Post on 15-Jan-2016
transcript
T.Sharon-A.Frank
3
Internet/Multimedia Assumptions
• Internet – Point-to-Point (unicast)– Best-Effort Delivery– Elastic Applications– FIFO Packet Scheduling– Provides average Packet
Delay– End-to-End Reliability– Statistical Multiplexing
Gain
• Multimedia– Multipoint
– Soft RT Constraints
– Inelastic Applications
– Need Control over Delay and Jitter
– Various Traffic Classes
– Need QoS Guarantees
T.Sharon-A.Frank
4
Application Taxonomy (1)
Elastic Inelastic
Applications
Elastic Applications:
Can tolerate relatively large delay variance – essentially the traditional data application.
Inelastic Applications:
Comparatively intolerant to delay, delay variance, throughput variance and errors.
T.Sharon-A.Frank
5
Examples of Elastic Applications
• Email:– asynchronous
– message is not real-time
– delivery in several minutes is acceptable
• File transfer:– interactive service
– require “quick” transfer
– “slow” transfer acceptable
• Network file service:– interactive service– similar to file transfer– fast response required– (usually over LAN)
• WWW:– interactive– file access mechanism– fast response required– QoS sensitive content on
WWW pages
T.Sharon-A.Frank
6
Examples of Inelastic Applications
• Streaming voice:– not interactive– end-to-end delay
not important– end-to-end jitter not
important– data rate and loss
very important
• Real-time voice:– person-to-person– interactive– important to control:
• end-to-end data rate• end-to-end delay• end-to-end jitter• end-to-end loss
T.Sharon-A.Frank
7
Application Taxonomy (2)
Elastic Inelastic
Applications
Tolerant
Loose Delay Bounds
Firm Delay Bounds
IntolerantInteractiveBurst
Best EffortLevel 1
InteractiveBulk
Best EffortLevel 2
AsynchronousBulk
Best EffortLevel 3
TelnetX
NFS Web
FTP E-MailMM-Mail
Fax
Streaming VOD
Medical ImagingCAD Schemes
T.Sharon-A.Frank
8
QoS Types of Service
Best-effort Serviceno/partial guarantees/bounds
Predictive Serviceestimation based on past network behavior
Guaranteed Servicedeterministicstatistical
Current service in most
protocols
T.Sharon-A.Frank
9
Soft RT QoS Guarantees
• DeterministicProvide Bounds on Performance of all
Packets in a Session.
• StatisticalNo more than a Specified Fraction of
Packets will see Performance Below a Certain Specified Value.
T.Sharon-A.Frank
10
Deterministic RT QoS Guarantee
• Delay: no packets delayed more than D time units on E2E basis (T<=D).
• Loss: no packet loss occurs.
• Transit Window: bound transit window(Tmax-Tmin<=E).
• Queuing: the delay of every packet from session i is less than x at queue j.
T.Sharon-A.Frank
11
Statistical RT QoS Guarantee
• Delay: no more than x% of packets have a delay larger than D (PR[T>D]<epsilon)
• Loss: no more than x% of packets in a session are lost PR[Packet-loss]<epsilon
• Queuing: the probability that a packet from session i has a delay greater than x is guaranteed to be less than y at queue j.
T.Sharon-A.Frank
12
Application Taxonomy (3)
Elastic InelasticApplications
Tolerant
Loose Delay Bounds
Firm Delay Bounds
IntolerantInteractiveBurst
Best EffortLevel 1
InteractiveBulk
Best EffortLevel 2
AsynchronousBulk
Best EffortLevel 3
TelnetX
NFS Web
FTP E-MailMM-Mail
Fax
Streaming VOD
Medical ImagingCAD Schemes
Best-effort Service Predictive GuaranteedGrab BandwidthNo Certain Arrival TimeUses Data ImmediatelyNo Admission Control
The Opposite
Care About Average Packet Delay Quantitative Maximum Delay
T.Sharon-A.Frank
13
Example: Playback Applications
• Audio/Video Services
• Soft Real-Time Tolerant Constraints
sender receiver
bufferNetwork
Varying delay transmit
Buffer, Decompress, PlaybackAcquire signal, Digitize, Compress
If arrives late – useless/loss. Playback point: Signal generation time + Fixed offset delay.
