Abstract:Maintaining scheduling parameters onlyineach packet's header makeitdifficult forcore-stateless networks toprovide deterministic (zero loss) fair bandwidth allocation toeachflow. BasedonVCSGR,thispaperpresents anew algorithm, called FVCSGR (Fair-VCSGR). Inanetwork of FVCSGR,thearriving traffic ofeachflowisdivided atthe ingress nodeintoa setoflayers according toitsreserved rate. Everypacket ismarkedataningress nodewithalayer label andtwovirtual delay vectors: thebasic vector andthe fair vector. Thebasic vector whichcomesfromVCSGR is usedtoprovide accurate basic bandwidth guarantees and thesameupperboundon end-to-end delayasitsGR counterpart. Ontheotherhand,thefair vector isusedwith layerlabelto guarantee proportional fairbandwidth allocation. We show thatthe out-of-order problem originated fromlayer scheduling canberestricted within a narrowrange. I.INTRODUCTION DPS(Dynamic Packet State) based core-stateless schemes that canachieve bothfine-grained QoSguarantees and scalability tosomeextent havebeenproposed (2,3,4). However, notperforming per-flow management inside the network raises somenewissues. Thatis, thecurrent packet can't knowthetransmission states oftheprevious packet. Theresult isthat theservice tagsusedbycorenodes for sorting packets areonly gotintheworst-case assumption (i.e. assuming that all packets before thecurrent oneswere served exactly atthereserved rate). Thelost information won'taffect theupper boundontheend-to-end delay of thescheduling algorithm, butwill result inthelost ofthe fairness characteristic, whichisrequired by many applications. We aimatimplementing acore-stateless fair scheduling algorithm inascalable manner. To achieve sucha purpose, we divide thebandwidth allocation intotwoparts conceptive: basic bandwidth allocation andextra bandwidth allocation, where thebasic bandwidth isallocated tothose active flows that always havepackets inqueuewaiting fortransmission according totheir reserved rate. Whenthebasic bandwidth allocation isfinished, theleft extra bandwidth should beshared in proportional totheir reserved rateamongalltheactive flows attheservers that those flows traverse. Thecore- stateless fair scheduling algorithm canbeimplemented via such twosteps. In(1), wepropose anewcorestateless packet scheduling algorithm, called VCSGR (Vector-based Core-Stateless Guaranteed Rate). Through thevirtual delayvector technique, anetwork ofVCSGRservers canachieve the goalthat provide delay guarantee sameasanetwork of corresponding GR servers aswellasaccurate rate guarantee closely similar tothosestateful ones,like VirtualClock. Inthis paper, based onVCSGR,wepropose anewcore- stateless fairpacketscheduling algorithm, called FVCSGR(Fair Vector-based CoreStateless Guaranteed Rate), toachieve deterministic fairness andend-to-end delay guarantees incore-stateless networks. FVCSGRisa combination ofavirtual delay vector scheme ofVCSGR andalayer marking/scheduling scheme. First, thearriving traffic ofeachflowisdivided atthe ingress nodeinto asmall number ofthin layers according toitsreserved rateandevery packet isstamped alayer label. Then, thebasic andfair virtual delay vector are calculated andalsocarried bypacket itself asper-flow state information. Queuing server ofeachnodealong a packet's path computes twoservice tagsforeachpacket respectively: thebasic stampandthefair stampaccording tothebasic andfair vector carried inthepacket header. Eachserver maintains a separate priority queueforeachlayer (numbered from1)withthefair stamp askey. We also set anextra queuewiththebasic stamp askey, called layer 0. Therefore, apacket fromlayer Q havetoqueuetwice whenitarrives: its basic stampisinserted inlayer 0and itsfair stampisinserted inlayer Q.FVCSGRsetan eligible scheduling areatoavoid packets oflower layers fromoccupying toomachbandwidth. Eachserver ina FVCSGRnetwork picks thepacket withthelowest stamp ofthelowest non-empty layer (including layer 0)inthe eligible scheduling areaforservicing next.

VCSVC: Vector Core Stateless Virtual Clock

Virtual Delay Vector Based Core-Stateless Packet Scheduling Algorithm

Rate-Guaranteed Core-Stateless Packet Scheduling Algorithm

Core-stateless Fair Bandwidth Allocation for Guaranteed Services, Part I: End-to-end Precise Basic Bandwidth Guarantees

Core-stateless FairBandwidth Allocation Forguaranteed Services, PartII: Proportional Allocation Ofexcess Bandwidth

Core-stateless Virtual Clock Scheduling Algorithms

Core-stateless Fair Bandwidth Allocation for Guaranteed Services, Part II: Proportional Allocation of Excess Bandwidth

An Efficient Virtual Network Embedding Algorithm with Delay Constraints

Scheduling Reconfigurable Service Function Chain

EVCP: a Convergence Time Improved High-Speed Transport Congestion Control Protocol

VQ-CSMA: Throughput-Optimal Low-Delay Random Access

Awareness scheduling and algorithm implementation for collaborative virtual environment

Twenty Years After: Hierarchical Core-Stateless Fair Queueing

Congestion Control with Deterministic Service Delay Guarantee

Theoretical Analysis and Improvement on the Core-Stateless Fair Queueing Algorithm

Fair Scheduling for Delay-Sensitive VoIP Traffic

A Channel-State Aware Scheduling Mechanism for Wireless Local Area Networks

A Scheduling Algorithm Using Compensating Round Robin in Packet-Swtiching Broadband Networks

Dynamic TDM Virtual Circuit Implementation for NoC

Bypassed channels based differentiated service implementation method for network-on-chip

A Scalable Per-flow Priority Scheduling Scheme for High-Speed Network