SlideShare a Scribd company logo
CAS CS 835
QoS Networking Seminar
What to expect?
• Discuss latest developments and research issues,
current and possible future QoS/CoS technologies
• Queue management, traffic monitoring/shaping,
admission control, routing, …
• Focus on Internet extensions and QoS-sensitive
protocols/applications
• Integrated Services and RSVP
• Differentiated Services
• Multi Protocol Label Switching
• Traffic Engineering (or QoS Routing)
What to expect?
• Proposal:
pre-proposal, full proposal, presentation
• Paper presentation
• 2 quizzes
• Active participation
What is QoS?
• A broad definition: methods for differentiating traffic
and services
• To some, introducing an element of predictability and
consistency into a highly variable best-effort network
• To others, obtaining higher network throughput while
maintaining consistent behavior
• Or, offering resources to high-priority service classes at
the expense of lower-priority classes (conservation law)
• Or, matching network resources to application demands
What is Quality? Service? Why Integrated?
• Quality encompasses data loss, induced delay or latency,
consistency in delays (jitter), efficient use of network
resources, …
• Service means end-to-end communication between
applications (e.g., audio, video, Web browsing), a
protocol type (e.g., TCP, UDP), …
• A single network is better --- unused capacity is available
to others, one network to manage
• Better to give each user/traffic what it wants
• QoS mechanisms for unfairness: managing queuing
behavior, shaping traffic, control admission, routing, …
• Engineering the network is hard, and over-engineering it
is expensive (especially for long-distance links)
Why the recent interest in QoS?
• No longer enough for an ISP to keep traffic flowing, links
up, routing stable --- offer QoS to be more competitive
• QoS is becoming more cost-effective with improved
implementation of differentiation tools (classification,
queue-manipulation, …), more reliable tools for
measuring delivered QoS
• Hard to dimension the network as it is becoming
increasingly hard (if not impossible) to predict traffic
patterns (e.g., 80 local/20 remote rule no longer reliable,
now mostly reversed 25/75)
• If you throw more bandwidth, there will be more demand
(a “vicious cycle”)
Applications of a QoS Network
• Real-time: voice, video, emergency control, stock quotes ...
• Non-real-time (or best-effort): telnet, ftp, …
• Real-time:
- hard with deterministic or guaranteed QoS: no loss, packet
delay less than deadline, difference in delays of 2 packets
less than jitter bound, …
Note: reducing jitter reduces buffers needed to absorb delay
variation at receiving host
- soft with statistical or probabilistic QoS: no more than x%
of packets lost or experience delay greater than deadline, …
• Best-effort do not have such timing constraints
Why end-to-end control not enough?
• Problem: with common FCFS schedulers at routers, delay and delay
variance increase very rapidly with load
• For an M/M/1 model:
average delay = 1 / [ServiceRate - ArrivalRate]
= 1 / [ServiceRate (1 - Load)]
delay variance = 1 / [ (1 - Load)]
• As load increases, buffer overflows and router starts dropping packets
• Solution: TCP reduces load (slow start and congestion avoidance
algorithm)
• 2 TCP users on different hosts sharing the same bottleneck may get
different share of the bandwidth (uncontrolled unfairness) ==>
users should not trust network
• Some users may not “play by the rules” and reduce their sending rates
upon congestion, i.e. not TCP-friendly sources like a voice or video
UDP-based application ==> network should not trust users
2
e
ServiceRat
How to provide guarantees?
• Some form of resource reservation within the network;
hosts can’t “hide” delays
• Challenge: do it asynchronously (i.e., as needed), don’t
reserve peak rate for a voice connection/flow active 30% of
the time
• Want to support as many real-time flows as possible to
increase revenue
• Key question:
Can the network admit a new flow at its requested QoS,
without violating QoS of existing flows?
• A flow has to specify its QoS (Rspec) and also its traffic
pattern (Tspec)
Contract or SLA
• Service Level Agreement between client (subscriber) and
network (provider): the network keeps its promise as long as
the flow conforms to the traffic specification
• The network must monitor/police/shape incoming traffic
• The shape is important: E.g. a gigabit network contracting
