A router that connects the customer's internal network to a service provider. This can be a stand-alone device or integrated with other access devices.
 

A packet sent by a receiving station to tell the sending station that the packet was received correctly and the sequence number of the next packet that it would expect. The sending station then knows whether to send the next packet or to resend a missing packet. A form of handshaking. 
SYN - Under normal circumstances, the application that initiates a session sends a TCP SYN synchronization packet to the receiving application. The receiver sends back a TCP SYN-ACK acknowledgment packet and then the initiator responds with an ACK acknowledgment. After this handshake, the applications are set to send and receive data.
 

• Global Definition


A hard decision applied to connections that want to send data over the network. This decision is made at the time a connection request is made, typically to an access router. Connection denial is triggered when the available bandwidth for a particular traffic class is exhausted.
There are multiple ways to enforce admission control:
1. The new connection can be denied silently with no information sent back to the requestor.
2. The new connection can be rejected with a reject notice sent back to the requestor.
3. The new connection can be accomodated with the existing connections by forcing them all to shared the available bandwidth equally.

 
• RSVP Definition

Traffic packets are allowed to flow because of some decision based on past, current and expected conditions:

• The admission decision can be based on parameters in the IP packet such as:
• simplest is the destination address - also called destination based routing
•  ToS
• IP Precedence
• current traffic load (e.g. , buffer overload calls for "call blocking" or packet dropping. Determining which packets get dropped from an overloaded queue is determined by congestion control mechanisms such as  RED.

 
• The admission decision can also be based on out-of-band signaled parameters regarding the traffic flow, and on downstream conditions.

 

A routing technique whereby multiple paths, rather than just the shortest path, between a source and a destination are utilized to route traffic. One of the objectives of alternate path routing may be to distribute load among multiple paths in the network (or, conversely,  to reduce dependence on single mode/path failure).
 

The maximum data carrying capacity of a transmission link. For example, a T1 link purchased from a service provider has a bandwidth of 1.544 megabits per second.
 

A network device that allocates specific portions of the available network bandwidth to different types of connections based on their classification.
 

A data packet-forwarding mechanism which is stateless. That is, the device performing the packet forwarding does not keep track of the data that arrived before the current packet. The packets are forwarded as long as there is bandwidth available to transport the packet. If there is no bandwidth available, the packets are simply discarded.
 

A bridge is a product that connects a local area network (LAN) to another local area network that uses the same protocol (for example, Ethernet or token ring). A bridge examines each message on a LAN, passing those known to be within the same LAN, and forwarding those known to be on the other interconnected LAN (or LANs). 

In bridged networks, computer or node addresses have no specific relationship to the location. For this reason, messages are sent out to every address on the network and accepted only by the intended destination node.
 

A collection of data flows that are grouped together based on a specific definition. The most common definition is the combination of source and destination IP address, source and destination port, and Protocol.
 

A technique that combines classification and queuing of data packets based on rules defined by an administrator. Packets are divided into a hierarchy of classes based on any combination of IP address, protocol, and application type. Each class is assigned a set of bandwidth priorities.
 

This is analagous to address aggregation and is currently used to compact destination based forwarding tables.
 

A 16-bit field in the IPv6 packet header that identifies the relative priority of one packet over another. Network nodes use this field to mark packets according to a set of rules and prioritize the packets appropriately.  This marking and prioritization is used to determine which queue, and hence,  which queue is used to send the packets out the engress device.  This field is used extensively by DiffServ.
 

Guaranteed minimum rate that the service provider commits to providing the customer. CIRs are often purchased separately from the service.
 

An IETF standard that defines a query and response protocol that can be used to exchange policy information between a policy server (Policy Decision Point or PDP, see below) and its clients (Policy Enforcement Points or PEPs, see below).  
 

Network state and load conditions where network traffic is partially or largely not being carried in the most efficient manner.  That is, traffic flows across a link to a bottleneck only to be dropped. Also, in multiservice networks with reservations, this can refer to a high call blocking probability.
 

A mechanism that ensures that the network is able to carry a specified amount of traffic sent through it.
 

See Latency. 
 

An IETF standard, developed by the working group of the same name, that defines a framework on which differentiated service level offerings can be provided for different traffic classes.  It defines bits in the IPv4  ToS  and IPv6  CoS fields that are used to mark and prioritize packets.

The standard defines per-hop-behaviors (PHB) that can be implemented on a node-to-node basis, rather than end-to-end behaviors such as that defined by  IntServ  and RSVP.
 

Traffic policing done at the first RSVP-(and policing-) capable router on a data path. Also known as ingress policing.

A sequence of packets from a source IP address to a destination IP address constituting a single activity (e.g., transfer of a file, a web page, a audio/video stream, etc.) that are usually temporally as well as spatially correlated.
 

A process by which it is determined whether a link or a node has sufficient resources to satisfy the QoS required for a flow. FAC is typically applied by each node in the path of a flow during flow set-up to check local resource availability.
 