Compute offset based on max delay:Offset delay can be adjusted
provided by network based on observed delays
T.Sharon-A.Frank
14
Internet QoS Models
• Adaptation Model– Adapt applications
• hide Internet service from the users – scaling
– Adapt Internet• Differentiated Services (DiffServ) – simple priority
• Extension Model• Integrated Services (IntServ) – resource reservation
T.Sharon-A.Frank
15
Adaptation Model
• Use network Feedback/Scaling• Adapt applications (Scaling)• Minimal changes to Internet (DiffServ)• No need for Resource Reservation:
– “Bandwidth will be infinite”When? Everywhere? Overload?
– “Applications can be adaptive”Too slow? Can users adapt?
– “Simple priority is sufficient”All high priority? Overload?
T.Sharon-A.Frank
16
Scaling
Transparent Scaling - usually by dropping some portion of the data
stream.
Non-transparent Scaling - usually by adjusting parameters in the coding
algorithm.
Means to sub-sample a data streamand only present a fraction of its original content.Scaling types:
T.Sharon-A.Frank
17
Scaling in Audio and Video
Audio– Transparent scaling is difficult because human ear is
sensitive– usually done by changing sampling rate
Video– Temporal scaling (drop frames)– Spatial scaling (reduce resolution)– Frequency scaling (reduce number of DCT coefficients)– Amplitude scaling (reduce color depth)– Color space scaling (reduce number of color entries or even
switch to gray scale)
T.Sharon-A.Frank
21
Stream Management
• Managing streams is all about managing bandwidth, buffers, processing capacity and scheduling priorities – which are all needed in order to realize QoS guarantees.
• This is not as simple as it sounds, and there’s no general agreement as to “how” it should be done.
• For instance: ATM’s QoS (which is very “rich”) has proven to be unworkable (difficult to implement).
• Another technique is the Internet’s RSVP.
T.Sharon-A.Frank
22
Improving QoS in IP Networks
• IETF groups are working on proposals to provide better QoS control in IP networks, i.e., going beyond best effort to provide some assurance for QoS.
• Work in Progress includes Differentiated Services (DiffServ), RSVP and Integrated Services (IntServ).
T.Sharon-A.Frank
23
Differentiated Services (DiffServ)
• Relatively simple, coarse-grained QoS mechanism.
• Deployed in networks without needing to change the operation of the end system application.
• Based around marking packets with a small-fixed bit-pattern, which maps to certain handling and forwarding criteria at each hop.
T.Sharon-A.Frank
24
Extension Model
• Single Service Model– Best-effort services
– Soft real-time services
• Keep Internet Philosophy– Downward compatible
– Common infrastructure
– Unified protocol stack
– Open/public access
– User usage-based pricing
Need New Integrated Services (IntServ) Model?
T.Sharon-A.Frank
25
Resource Reservation
• Pre-allocation of needed resources to guarantee deterministic QoS.
• Allocated resources are dedicated; if not used – remain idle.
• Example: Internet RSVP – Resource reSerVation Protocol.
• If resources cannot be reserved, scaling can be used.
T.Sharon-A.Frank
26
Internet RSVP QoS
The basic organization of RSVP for resource reservation in a distributed system – transport-level control protocol for enabling resource reservations in routers. Interesting characteristic: receiver initiated.
T.Sharon-A.Frank
27
Specifying QoS with Flow Specifications
A flow specification – one way of specifying QoS – a little complex, but it does work (but not via a user controlled interface).
Characteristics of the Input Service Required
• maximum data unit size (bytes)• Token bucket rate (bytes/sec)• Toke bucket size (bytes)• Maximum transmission rate (bytes/sec)
• Loss sensitivity (bytes)• Loss interval (sec)• Burst loss sensitivity (data units)• Minimum delay noticed (sec)• Maximum delay variation (sec)• Quality of guarantee
T.Sharon-A.Frank
28
An Approach to Implementing QoS
The principle of a token bucket algorithm – a “classic” technique for controlling the flow of data (and implementing QoS characteristics).
T.Sharon-A.Frank
29
Integrated Services (IntServ)
• An architecture for providing QOS guarantees in IP networks for individual application sessions.
• Relies on resource reservation.
• Routers need to maintain state info, maintaining records of allocated resources and responding to new Call setup requests on that basis.