with a 100Mbps flow. A big difference between sending one
100Mb packet every second and sending 1Kb packet every
10 microsec.
Traffic Shaping
• Leaky bucket:
- Data packets leak from a bucket of depth “sigma” at a rate
“rho”.
- Network knows that the flow will never inject traffic at a
rate higher than “rho” --- incoming traffic is bounded
- Easy to implement
- Easy for the network to police
- Accommodates well fixed-rate flows (“rho” set to rate),
but does not accommodate variable-rate (bursty) flows
unless “rho” is set to peak rate, which is wasteful
Token Bucket
• Allows “bounded” burstiness
• Tokens generated at rate “rho” are placed in bucket with
depth “sigma”
• An arriving packet has to remove token(s) to be allowed
into the network
• A flow can never send more than [sigma + rho * t] over an
interval of time t
• Thus, the long-term average rate will not exceed rho
• More accurate characterization of bursty flows means better
management of network resources
• Easy to implement, but harder to police
• Can add a peak-rate leaky bucket after a token bucket
Token Bucket
• Example:
Flow A always send at 1 MBps
==> rho = 1 MBps, sigma = 1 byte
Flow B sends at 0.5 MBps for 2 sec, then at 2
MBps for 1sec, then repeats
==> rho = 1 MBps, sigma = 1 MB
We can also this Tspec for Flow A, but that would
be an inaccurate characterization
Link Scheduling
• Challenge: Tspec may change at exit of scheduler (e.g., FCFS)
because of interactions among flows
• FCFS increases worst-case delay and jitter, so we admit less flows
• Solution: non-FCFS scheduler that isolates different flows or
classes of flows (hard, soft, best-effort)
• Emulate TDM without wasting bandwidth
• Virtual Clock provides flows with throughput guarantees and
isolation
• Idea: use logical (rather than real) time
• AR: average data rate required by a flow
• When a packet of size P arrives, VC = VC + P/AR
• Stamp packet with VC
• Transmit packets in increasing order of time stamps
• If a flow has twice AR, it will get double a double rate
Virtual Clock
• If buffer is full, the packet with largest timestamp is
dropped
• Problem: A flow can save up credits and use them to
bump other flows in the future
• Fix: when a packet arrives, catch up with real time first
VC = MAX (real time, VC)
VC = VC + P/AR
• Also, if AI is averaging interval, upon receiving AR*AI
bytes on this flow, if VC > real time + Threshold, then
send advisory to source to reduce its rate
WFQ
• WFQ provides isolation and delay guarantees
• FQ simulates fair bit-by-bit RR by assigning packets
priority based on finishing times under bit-by-bit RR
- E.g. Flow 1 packets of size 5 and 8, Flow 2 packet of size
10: size 5 first, then size 10, then size 8
• Round number increases at a rate inversely proportional to
number of currently active flows
• On packet arrival: recompute round number, compute finish
number, insert in priority queue, if no space drop packet
with largest finish number (max-min fairness)
• Approximation error bounded by max_pkt_size / capacity
• WFQ can assign different weights to different flows
Computing Deterministic Delay Bounds
• If flow is shaped by a token bucket (sigma, rho), all routers
along the “h”-hop path employ WFQ schedulers, and the
flow is assigned a rate of rho at each hop, then end-to-end
delay of a packet is bounded by:
(sigma / rho) + (h * max_pkt_size / rho) +
total approximation error + total propagation delay
• A flow is totally isolated, even if other traffic is not shaped
at all
• Cons: bandwidth and delay are coupled
• WFQ does not bound jitter
• A non-work-conserving scheduler (e.g., Stop-and-Go, jitter-
EDD) may be used to bound jitter
Earliest Due Date (EDD)
• Unlike WFQ, delay bound is independent of bandwidth
requirement
• Packet with earliest deadline selected
• EDD prescribes how to assign deadlines to packets
• Deadline = expected arrival time + delay bound
• “Expected” arrival time is the time the packet should
arrive according to traffic specification (to deal with
bunching of packets downstream)
• Delay bound is the smallest feasible bound, computed
assuming worst-case arrival pattern from all flows
Xmin1 = 10
Xmin2 = 4
3
3 2
Bounding jitter
• Assume we want to eliminate all jitter
• We can achieve this by making the network look like a
constant delay line
• Idea: At the entry of each switch/router, have a regulator
that absorbs delay variations by delaying a packet that
arrived ahead of its local deadline at previous switch