A shortcut key for looking up a flow rather than the set of typical IP  (TCPand UDP  protocol fields.
 

A field in the IPv6 packet header that could serve as a globally allocated flow id.
 

In RSVP, the act of determining the path for a flow, and attempting to establish state in routers along the flow path to satisfy its QoS requirement.
 

Used to identify an IPv4 flow in RSVP. A component of a session definition that provides further transport or application protocol layer demultiplexing beyond Desination Address. See Session.

A process by which it is determined whether or not a flow set-up should proceed, based on estimates of the overall resource usage by the flow. Higher-level admission control may result in the failure of a flow set-up even when FAC at each node along the flow path indicates resource availability.

An IP packet stream from a source to a destination (unicast or multicast) with an associated Quality of Service (QoS) (see below) and higher level demultiplexing information. The associated QoS could be best-effort.
 

Variation in the delay between packets.

The killer reservation problem describes a case where a RSVP Receiver attempting and failing to make a large QoS reservation prevents smaller QoS reservations from being established.
 

The overall delay experienced in response to a query, request, etc. The sum total of delays from individual transmission devices, the network, and servers.
 

A traffic-shaping mechanism that allows only a predetermined amount of data to be transmitted over the network. If the data arrives faster that can be transmitted, it is held in buffers until the buffers reach capacity, after which the data is unceremoniously dropped.
 

Allows RSVP to rapidly adapt its reservations to changes in routing.
 

The process of taking the maximum (or more generally the least upper bound) of the RSVP reservations arriving on outgoing interfaces, and forwarding this maximum on the incoming interface.
 

Traffic policing in RSVP that takes place at data merge point of a shared reservation.
 

A subtree of the network topology in which all the leaves and zero or more interior nodes are members of the same multicast group. A multicast path may be per-source, in which case the subtree is rooted at the source.
 

An  RSVP object that carries the Next Hop information in RSVP control messages.
 

The way in which packets are segregated and marked. This is done by matching a set of parameters such as source and destination IP address, source and destination port, and Protocol with a predetermined set. The packets are then appropriately handled by other packet-handling entities such as queues, etc. 
 

When traffic specifies a service contract, it should keep to it (especially if there are tariff (per use or per subscription based) considerations. To check this, traffic is typically monitored as it is placed into appropriate queues for appropriate schedules.

If a service contract is exceeded, packets can be treated in a number of ways:
 

• charge for additional traffic
• mark packets as belonging to a traffic class with a lower precedence

 
• drop out-of-profile packets

 
• mark  packets as out-of-profile

 

A set of precise rules by which the needs of the application are matched to the resources available from the network. These rules may specify the amount of bandwidth to be allocated for a specific application, a relative priority, or the time of day when the policy is to be enforced. 
 

The network entity that decides which policy to apply to a particular connection.
 

The network entity that enforces predetermined policies. The PEP includes classification, traffic policing, and admission control. 
 

Policy-based networking allows various kinds of traffic to get the priority and bandwidth needed to serve the network's users effectively. Using policy statements, network administrators can specify which kinds of service to give priority at what times of day on what parts of their network.

• Global Definition

QoS is not any one technology, but a combination of different complementary technologies that combine to provide different users and different applications with the specific network performance guarantees that allow them to get their jobs done efficiently and quickly.
 

• RSVP Definition

As it relates specifically to RSVP, QoS signifies a set of service requirements that must be met by the networkl when ransporting a RSVP flow.

RSVP itself defines the parameters that determine QoS descriptions. The actual meaning of the parameters is determined by the IETF Integrated Services service class definitions (best effort, controlled load, guaranteed service).  Given a service class, a RSVP flow is also determined by the following RSVP parameters.
 

• Tspec

Traffic specification.
 
• Rspec
  The QoS desired.
• Adspec

Advertised flow spec used by OPWA.
 
• OPWA

One Path With Advertising

An IP routing mechanism whereby flow paths are determined based on some knowledge of resource availability in the network as well as the QoS requirement of flows.
 

In packet processing models, a queue is a data structure which is used to determine the order in which packets are processed, removed and distributed on the output device. Typically, when packets are received on the input device, they are placed on a stack in the order in which they were received. They are removed from the stack and placed on the queue in an order determined by the scheduler.
 

Queue management exists to shape and police traffic in a statistically multiplexed network.

RED  is a technique for controlling best-effort queue lengths.  It is also used for determining out-of-profile packets sent within a service class better than best-effort, such as expedited forwarding (EF).
 

Describes a set of RSVP attributes for a reservation, including the shared attributes and sender selection attributes.
 

ToS (Type of service ) is a fairly well understood field in the IPv4 header. That is, some systems even implement setting it, and some forwarding software (e.g., slip drivers) use it.

QoS refers to a route derived from consideration of network topology, edge costs and traffic conditions.

