network support for multimedia - brigham young university · 2017-04-15 · leaky (token) bucket...
TRANSCRIPT
![Page 1: Network Support for Multimedia - Brigham Young University · 2017-04-15 · Leaky (Token) Bucket tokens generated at rate of r tokens/s bucket holds at most b tokens, excess is lost](https://reader034.vdocuments.mx/reader034/viewer/2022042806/5f6a5388889f4146b92b3ab2/html5/thumbnails/1.jpg)
Network Support forMultimedia
Daniel Zappala
CS 460 Computer NetworkingBrigham Young University
![Page 2: Network Support for Multimedia - Brigham Young University · 2017-04-15 · Leaky (Token) Bucket tokens generated at rate of r tokens/s bucket holds at most b tokens, excess is lost](https://reader034.vdocuments.mx/reader034/viewer/2022042806/5f6a5388889f4146b92b3ab2/html5/thumbnails/2.jpg)
Classes Scheduling Policing Differentiated Services Integrated Services Retrospective
Network Support for Multimedia
• make the best of best effort• use application-level techniques• use CDNs for stored content• provision the network• simple and broadly supported
• differentiated services• treat different classes of traffic differently, but no circuit-like
guarantees• use packet marking, policing, scheduling• incremental, some deployment
• integrated services• per-flow virtual circuits to guarantee bandwidth, delay• use packet marking, policing, scheduling, call admission and
signalling• needs widespread adoption, little deployment
2/33
![Page 3: Network Support for Multimedia - Brigham Young University · 2017-04-15 · Leaky (Token) Bucket tokens generated at rate of r tokens/s bucket holds at most b tokens, excess is lost](https://reader034.vdocuments.mx/reader034/viewer/2022042806/5f6a5388889f4146b92b3ab2/html5/thumbnails/3.jpg)
Classes Scheduling Policing Differentiated Services Integrated Services Retrospective
Making the Best of Best Effort
• provide bandwidth to eliminate queueing delay and loss• bandwidth provisioning: given a topology, determine the
capacity required on all links to meet performance requirements• network dimensioning: design a topology (place routers, create
links, assign bandwidth to links) to meet performancerequirements
• provisioning process• model traffic between network end points (e.g. arrival rate of
users, arrival rate of packets for a given application)• define performance requirements• develop a model to predict performance given a workload• find a minimal cost bandwidth allocation that will meet all
user requirements
• who will pay the ISPs to install sufficient bandwidth?
• will ISPs cooperate to ensure the entire path is properlyprovisioned?
3/33
![Page 4: Network Support for Multimedia - Brigham Young University · 2017-04-15 · Leaky (Token) Bucket tokens generated at rate of r tokens/s bucket holds at most b tokens, excess is lost](https://reader034.vdocuments.mx/reader034/viewer/2022042806/5f6a5388889f4146b92b3ab2/html5/thumbnails/4.jpg)
Multiple Classes of Service
![Page 5: Network Support for Multimedia - Brigham Young University · 2017-04-15 · Leaky (Token) Bucket tokens generated at rate of r tokens/s bucket holds at most b tokens, excess is lost](https://reader034.vdocuments.mx/reader034/viewer/2022042806/5f6a5388889f4146b92b3ab2/html5/thumbnails/5.jpg)
Classes Scheduling Policing Differentiated Services Integrated Services Retrospective
Providing Multiple Classes of Service
• simple solution to provide different levels of service todifferent types of applications
• analogous to first-class airline service• provided for aggregates of traffic, not individual flows
• dates back to 1979, when Type of Service (ToS) bits wereprovided in the IPv4 header
5/33
![Page 6: Network Support for Multimedia - Brigham Young University · 2017-04-15 · Leaky (Token) Bucket tokens generated at rate of r tokens/s bucket holds at most b tokens, excess is lost](https://reader034.vdocuments.mx/reader034/viewer/2022042806/5f6a5388889f4146b92b3ab2/html5/thumbnails/6.jpg)