• Traffic characteristics are preserved as traffic passes through
the network
• Schedulers with regulators are called “rate-controlled”
schedulers
• Reduce burstiness within network, thus less buffers needed
• Average packet delay is higher than with work-conserving
schedulers, but that’s fine for hard real-time traffic
Statistical/soft/predictive QoS
• Goal: bound the tail of the measure distribution
• Not a good idea to use worst-case delay bounds since
very few packets (or none!) will actually experience
this worst-case delay
• Computing statistical bounds (e.g., using effective
bandwidths) is usually approximate and often
conservative
• FIFO+ attempts to reduce worst-case delay and jitter
using minimal isolation (and maximal statistical gain)
• At each router, a packet is assigned lower priority if it
left previous routers ahead of measured average delay,
and higher priority if behind average delay
Admission Control with Statistical Guarantees
• Key idea (law of large numbers): as capacity increases,
number of flows that can be supported increases, the
probability that a significant number of flows burst
simultaneously decreases, so the sum of peak rates can
be higher than the available capacity
• As the number of flows increases, the capacity allocated
to each flow is “effectively” its average rate
• Put in enough buffers to make probability of loss low
Measurement-based Admission
• Key assumption: past behavior is a good indicator of
the future
• User tells peak rate
• If peak rate + measured average rate < capacity, admit
• Over time, new call becomes part of average
• Can afford to make some errors for predictive (or
controlled load) service
• Obviously, can admit more calls than admission at
peak rate
Summary
• Performance guarantees can be achieved by combining
traffic shaping and scheduling schemes
• How good the bounds are? Loose or tight?
• How easy to implement these schemes?
• What kind of guarantees they provide?
• How good is the utilization of the network?
• How do clients signal their QoS requirements?
• What is the best path to route a flow?
• How to achieve QoS for multicast flows and with
mobility?
RSVP ReSerVation Protocol
• Designed to handled multicast flows, where multiple receivers may
have different capabilities
• RSVP is receiver-initiated, i.e. receiver generates reservation
• PATH message tentatively reserves resources, RESV makes
reservation (travels as far back up as necessary)
• RSVP supports merging of reservations
• RSVP is decoupled from routing
• PATH and RESV messages periodically sent to refresh state before
timeout (i.e. soft state)
• Multiple senders can share the same reservation
• In contrast, ATM signaling (Q.2931) is sender-initiated, coupled with
routing, uses hard state (i.e. explicit delete), and handles only uniform
QoS
Where to Implement Policies in a Network Topology?
Or, QoS Architectures
• Key to scaling is to maintain levels of hierarchy: core, distribution,
and access
• Link speeds should be faster toward the core of the network
• Access routers generally do not have to handle high packet switching
rates and thus can do complex traffic identification, classification,
policing and marking
• The overhead of implementing QoS policies in the core would affect a
large amount of traffic
• Access routers can mark the IP precedence field in the packet header
• Core routers simply map precedence to queuing or drop actions
• This is similar to the IETF DiffServ architecture (as opposed to the
“less scalable” per-flow IETF IntServ architecture)
Tension: Scalability versus QoS
• Scalability calls for aggregating flows into precedence
classes
• Also, aggregating destinations results in fewer routing
entries, but less granularity of routing decisions
• Performing disaggregation gives finer granularity at a
higher cost
• E.g. more detailed routing advertisments can allow more
granular choice of paths that better satisfy QoS
requirements
• OSPF can advertise multiple costs per link, and compute
multiple shortest paths
• Or, loose source route through a particular service
provider, or multiple addresses/names per host
Other QoS Requirements/Issues
• Pricing/Billing
- can shift demand to off-peak times
- charge more during peak hours
- rational users help the network by helping themselves
• Privacy
• Reliability and availability
• Operating systems support
- reduce costs of data copying, interrupts, …
- real-time CPU scheduling, ...

