.
- Buffering and queuing at each node is still possible.
- Comparisons.
- Virtual circuits amortize a single routing cost over all packets in
through the circuit.
- Effectively driving the routine costs to zero.
- Datagrams occur routing costs on each packet on each node.
- Virtual circuits have a call-setup cost; datagrams do not.
- Important for short communications.
- Virtual circuits are static, datagrams are dynamic.
- Datagrams can work around node or link outages or congestion;
virtual circuits can't (or can't easily).
- Routing (or call-setup) requires global information.
- Routing has to be adaptive to deal with changes in network
topologies and conditions.
- X.25 is an early, public packet-switched network.
- X.25 is a protocol specifying the host-psn interface at the physical,
link, and packet levels.
- The physical level is specified by X.21 or others (EIA-232).
- The link level is specified by LAPB (Link Access Protocol-Balanced) a
subset of HDLC (High-level Data Link Control).
- The packet level specifies end-to-end virtual circuits.
- These go from outside the psn to outside the psn.
- Not to be confused with packet-train virtual circuits, which go
from inside to inside (routine node to routing node).
- X.25 has a three-layer architecture: physical layer, the LAPB pipe,
and X.25 packets and virtual circuits.
- X.25 can work with datagram or virtual-circuit routing.
- Routine nodes are should exploit X.25 virtual-circuit information to
provide, for example sequenced delivery and error detection and
correction (usually retransmission).
- Frame-relay networks.
- X.25 has in-band signaling, layer 2 virtual circuit multiplexing, and
layer 2 and 3 flow and error control.
- This is a high-overhead protocol.
- Frame-relay characteristics.
- Out-of band signaling.
- Layer 2 only switching and multiplexing.
- No hop-by-hop flow or error control
- Compared to X.25, frame relay presents a simpler network interface and
reduced internal network processing.
- Frame-relay architecture.
- Frame relay has a two-layer architecture: physical and data-link.
- The link-level protocol is LAPF (Link access procedure for frame-mode
bearer services)
- LAPF functions include
- Frame delimiting, alignment, and transparency.
- Frame multiplexing and demultiplexing on the address field.
- Frame integrity checks before and after bit-stuffing (integer
sizes, size bounds).
- Transmission error detection.
- Congestion control.
- LAPF has two sub layers: the core and control.
- All components (endpoint and network node) implement the core.
- The core provides sequenced delivery and better effort
delivery.
- The control sublayer can be implemented by the endpoint, not the
network node.
- The control sublayer provides error and flow control.
- LAPF provides virtual connections, which are less protected than an
X.25 virtual circuit.
- Frame relay has a separate control virtual connection.
- Per frame processing is less for frame relay.
- Frame relay data transfer.
- Frame format.
- Flag (1), Address (2-4), Information (n), Frame check sequence (2),
Flag (1).
- LAPF frames have no control field.
- The address field is extensible to allow for 10, 17, or 24-bit
data-link connection identifiers.
- Also included are bits for address extension, congestion
notification, discard eligibility, and command-response
indicators.
- Frame control.
- Control is context dependent.
- Point-to-point connections can be simpler than ISDN.
- General procedure:
- Establish a logical point-to-point connection with a unique
data-link connection identifier (DLCI).
- Exchange information via data frames through the logical
connection.
- Release the logical connection when done.
- Connection control involves four messages: setup, connect, release,
and release complete.
- These are sent over the logical connection with dlci 0.
- A sends setup to b.
- B responds with release complete to deny or connect to accept.
- Either side may select the DCLI; the default is to let the acceptor
select.
- Connections are broken with a release from either side.
- The receiver of a release message responds with a release complete.
- Frame relay congestion control.
- Minimize frame discard, maintain negotiated qos; balance resources.
- LAPF provides minimal help for congestion control.
- The end users and the network must cooperate.
- The network monitors and reports, the end-points react.
- Congestion control techniques include discard, avoidance, and
recovery.
- Avoidance uses the B and FECN bits; this is explicit signaling.
- Discard uses the DE bit
- Congestion recovery uses implicit signaling (dropped frames).
- This is usually the responsibility of the higher-level protocols.
- Frame discard when buffers are full.
- Arbitrary discard.
- Discard to maintain the committed information rate (in b/s).
- Assigned for permanent circuits, negotiated for switched circuits.
- Three ranges: under, over, and past maximum.
- Congestion avoidance.
- The backward and forward explicit congestion notification bits.
- The bits are set as the frame travels through the network.
This page last modified on 14 November 2004.