Classes Scheduling Policing Differentiated Services Integrated Services Retrospective
Motivating Scenarios
• we will use the figure below to explore general principles• H1 is talking to H3, H2 is talking to H4• classic example used for TCP and QoS
6/33
![Page 7: Network Support for Multimedia - Brigham Young University · 2017-04-15 · Leaky (Token) Bucket tokens generated at rate of r tokens/s bucket holds at most b tokens, excess is lost](https://reader034.vdocuments.mx/reader034/viewer/2022042806/5f6a5388889f4146b92b3ab2/html5/thumbnails/7.jpg)
Classes Scheduling Policing Differentiated Services Integrated Services Retrospective
Packet Marking
• example: 1 Mbps VoIP, competing with FTP application• bursts of FTP packets can cause audio loss• want to give priority to audio over FTP
• packet marking: allows a router to distinguish among packetsbelonging to different classes of traffic
• can be done by application or by router
7/33
![Page 8: Network Support for Multimedia - Brigham Young University · 2017-04-15 · Leaky (Token) Bucket tokens generated at rate of r tokens/s bucket holds at most b tokens, excess is lost](https://reader034.vdocuments.mx/reader034/viewer/2022042806/5f6a5388889f4146b92b3ab2/html5/thumbnails/8.jpg)
Classes Scheduling Policing Differentiated Services Integrated Services Retrospective
Traffic Isolation
• problems• what if audio application sends as many high-priority packets
as it wants?• what if many audio applications share the same class but one
is not well behaved?
• traffic isolation : protect classes and flows from each other sothat they are not adversely affected when someone misbehaves
• two approaches: policing and resource reservation
8/33
![Page 9: Network Support for Multimedia - Brigham Young University · 2017-04-15 · Leaky (Token) Bucket tokens generated at rate of r tokens/s bucket holds at most b tokens, excess is lost](https://reader034.vdocuments.mx/reader034/viewer/2022042806/5f6a5388889f4146b92b3ab2/html5/thumbnails/9.jpg)
Classes Scheduling Policing Differentiated Services Integrated Services Retrospective
Policing
• policing: enforce a rate on a network flow, usually at networkedge
• consequences if you exceed your rate: marking or droppingextra packets
9/33
![Page 10: Network Support for Multimedia - Brigham Young University · 2017-04-15 · Leaky (Token) Bucket tokens generated at rate of r tokens/s bucket holds at most b tokens, excess is lost](https://reader034.vdocuments.mx/reader034/viewer/2022042806/5f6a5388889f4146b92b3ab2/html5/thumbnails/10.jpg)
Classes Scheduling Policing Differentiated Services Integrated Services Retrospective
Resource Reservation
• resource reservation: allocate resources (e.g. bandwidth) toeach flow
• ideally, if audio application goes silent, data application canutilize unused bandwidth
10/33
![Page 11: Network Support for Multimedia - Brigham Young University · 2017-04-15 · Leaky (Token) Bucket tokens generated at rate of r tokens/s bucket holds at most b tokens, excess is lost](https://reader034.vdocuments.mx/reader034/viewer/2022042806/5f6a5388889f4146b92b3ab2/html5/thumbnails/11.jpg)
Scheduling
![Page 12: Network Support for Multimedia - Brigham Young University · 2017-04-15 · Leaky (Token) Bucket tokens generated at rate of r tokens/s bucket holds at most b tokens, excess is lost](https://reader034.vdocuments.mx/reader034/viewer/2022042806/5f6a5388889f4146b92b3ab2/html5/thumbnails/12.jpg)
Classes Scheduling Policing Differentiated Services Integrated Services Retrospective
FIFO: First In First Out
• transmit in order of arrival to queue
• discard policy: if packet arrives and queue is full, which packetshould the scheduler discard?