More Related Content

Similar to QoSintro.PPT

Advanced networking - scheduling and QoS part 1
Advanced networking - scheduling and QoS part 1Advanced networking - scheduling and QoS part 1
Advanced networking - scheduling and QoS part 1
GIST (Gwangju Institute of Science and Technology)
 
Congestion control and quality of service
Congestion control and quality of serviceCongestion control and quality of service
Congestion control and quality of service
JUW Jinnah University for Women
 
RIPE 80: Buffers and Protocols
RIPE 80: Buffers and ProtocolsRIPE 80: Buffers and Protocols
RIPE 80: Buffers and Protocols
APNIC
 
CN Module 5 part 2 2022.pdf
CN Module 5 part 2 2022.pdfCN Module 5 part 2 2022.pdf
CN Module 5 part 2 2022.pdf
MayankRaj687571
 
Traffic Characterization
Traffic CharacterizationTraffic Characterization
Traffic Characterization
Ismail Mukiibi
 
8. TDM Mux_Demux.pdf
8. TDM Mux_Demux.pdf8. TDM Mux_Demux.pdf
8. TDM Mux_Demux.pdf
Tabrezahmed39
 
MULTIMEDIA COMMUNICATION & NETWORKS
MULTIMEDIA COMMUNICATION & NETWORKSMULTIMEDIA COMMUNICATION & NETWORKS
MULTIMEDIA COMMUNICATION & NETWORKS
Kathirvel Ayyaswamy
 
RIPE 76: TCP and BBR
RIPE 76: TCP and BBRRIPE 76: TCP and BBR
RIPE 76: TCP and BBR
APNIC
 
SGSGS
SGSGSSGSGS
Quality of Service
Quality of ServiceQuality of Service
Quality of Service
Abhishek Wadhwa
 
Congestion control
Congestion controlCongestion control
Congestion control
Aman Jaiswal
 
Qo s rsvp......
Qo s rsvp......Qo s rsvp......
Qo s rsvp......
Abhishek Kesharwani
 
Congestion control 1
Congestion control 1Congestion control 1
Congestion control 1
Aman Jaiswal
 
AusNOG 2019: TCP and BBR
AusNOG 2019: TCP and BBRAusNOG 2019: TCP and BBR
AusNOG 2019: TCP and BBR
APNIC
 
High performance browser networking ch1,2,3
High performance browser networking ch1,2,3High performance browser networking ch1,2,3
High performance browser networking ch1,2,3
Seung-Bum Lee
 
9_Network.ppt
9_Network.ppt9_Network.ppt
9_Network.ppt
SushmaShivani
 
Network performance overview
Network  performance overviewNetwork  performance overview
Network performance overview
My cp
 
qualityofserviceabhishek-131013165513-phpapp01.pptx
qualityofserviceabhishek-131013165513-phpapp01.pptxqualityofserviceabhishek-131013165513-phpapp01.pptx
qualityofserviceabhishek-131013165513-phpapp01.pptx
AshwiniKatkar3
 
Network protocols and vulnerabilities
Network protocols and vulnerabilitiesNetwork protocols and vulnerabilities
Network protocols and vulnerabilities
G Prachi
 
Availability Computer Communication Network .pdf
Availability Computer Communication Network .pdfAvailability Computer Communication Network .pdf
Availability Computer Communication Network .pdf
YashShirude1
 

Similar to QoSintro.PPT (20)

Advanced networking - scheduling and QoS part 1
Advanced networking - scheduling and QoS part 1Advanced networking - scheduling and QoS part 1
Advanced networking - scheduling and QoS part 1
 
Congestion control and quality of service
Congestion control and quality of serviceCongestion control and quality of service
Congestion control and quality of service
 
RIPE 80: Buffers and Protocols
RIPE 80: Buffers and ProtocolsRIPE 80: Buffers and Protocols
RIPE 80: Buffers and Protocols
 
CN Module 5 part 2 2022.pdf
CN Module 5 part 2 2022.pdfCN Module 5 part 2 2022.pdf
CN Module 5 part 2 2022.pdf
 
