QoS models themselves may be divided into two major classes:
?– Soft QoS (Class of Service or CoS) In this model (epitomised by the IETF
DiffServ architecture), forwarding plane QoS techniques are used to provide different
forwarding treatment to different classes of traffic. This model scales well, since
there is no need for per-flow state, and is applicable to connectionless and connectionorientated
forwarding, but is unable to provide guaranteed forwarding behavior.
?– Hard QoS In this model (epitomised by the IETF IntServ architecture) resources
are reserved for the each connection using the control plane, and then forwarding
plane techniques are applied to ensure that each connection gets the required level
of QoS. This model is less scalable (as per-flow reservations are required), but can
be used to enable a firm guarantee that a traffic contract will be met.
Control Plane QoS When signalling an MPLS LSP with RSVP-TE, an LSR signals the
bandwidth required for the LSP. Each LSR in the path checks the bandwidth requested
against the bandwidth available on its outbound link and only forwards the Path message
if there is sufficient bandwidth available for the LSP on that link.
Pages:
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977