• drop-tail: drop the arriving packet• priority: remove a packet based on priority• random: remove a packet randomly
12/33
![Page 13: Network Support for Multimedia - Brigham Young University · 2017-04-15 · Leaky (Token) Bucket tokens generated at rate of r tokens/s bucket holds at most b tokens, excess is lost](https://reader034.vdocuments.mx/reader034/viewer/2022042806/5f6a5388889f4146b92b3ab2/html5/thumbnails/13.jpg)
Classes Scheduling Policing Differentiated Services Integrated Services Retrospective
FIFO in Operation
13/33
![Page 14: Network Support for Multimedia - Brigham Young University · 2017-04-15 · Leaky (Token) Bucket tokens generated at rate of r tokens/s bucket holds at most b tokens, excess is lost](https://reader034.vdocuments.mx/reader034/viewer/2022042806/5f6a5388889f4146b92b3ab2/html5/thumbnails/14.jpg)
Classes Scheduling Policing Differentiated Services Integrated Services Retrospective
Priority Queuing
• transmit the highest priority queued packet
• may have multiple classes with different priorities, based onmarking or IP addresses, ports
14/33
![Page 15: Network Support for Multimedia - Brigham Young University · 2017-04-15 · Leaky (Token) Bucket tokens generated at rate of r tokens/s bucket holds at most b tokens, excess is lost](https://reader034.vdocuments.mx/reader034/viewer/2022042806/5f6a5388889f4146b92b3ab2/html5/thumbnails/15.jpg)
Classes Scheduling Policing Differentiated Services Integrated Services Retrospective
Priority Queue in Operation
15/33
![Page 16: Network Support for Multimedia - Brigham Young University · 2017-04-15 · Leaky (Token) Bucket tokens generated at rate of r tokens/s bucket holds at most b tokens, excess is lost](https://reader034.vdocuments.mx/reader034/viewer/2022042806/5f6a5388889f4146b92b3ab2/html5/thumbnails/16.jpg)
Classes Scheduling Policing Differentiated Services Integrated Services Retrospective
Round Robin Scheduling
• round robin among multiple classes of traffic, either throughmarking or IP header
16/33
![Page 17: Network Support for Multimedia - Brigham Young University · 2017-04-15 · Leaky (Token) Bucket tokens generated at rate of r tokens/s bucket holds at most b tokens, excess is lost](https://reader034.vdocuments.mx/reader034/viewer/2022042806/5f6a5388889f4146b92b3ab2/html5/thumbnails/17.jpg)
Classes Scheduling Policing Differentiated Services Integrated Services Retrospective
Weighted Fair Queuing
• generalized round robin among classes or flows of traffic
• each flow is weighted with wi and is guaranteed to receivethroughput of R ∗ wi/
∑wj
• takes into account packet sizes
17/33
![Page 18: Network Support for Multimedia - Brigham Young University · 2017-04-15 · Leaky (Token) Bucket tokens generated at rate of r tokens/s bucket holds at most b tokens, excess is lost](https://reader034.vdocuments.mx/reader034/viewer/2022042806/5f6a5388889f4146b92b3ab2/html5/thumbnails/18.jpg)
Classes Scheduling Policing Differentiated Services Integrated Services Retrospective
Work-Conserving
• work-conserving: indicates that a scheduling algorithm willnever allow the link to become idle if there are packets to besent
• FIFO• priority queuing• WFQ
• non-work-conserving: link may be left idle even when thereare packets ready
• round robin : if each class allocated a fixed number of slots(similar to TDMA)
18/33
![Page 19: Network Support for Multimedia - Brigham Young University · 2017-04-15 · Leaky (Token) Bucket tokens generated at rate of r tokens/s bucket holds at most b tokens, excess is lost](https://reader034.vdocuments.mx/reader034/viewer/2022042806/5f6a5388889f4146b92b3ab2/html5/thumbnails/19.jpg)
Policing
![Page 20: Network Support for Multimedia - Brigham Young University · 2017-04-15 · Leaky (Token) Bucket tokens generated at rate of r tokens/s bucket holds at most b tokens, excess is lost](https://reader034.vdocuments.mx/reader034/viewer/2022042806/5f6a5388889f4146b92b3ab2/html5/thumbnails/20.jpg)
Classes Scheduling Policing Differentiated Services Integrated Services Retrospective
Policing Parameters
• goal: limit traffic to a set of declared parameters
• common parameters:• average rate: number of bits per second, over some time
interval• 100 bits/second = 6000 bits/minute• what interval do you want to average over?