Traffic Characterization
Traffic CharacterizationTraffic Characterization
Traffic Characterization
 
8. TDM Mux_Demux.pdf
8. TDM Mux_Demux.pdf8. TDM Mux_Demux.pdf
8. TDM Mux_Demux.pdf
 
MULTIMEDIA COMMUNICATION & NETWORKS
MULTIMEDIA COMMUNICATION & NETWORKSMULTIMEDIA COMMUNICATION & NETWORKS
MULTIMEDIA COMMUNICATION & NETWORKS
 
RIPE 76: TCP and BBR
RIPE 76: TCP and BBRRIPE 76: TCP and BBR
RIPE 76: TCP and BBR
 
SGSGS
SGSGSSGSGS
SGSGS
 
Quality of Service
Quality of ServiceQuality of Service
Quality of Service
 
Congestion control
Congestion controlCongestion control
Congestion control
 
Qo s rsvp......
Qo s rsvp......Qo s rsvp......
Qo s rsvp......
 
Congestion control 1
Congestion control 1Congestion control 1
Congestion control 1
 
AusNOG 2019: TCP and BBR
AusNOG 2019: TCP and BBRAusNOG 2019: TCP and BBR
AusNOG 2019: TCP and BBR
 
High performance browser networking ch1,2,3
High performance browser networking ch1,2,3High performance browser networking ch1,2,3
High performance browser networking ch1,2,3
 
9_Network.ppt
9_Network.ppt9_Network.ppt
9_Network.ppt
 
Network performance overview
Network  performance overviewNetwork  performance overview
Network performance overview
 
qualityofserviceabhishek-131013165513-phpapp01.pptx
qualityofserviceabhishek-131013165513-phpapp01.pptxqualityofserviceabhishek-131013165513-phpapp01.pptx
qualityofserviceabhishek-131013165513-phpapp01.pptx
 
Network protocols and vulnerabilities
Network protocols and vulnerabilitiesNetwork protocols and vulnerabilities
Network protocols and vulnerabilities
 
Availability Computer Communication Network .pdf
Availability Computer Communication Network .pdfAvailability Computer Communication Network .pdf
Availability Computer Communication Network .pdf
 

Recently uploaded

বাংলাদেশ অর্থনৈতিক সমীক্ষা (Economic Review) ২০২৪ UJS App.pdf
বাংলাদেশ অর্থনৈতিক সমীক্ষা (Economic Review) ২০২৪ UJS App.pdfবাংলাদেশ অর্থনৈতিক সমীক্ষা (Economic Review) ২০২৪ UJS App.pdf
বাংলাদেশ অর্থনৈতিক সমীক্ষা (Economic Review) ২০২৪ UJS App.pdf
eBook.com.bd (প্রয়োজনীয় বাংলা বই)
 
Digital Artefact 1 - Tiny Home Environmental Design
Digital Artefact 1 - Tiny Home Environmental DesignDigital Artefact 1 - Tiny Home Environmental Design
Digital Artefact 1 - Tiny Home Environmental Design
amberjdewit93
 
C1 Rubenstein AP HuG xxxxxxxxxxxxxx.pptx
C1 Rubenstein AP HuG xxxxxxxxxxxxxx.pptxC1 Rubenstein AP HuG xxxxxxxxxxxxxx.pptx
C1 Rubenstein AP HuG xxxxxxxxxxxxxx.pptx
mulvey2
 
How to Fix the Import Error in the Odoo 17
How to Fix the Import Error in the Odoo 17How to Fix the Import Error in the Odoo 17
How to Fix the Import Error in the Odoo 17
Celine George
 
Chapter 4 - Islamic Financial Institutions in Malaysia.pptx
Chapter 4 - Islamic Financial Institutions in Malaysia.pptxChapter 4 - Islamic Financial Institutions in Malaysia.pptx
Chapter 4 - Islamic Financial Institutions in Malaysia.pptx
Mohd Adib Abd Muin, Senior Lecturer at Universiti Utara Malaysia
 
