Lic.(Tech.) Marko Luoma (1/24) Lic.(Tech.) Marko Luoma (2/24) Scheduling Task of a scheduler is to decide the order of packets which are transmitted from � the queue S−38.180 Palvelunlaatu Internetissä S−38.180 Quality of Service in Internet A B C Luento 4: Mekanismit − osa 2 D E ? Lecture 4: Mechanisms − part 2 I D H C G A J E B F F G H I J Lic.(Tech.) Marko Luoma (3/24) Lic.(Tech.) Marko Luoma (4/24) Scheduling Scheduling Selecting the order of packets means One end in this continum is that There are vast amount of schedulers � � � that resource sharing is controlled predefined amount of resources is developed for different purposes with predefined policy. allocated to the connection. Generally they can be divided into � Policy defines the amount of Other end is that no allocation is categories of � � Work−conserving Non−work−conserving resources which are allocated to the done and resources are shared on the � Work−conserving vs non−work− connections / classes for which basis of the need conserving single packets belong to. � Time−based vs frame−based Sorted−priority SCHEDULERS Frame−based � Continuous vs packetized � Priority vs no priority Fluid−flow Packet−per−packet
Lic.(Tech.) Marko Luoma (5/24) Lic.(Tech.) Marko Luoma (6/24) Scheduling Scheduling Conservation of work means that Why we would want to have non− Continuous time Packetized � � � � work conserving scheduler ? scheduler is executing its task as � Scheduling decissions and � Scheduling decissions and long as it has some work to do. Conservation of work means that calculations are done based on calculations are based on packet � packets are sent to the link even continuous time units per packet analysis Techinacally this means that there � though for the receiving would are packets in the queue which has to � Fluid−Flow modeling − packets � Distorts fluid flow model prefere it to come a little bit later. be sent into the link before scheduler are infinitesimally small can take a break i.e. change to the This can happen with real−time � � Assumes that number of packets idle state. applications which send packets with could be served on same time constant time intervals. However, Non−work conserving scheduler can � (not possible) network can multiplex them so that idle even though it has packets in the they form bursts. Non−work queue. B B B conserving scheduler may delay B A B A B A A A A packets so that intervals structure is maintained throughout the network. Lic.(Tech.) Marko Luoma (7/24) Lic.(Tech.) Marko Luoma (8/24) Scheduling Scheduling Time based scheduling Frame based scheduling Scheduling can happen: � � � � Uses either arrival time or � Uses fixed frame which is � Within one queue , sorting packets inside queue to appropriate transmission finishing time as a criteria for partitioned for scheduled items order ordering based on their weights. � Between several queues , dispatching head of line packets from different � Time may be virtual or real−time � During rotation if partition and queues depending on scheduler time left overs from previous partition � Hierarchically over several schedulers , combination of previous ones aggregate enough token for a � Virtual time is usually finishing Many of scheduling algorithms can be used to produce QoS in each of these item then it is served. If not � time in ideal scheduler i.e. cases tokens are added for next round. Scheduler which is not packetized � Number of packets may be served from a single class if frame is big.
Lic.(Tech.) Marko Luoma (9/24) Lic.(Tech.) Marko Luoma (10/24) Scheduling Scheduling First Come First Served (FCFS) is prevalent scheduling method in routers. Simple priority scheduler extends FCFS to be able to distinguish between � � more and less important traffic. FCFS uses arrival time information as sorting criteria for packet dispatching. � Packets are ordered first based on their priority and second on their arrival time. � FCFS is not able to offer any QoS as time is the only parameter that has � influence to the order of packets. Low priority High priority Medium priority Class Arrival Time Service Time C1 C2 C1 C2 C3 C1 C3 C1 C2 C1 C1 C2 C3 C3 Class Arrival Time Service Time C1 0 1 C1 0 1 C2 0 3 C2 0 3 C1 3 1 Departure time Departure time C1 3 1 C3 3 3 C3 3 3 C2 3 2 Scheduling time C2 3 2 Scheduling time C1 5 1 C1 5 1 C3 7 3 C3 7 3 Arrival time Arrival time C1 C1 C1 C3 C1 C1 C1 C3 C2 C2 C2 C2 C3 C3 Lic.(Tech.) Marko Luoma (11/24) Lic.(Tech.) Marko Luoma (12/24) Scheduling Scheduling Prioritized ordering may lead to starvation of resources in low priority classes if Deadline based scheduling schemes (e.q. Earlies Due Date) are based on the � � traffic in high priority classes is not limited. calculation of finishing time if packet would have been scheduled when it arrived to the queue. This can be accomplished by using � Packets are transmitted on the order of finishing times. � Connection admission control � � Over provisioning � Rate control Class Arrival Time Service Time C1 C2 C1 C2 C3 C1 C3 C1 0 1 � Modifying priority scheduler to take class rates into account (token based C2 0 3 C1 3 1 Departure time operation) C3 3 3 C2 3 2 Scheduling time C1 5 1 C3 7 3 High priority High priority Arrival time Low priority Low priority C1 C1 C1 C3 C2 C2 C3
Lic.(Tech.) Marko Luoma (13/24) Lic.(Tech.) Marko Luoma (14/24) Scheduling Scheduling Generalized Processor Sharing is ideal fair queueing algorithm which is Disadvantages of GPS are: � � based on fluid flow model. � Departures from GPS are colliding which makes the use of GPS based scheduler impossible GPS provides service to the individual connections based on their weights. � � However it may be used as backgroud scheduler if collisions are GPS is work conserving scheduler and thus distributes excess capacity to � resolved in some manner connections which are able to utilize it. C1 � Heavy calculation of departure times C2 Class Arrival Time Service Time C1 C2 C2 C1 C3 C3 C1 0 1 � Departure time of every packet in scheduler changes whenever a packet C2 0 3 C1 3 1 arrives or departs the scheduler Departure time C3 3 3 C2 3 2 Scheduling time C1 5 1 C3 7 3 Arrival time Weights are all equal C1 C1 C1 C3 C2 C2 C3 Lic.(Tech.) Marko Luoma (15/24) Lic.(Tech.) Marko Luoma (16/24) Scheduling Scheduling Advantages of GPS are: Packetized Generalized Processor Sharing is packet per packet � � approximation of GPS scheduling. � Fairness which it provides for the sharing connections Most prevalent implementation of PGPS is weighted fair queueing (WFQ) � Se r vi c e t , t � ∆ t ≥ W e i g h t i i Se r vi c e t , t � ∆ t W e i g h t j WFQ uses calculation of finishing time in corresponging GPS system as a � j criteria for sorting the packets. � Strict delay bound caused by scheduling when traffic is constrained by a Class Arrival Time Service Time C1 C2 C1 C1 C3 C2 C3 token bucket of token rate r and bucket depth b C1 0 1 C2 0 3 C1 3 1 Departure time C3 3 3 W e i g h t i Se r v i c e r a t e f o r c o n n e c t i o n i : r i ≥ ⋅ L i n k R a t e C2 3 2 ∑ Scheduling time W e i g h t h j C1 5 1 j C3 7 3 b i D i ≤ Arrival time D e l a y f o r c o n n e c t i o n i : r i Weights are all equal Remember these results were derived from the assumption that packets flow like C1 C1 C1 C3 C2 C2 fluid through the system i.e. there would be a dedicated link with capacity r between endpoints. C3
Recommend
More recommend