• peak rate: highest rate allowed, over some time interval• typically over a shorter interval• example: 256 kbps average per minute, 1 Mbps peak per
second
• burst size: number of packets sent consecutively
20/33
![Page 21: Network Support for Multimedia - Brigham Young University · 2017-04-15 · Leaky (Token) Bucket tokens generated at rate of r tokens/s bucket holds at most b tokens, excess is lost](https://reader034.vdocuments.mx/reader034/viewer/2022042806/5f6a5388889f4146b92b3ab2/html5/thumbnails/21.jpg)
Classes Scheduling Policing Differentiated Services Integrated Services Retrospective
Leaky (Token) Bucket
• tokens generated at rate of r tokens/s
• bucket holds at most b tokens, excess is lost
• maximum burst size: b
• average rate over interval t: rt + b
21/33
![Page 22: Network Support for Multimedia - Brigham Young University · 2017-04-15 · Leaky (Token) Bucket tokens generated at rate of r tokens/s bucket holds at most b tokens, excess is lost](https://reader034.vdocuments.mx/reader034/viewer/2022042806/5f6a5388889f4146b92b3ab2/html5/thumbnails/22.jpg)
Classes Scheduling Policing Differentiated Services Integrated Services Retrospective
Guaranteed Delay
• leaky bucket + weighted fair queuing = provable maximumdelay in a queue
• dmax = biR∗wi/
∑wj
22/33
![Page 23: Network Support for Multimedia - Brigham Young University · 2017-04-15 · Leaky (Token) Bucket tokens generated at rate of r tokens/s bucket holds at most b tokens, excess is lost](https://reader034.vdocuments.mx/reader034/viewer/2022042806/5f6a5388889f4146b92b3ab2/html5/thumbnails/23.jpg)
Differentiated Services
![Page 24: Network Support for Multimedia - Brigham Young University · 2017-04-15 · Leaky (Token) Bucket tokens generated at rate of r tokens/s bucket holds at most b tokens, excess is lost](https://reader034.vdocuments.mx/reader034/viewer/2022042806/5f6a5388889f4146b92b3ab2/html5/thumbnails/24.jpg)
Classes Scheduling Policing Differentiated Services Integrated Services Retrospective
Differentiated Services
• scalability• need to handle hundreds of thousands of simultaneous flows• keep the network core simple, with complexity at the edges
• flexibility• may develop new services class or stop using old classes• provide functional components, rather than specific classes• generic ability to map flows into classes
24/33
![Page 25: Network Support for Multimedia - Brigham Young University · 2017-04-15 · Leaky (Token) Bucket tokens generated at rate of r tokens/s bucket holds at most b tokens, excess is lost](https://reader034.vdocuments.mx/reader034/viewer/2022042806/5f6a5388889f4146b92b3ab2/html5/thumbnails/25.jpg)
Classes Scheduling Policing Differentiated Services Integrated Services Retrospective
Architecture
• edge routers: packet classification and traffic conditioning(policing)
• core routers: schedule packets based on traffic class (per-hopbehavior), no discrimination between flows in the same class
25/33
![Page 26: Network Support for Multimedia - Brigham Young University · 2017-04-15 · Leaky (Token) Bucket tokens generated at rate of r tokens/s bucket holds at most b tokens, excess is lost](https://reader034.vdocuments.mx/reader034/viewer/2022042806/5f6a5388889f4146b92b3ab2/html5/thumbnails/26.jpg)
Classes Scheduling Policing Differentiated Services Integrated Services Retrospective
Packet Classification and Traffic Conditioning
• occurs at edge routers
• classifier marks traffic class
• marker marks packets in or out of profile based on apre-negotiated rate (e.g. leaky bucket)
• shaper may force flow to conform to bucket parameters ordrop some traffic
26/33
![Page 27: Network Support for Multimedia - Brigham Young University · 2017-04-15 · Leaky (Token) Bucket tokens generated at rate of r tokens/s bucket holds at most b tokens, excess is lost](https://reader034.vdocuments.mx/reader034/viewer/2022042806/5f6a5388889f4146b92b3ab2/html5/thumbnails/27.jpg)
Classes Scheduling Policing Differentiated Services Integrated Services Retrospective
Per-Hop Behaviors
• forwarding behaviors performed at core routers
• resulting service must be measurable
• expedited forwarding• rate given to a class must equal or exceed a specified rate• provides a logical link with a minimum guaranteed rate to a
particular class of traffic
• assured forwarding• divides traffic into four classes, each with a minimum
guaranteed rate• each class has three drop preference categories• provides a way for customers to allocate traffic into classes and
to exceed traffic specifications while also specifying whichpackets get dropped first
27/33
![Page 28: Network Support for Multimedia - Brigham Young University · 2017-04-15 · Leaky (Token) Bucket tokens generated at rate of r tokens/s bucket holds at most b tokens, excess is lost](https://reader034.vdocuments.mx/reader034/viewer/2022042806/5f6a5388889f4146b92b3ab2/html5/thumbnails/28.jpg)
Integrated Services
![Page 29: Network Support for Multimedia - Brigham Young University · 2017-04-15 · Leaky (Token) Bucket tokens generated at rate of r tokens/s bucket holds at most b tokens, excess is lost](https://reader034.vdocuments.mx/reader034/viewer/2022042806/5f6a5388889f4146b92b3ab2/html5/thumbnails/29.jpg)
Classes Scheduling Policing Differentiated Services Integrated Services Retrospective
Integrated Services
• goal: provide quality of service guarantees to individualconnections on the Internet
• major components• scheduling: meet performance requirements at each hop, e.g.