ISO/IEC 27001, ISO/IEC 42001, and GDPR: Best Practices for Implementation and...
ISO/IEC 27001, ISO/IEC 42001, and GDPR: Best Practices for Implementation and...ISO/IEC 27001, ISO/IEC 42001, and GDPR: Best Practices for Implementation and...
ISO/IEC 27001, ISO/IEC 42001, and GDPR: Best Practices for Implementation and...
PECB
 
The Diamonds of 2023-2024 in the IGRA collection
The Diamonds of 2023-2024 in the IGRA collectionThe Diamonds of 2023-2024 in the IGRA collection
The Diamonds of 2023-2024 in the IGRA collection
Israel Genealogy Research Association
 
Azure Interview Questions and Answers PDF By ScholarHat
Azure Interview Questions and Answers PDF By ScholarHatAzure Interview Questions and Answers PDF By ScholarHat
Azure Interview Questions and Answers PDF By ScholarHat
Scholarhat
 
DRUGS AND ITS classification slide share
DRUGS AND ITS classification slide shareDRUGS AND ITS classification slide share
DRUGS AND ITS classification slide share
taiba qazi
 
Assessment and Planning in Educational technology.pptx
Assessment and Planning in Educational technology.pptxAssessment and Planning in Educational technology.pptx
Assessment and Planning in Educational technology.pptx
Kavitha Krishnan
 
Your Skill Boost Masterclass: Strategies for Effective Upskilling
Your Skill Boost Masterclass: Strategies for Effective UpskillingYour Skill Boost Masterclass: Strategies for Effective Upskilling
Your Skill Boost Masterclass: Strategies for Effective Upskilling
Excellence Foundation for South Sudan
 
Liberal Approach to the Study of Indian Politics.pdf
Liberal Approach to the Study of Indian Politics.pdfLiberal Approach to the Study of Indian Politics.pdf
Liberal Approach to the Study of Indian Politics.pdf
WaniBasim
 
clinical examination of hip joint (1).pdf
clinical examination of hip joint (1).pdfclinical examination of hip joint (1).pdf
clinical examination of hip joint (1).pdf
Priyankaranawat4
 
South African Journal of Science: Writing with integrity workshop (2024)
South African Journal of Science: Writing with integrity workshop (2024)South African Journal of Science: Writing with integrity workshop (2024)
South African Journal of Science: Writing with integrity workshop (2024)
Academy of Science of South Africa
 
Natural birth techniques - Mrs.Akanksha Trivedi Rama University
Natural birth techniques - Mrs.Akanksha Trivedi Rama UniversityNatural birth techniques - Mrs.Akanksha Trivedi Rama University
Natural birth techniques - Mrs.Akanksha Trivedi Rama University
Akanksha trivedi rama nursing college kanpur.
 
Lapbook sobre os Regimes Totalitários.pdf
Lapbook sobre os Regimes Totalitários.pdfLapbook sobre os Regimes Totalitários.pdf
Lapbook sobre os Regimes Totalitários.pdf
Jean Carlos Nunes Paixão
 
A Survey of Techniques for Maximizing LLM Performance.pptx
A Survey of Techniques for Maximizing LLM Performance.pptxA Survey of Techniques for Maximizing LLM Performance.pptx
A Survey of Techniques for Maximizing LLM Performance.pptx
thanhdowork
 
Pollock and Snow "DEIA in the Scholarly Landscape, Session One: Setting Expec...
Pollock and Snow "DEIA in the Scholarly Landscape, Session One: Setting Expec...Pollock and Snow "DEIA in the Scholarly Landscape, Session One: Setting Expec...
Pollock and Snow "DEIA in the Scholarly Landscape, Session One: Setting Expec...
National Information Standards Organization (NISO)
 
Executive Directors Chat Leveraging AI for Diversity, Equity, and Inclusion
Executive Directors Chat  Leveraging AI for Diversity, Equity, and InclusionExecutive Directors Chat  Leveraging AI for Diversity, Equity, and Inclusion
Executive Directors Chat Leveraging AI for Diversity, Equity, and Inclusion
TechSoup
 
ANATOMY AND BIOMECHANICS OF HIP JOINT.pdf
ANATOMY AND BIOMECHANICS OF HIP JOINT.pdfANATOMY AND BIOMECHANICS OF HIP JOINT.pdf
ANATOMY AND BIOMECHANICS OF HIP JOINT.pdf
Priyankaranawat4
 