QoR refers to routes derived soley from edge fixed parameters.
 

A mechanism to keep a flow path fixed for a duration of time.
 

The proposed Internet signaling protocol.

RSVP Session defines one simplex unicast or multicast data flow for which reservations are required. A session is identified by the destination address, transport-layer protocol, and an optional (generalized) destination port.
 

RSVP control state in hosts and routers that will expire if not refreshed within a specified amount of time.
 

The paradigm of packet schedulers is to determine historical resource sharing over a period of time in order to determine which sources are decreased by the request for special circumstances.
 

• First In , First Out (FIFO)



A service discipline in which packets are removed  from the queue  in the same order they were received.
 

• Last In. First Out  (LIFO)


In contrast,  this service discipline removes packets in the reverse order from when they were received and placed on the queue or stack.
 
• Drop Policy

Drop policy determines which packet is dropped when the output queue is full. Types of drop policies include drop tail, drop head, drop random (e.g., RED ).
 

• Priority Queue

Priority queue determines which packet gets fowarded when there is more than one packet in the queue. Note that priority queueing can starve lower priority packets from ever receiving service.  To deal effectivlely with potential starvation, QoS mechanisms such as shapers are required. A shaper typically will use a leaky bucket model; policers typically use a token bucket model.
 

• Precendence Queueing

Precendence queueing bases the priority of a queue on the preceedence bits in the IP header.
 

• Custom Queueing

A proprietary scheme to give certain flows a minimum performance through hardwired classsifications and minimum queue service rates.
 

• Fair Queueing (FQ)

Round robin queueing.
 

• Weighted Fair Queueing (WFQ)

Rounder robin queueing where the rounding error allows for variable length packets. The weight in WFQ sometimes refers to an unfair share.
 

• Earliest Due Date (EDD)

An alternative scheduler for delay bounding.
 

• Deficit Round Robin (DRR)

 

 

A single packet cannot be said to represent a service contract. A flow of packets can see a statistical properties with distributions such as interarrival and departure times, drop probabilities, re-ordering probabilities, erroring probabilities, and so forth.

The difference between traffic distributions before and after service realization are the result of the scheduling discipline, perhaps combined with other traffic (especially true for best effort and priority queueing). We cannot typically derive anything beyond mean and bounds (if they exist) for typical service classes and worst case traffic. The Internet largely survives because of FIFO and drop tail's amazing ability to survive many traffic arrival patterns and under engineering. RED  appears to improve this.
 

QoS management requires an amount of state, which in turn has overhead costs in terms of memory and in the messages required to establish it. The state may simply be a constant increase over the state needed for destination routing. However, destination routing is usually heavily aggregated (e.g. , at least by network number, but also by CIDR, etc).
 

Traffic may be within the bounds of a service contract, but may incur extra jitter due to bursts of other traffic. The question remains whether or not the traffic should then be re-shaped to its mean rate. Typically, if a path has buffering equivalent to:
 

2 * bandwidth * delay

A host or router that can be identified by an unique unicast IPv4 or IPv6 address.
 

Determined by  Route, QoS, QoR and RSVP (signalling) information.
 

A technique for optimizing the lookup for forwarding packets which reduces router memory limits by a constant factor.
 

A tariff structure contains a per QoS parameter, per service class price structure which may include time of day, or even other traffic-related factors. Obviously, the more complex the service model and the tarrif structure, the more expensive it is to collect usage based revenue.
 

A method by which the sending rate of TCP traffic is controlled at the source, such that it fits within the allocated bandwidth.
 

Actual average traffic throughput over a given time interval, expressed in bits per second (bps).
 

A traffic-policing mechanism that allows only a predetermined amount of data to be transmitted over the network. If the data arrives faster that can be transmitted, the data is unceremoniously dropped.
 

A host or router that can be identified by a unique unicast IP address.
 

A way for hosts that cannot use raw sockets to participate in RSVP by encapsulating the RSVP protocol (raw) packets in ordinary UDP packets.
 

A data network that is overlaid on top of a public shared network. Appropriate security and privacy-guarding mechanisms are used so that the network appears to be privately owned by the customer.
 

A framework by which voice conversations are digitized, compressed, and transported over an IP network.
 

A theoretical model for traffic control that limits packets going into the network based on known history (e.g., given a delayed feedback loop of what has arrived, what can we say about what can be sent?).

Currently, Van has implemented this algorithm three times:
 

• Congestion Control
• Adaptive Playout (in Vat)
• Class Based Queueing and RED

 

A data packet queuing technique that distributes incoming packets into different queues based on their relative priorities. Each of the queues is assigned a weight, and the queue with the higher weight gets to transmit its portion of the traffic first. This queuing algorithm ensures that no one class of traffic starves the others.
 

An enterprise network connecting branches to headquarters. WANs can be built using point-to-point leased lines or a shared network such as frame relay.