with Weighted Fair Queueing• admission control: decide whether there are enough resources
for the connection at each hop• resource reservation: a setup protocol to get approval from
admission control and reserve resources with the scheduleralong a path
• QOS routing: a routing protocol that can find a path withavailable resources
• pricing and billing: charge users for the resources they use
29/33
![Page 30: Network Support for Multimedia - Brigham Young University · 2017-04-15 · Leaky (Token) Bucket tokens generated at rate of r tokens/s bucket holds at most b tokens, excess is lost](https://reader034.vdocuments.mx/reader034/viewer/2022042806/5f6a5388889f4146b92b3ab2/html5/thumbnails/30.jpg)
Classes Scheduling Policing Differentiated Services Integrated Services Retrospective
RSVP
• Resource ReSerVation Protocol: connection setup protocol forthe Internet
• features• resource reservations for multicast trees• receiver-oriented: group member initiates and maintains
reservation on the tree• soft state: reservations must be refreshed or else they will
expire
• source must specify a traffic specification (e.g. leaky tokenbucket parameters) and a resource specification (e.g.minimum bandwidth, maximum delay)
• network provides guaranteed bandwidth + leaky bucket =guaranteed delay
30/33
![Page 31: Network Support for Multimedia - Brigham Young University · 2017-04-15 · Leaky (Token) Bucket tokens generated at rate of r tokens/s bucket holds at most b tokens, excess is lost](https://reader034.vdocuments.mx/reader034/viewer/2022042806/5f6a5388889f4146b92b3ab2/html5/thumbnails/31.jpg)
Classes Scheduling Policing Differentiated Services Integrated Services Retrospective
The Call Setup Process
31/33
![Page 32: Network Support for Multimedia - Brigham Young University · 2017-04-15 · Leaky (Token) Bucket tokens generated at rate of r tokens/s bucket holds at most b tokens, excess is lost](https://reader034.vdocuments.mx/reader034/viewer/2022042806/5f6a5388889f4146b92b3ab2/html5/thumbnails/32.jpg)
Retrospective
![Page 33: Network Support for Multimedia - Brigham Young University · 2017-04-15 · Leaky (Token) Bucket tokens generated at rate of r tokens/s bucket holds at most b tokens, excess is lost](https://reader034.vdocuments.mx/reader034/viewer/2022042806/5f6a5388889f4146b92b3ab2/html5/thumbnails/33.jpg)
Classes Scheduling Policing Differentiated Services Integrated Services Retrospective
Retrospective
• We’ve been trying to provide QoS on the Internet for 20 years– why have we failed?
• to provide end-to-end integrated or differentiated services, allISPs between two hosts must provide the service and haveservice agreements between each other (e.g. agree to carrytraffic and give it the specified service, for a fee)
• need to police, shape, schedule, admit traffic – extradeployment and management challenges
• need to bill users, probably by volume of traffic
• under low or moderate load, we can get most of the benefitsof QoS using best-effort service!
• provisioning is the easy way out
33/33