Recently uploaded (20)

বাংলাদেশ অর্থনৈতিক সমীক্ষা (Economic Review) ২০২৪ UJS App.pdf
বাংলাদেশ অর্থনৈতিক সমীক্ষা (Economic Review) ২০২৪ UJS App.pdfবাংলাদেশ অর্থনৈতিক সমীক্ষা (Economic Review) ২০২৪ UJS App.pdf
বাংলাদেশ অর্থনৈতিক সমীক্ষা (Economic Review) ২০২৪ UJS App.pdf
 
Digital Artefact 1 - Tiny Home Environmental Design
Digital Artefact 1 - Tiny Home Environmental DesignDigital Artefact 1 - Tiny Home Environmental Design
Digital Artefact 1 - Tiny Home Environmental Design
 
C1 Rubenstein AP HuG xxxxxxxxxxxxxx.pptx
C1 Rubenstein AP HuG xxxxxxxxxxxxxx.pptxC1 Rubenstein AP HuG xxxxxxxxxxxxxx.pptx
C1 Rubenstein AP HuG xxxxxxxxxxxxxx.pptx
 
How to Fix the Import Error in the Odoo 17
How to Fix the Import Error in the Odoo 17How to Fix the Import Error in the Odoo 17
How to Fix the Import Error in the Odoo 17
 
Chapter 4 - Islamic Financial Institutions in Malaysia.pptx
Chapter 4 - Islamic Financial Institutions in Malaysia.pptxChapter 4 - Islamic Financial Institutions in Malaysia.pptx
Chapter 4 - Islamic Financial Institutions in Malaysia.pptx
 
ISO/IEC 27001, ISO/IEC 42001, and GDPR: Best Practices for Implementation and...
ISO/IEC 27001, ISO/IEC 42001, and GDPR: Best Practices for Implementation and...ISO/IEC 27001, ISO/IEC 42001, and GDPR: Best Practices for Implementation and...
ISO/IEC 27001, ISO/IEC 42001, and GDPR: Best Practices for Implementation and...
 
The Diamonds of 2023-2024 in the IGRA collection
The Diamonds of 2023-2024 in the IGRA collectionThe Diamonds of 2023-2024 in the IGRA collection
The Diamonds of 2023-2024 in the IGRA collection
 
Azure Interview Questions and Answers PDF By ScholarHat
Azure Interview Questions and Answers PDF By ScholarHatAzure Interview Questions and Answers PDF By ScholarHat
Azure Interview Questions and Answers PDF By ScholarHat
 
DRUGS AND ITS classification slide share
DRUGS AND ITS classification slide shareDRUGS AND ITS classification slide share
DRUGS AND ITS classification slide share
 
Assessment and Planning in Educational technology.pptx
Assessment and Planning in Educational technology.pptxAssessment and Planning in Educational technology.pptx
Assessment and Planning in Educational technology.pptx
 
Your Skill Boost Masterclass: Strategies for Effective Upskilling
Your Skill Boost Masterclass: Strategies for Effective UpskillingYour Skill Boost Masterclass: Strategies for Effective Upskilling
Your Skill Boost Masterclass: Strategies for Effective Upskilling
 
Liberal Approach to the Study of Indian Politics.pdf
Liberal Approach to the Study of Indian Politics.pdfLiberal Approach to the Study of Indian Politics.pdf
Liberal Approach to the Study of Indian Politics.pdf
 
clinical examination of hip joint (1).pdf
clinical examination of hip joint (1).pdfclinical examination of hip joint (1).pdf
clinical examination of hip joint (1).pdf
 
South African Journal of Science: Writing with integrity workshop (2024)
South African Journal of Science: Writing with integrity workshop (2024)South African Journal of Science: Writing with integrity workshop (2024)
South African Journal of Science: Writing with integrity workshop (2024)
 
Natural birth techniques - Mrs.Akanksha Trivedi Rama University
Natural birth techniques - Mrs.Akanksha Trivedi Rama UniversityNatural birth techniques - Mrs.Akanksha Trivedi Rama University
Natural birth techniques - Mrs.Akanksha Trivedi Rama University
 
