60 likes | 215 Views
“A QoS Model for Signaling IntServ Controlled-Load Service with NSIS” draft-kappler-nsis-qosmodel-controlledload-01.txt. cornelia.kappler at siemens.com. What is IntServ Controlled-Load Service?. IntServ Controlled Load Service is (in NSIS terms) a QoS Model
E N D
“A QoS Model for Signaling IntServ Controlled-Load Service with NSIS” draft-kappler-nsis-qosmodel-controlledload-01.txt cornelia.kappler at siemens.com
What is IntServ Controlled-Load Service? • IntServ Controlled Load Service is (in NSIS terms) a QoS Model • RFC 2210 specifies how to signal for Ctrld Load using RSVP • This ID specifies how to signal for Ctrld Load using NSIS • Controlled-Load Service (RFC 2211) • Provides approximately service of an unloaded best-effort network • QoS parameters signaled are Token Bucket and MTU • Implemented per “network element”, i.e. per-router or per-subnet • Can be used for • Reserving resources per-flow per-router • Admission control at edge of DiffServ domains • Admission control into MPLS clouds • What is Controlled Load specific, what is RSVP specific? • Only Controlled Load specifics need to be reflected in this ID • “extra features” compared to RSVP are in green
How to signal for Controlled-Load Service with NSIS • Role of QNEs • One or more QNE per “network element” • Stateless QNEs? • “Provide approximately service of an unloaded best-effort network” may also be possible with stateless QNEs • QoS-model specific Control Information • <QOSM Hops>? • Even if don’t implement Ctrld Load, all QNEs must be able to interpret all parameters • because all parameters for Ctrld Load are mandatory parameters • <Excess Treatment> may be included • QSPEC Procedures • No restriction, i.e. sender-ini, receiver-ini with all combination of QSPEC objects
How to signal for Controlled-Load Service with NSIS • QSPEC objects: <QoS Desired> = <token bucket>, (<QoS Class>) <QoS Available> = (<bandwidth>,<path latency>,<token bucket>, <QoS Class>), <MTU> <Minimum QoS> = <token bucket> OR <MTU>, (<QoS Class>) <QoS Reserved> = <token bucket> • Of these, only <QoS Desired> and <QoS Reserved> will be used in all procedures • E.g. for sender-initiated reservation can use only <QoS Desired> • May include <QoS Class> to provide for DiffServ domains • In QoS Available, <bandwidth><path latency> are there because this is how RSVP does it. Usage unclear • For admission control? • For parameters appearing in QoS Available and/or Minimum QoS can accept downgrade • In all QSPEC objects, additional parameters may be included
How to signal for Controlled-Load Service with NSIS • How to deal with MTU? • In sender-ini procedures, if reservation fails… • E.g. because of too big <M> (Max packet size)? …how can receive meaningful feedback? • When reservation fails only receive feedback up to failure point • Could include lower bound for <M> in <Minimum QoS> to avoid failure of reservation because of MTU problem • <Minimum QoS> is not necessarily supported by all QNEs • Xiaoming: “I would prefer to remove it from IntServ discussion; instead, PMTU discovery or out-of-band MTU information exchange should be another process instead inside QoS signaling itself.”