Lapbook sobre os Regimes Totalitários.pdf
Lapbook sobre os Regimes Totalitários.pdfLapbook sobre os Regimes Totalitários.pdf
Lapbook sobre os Regimes Totalitários.pdf
 
A Survey of Techniques for Maximizing LLM Performance.pptx
A Survey of Techniques for Maximizing LLM Performance.pptxA Survey of Techniques for Maximizing LLM Performance.pptx
A Survey of Techniques for Maximizing LLM Performance.pptx
 
Pollock and Snow "DEIA in the Scholarly Landscape, Session One: Setting Expec...
Pollock and Snow "DEIA in the Scholarly Landscape, Session One: Setting Expec...Pollock and Snow "DEIA in the Scholarly Landscape, Session One: Setting Expec...
Pollock and Snow "DEIA in the Scholarly Landscape, Session One: Setting Expec...
 
Executive Directors Chat Leveraging AI for Diversity, Equity, and Inclusion
Executive Directors Chat  Leveraging AI for Diversity, Equity, and InclusionExecutive Directors Chat  Leveraging AI for Diversity, Equity, and Inclusion
Executive Directors Chat Leveraging AI for Diversity, Equity, and Inclusion
 
ANATOMY AND BIOMECHANICS OF HIP JOINT.pdf
ANATOMY AND BIOMECHANICS OF HIP JOINT.pdfANATOMY AND BIOMECHANICS OF HIP JOINT.pdf
ANATOMY AND BIOMECHANICS OF HIP JOINT.pdf
 

QoSintro.PPT

  • 1. CAS CS 835 QoS Networking Seminar
  • 2. What to expect? • Discuss latest developments and research issues, current and possible future QoS/CoS technologies • Queue management, traffic monitoring/shaping, admission control, routing, … • Focus on Internet extensions and QoS-sensitive protocols/applications • Integrated Services and RSVP • Differentiated Services • Multi Protocol Label Switching • Traffic Engineering (or QoS Routing)
  • 3. What to expect? • Proposal: pre-proposal, full proposal, presentation • Paper presentation • 2 quizzes • Active participation
  • 4. What is QoS? • A broad definition: methods for differentiating traffic and services • To some, introducing an element of predictability and consistency into a highly variable best-effort network • To others, obtaining higher network throughput while maintaining consistent behavior • Or, offering resources to high-priority service classes at the expense of lower-priority classes (conservation law) • Or, matching network resources to application demands
  • 5. What is Quality? Service? Why Integrated? • Quality encompasses data loss, induced delay or latency, consistency in delays (jitter), efficient use of network resources, … • Service means end-to-end communication between applications (e.g., audio, video, Web browsing), a protocol type (e.g., TCP, UDP), … • A single network is better --- unused capacity is available to others, one network to manage • Better to give each user/traffic what it wants • QoS mechanisms for unfairness: managing queuing behavior, shaping traffic, control admission, routing, … • Engineering the network is hard, and over-engineering it is expensive (especially for long-distance links)
  • 6. Why the recent interest in QoS? • No longer enough for an ISP to keep traffic flowing, links up, routing stable --- offer QoS to be more competitive • QoS is becoming more cost-effective with improved implementation of differentiation tools (classification, queue-manipulation, …), more reliable tools for measuring delivered QoS • Hard to dimension the network as it is becoming increasingly hard (if not impossible) to predict traffic patterns (e.g., 80 local/20 remote rule no longer reliable, now mostly reversed 25/75) • If you throw more bandwidth, there will be more demand (a “vicious cycle”)
  • 7. Applications of a QoS Network • Real-time: voice, video, emergency control, stock quotes ... • Non-real-time (or best-effort): telnet, ftp, … • Real-time: - hard with deterministic or guaranteed QoS: no loss, packet delay less than deadline, difference in delays of 2 packets less than jitter bound, … Note: reducing jitter reduces buffers needed to absorb delay variation at receiving host - soft with statistical or probabilistic QoS: no more than x% of packets lost or experience delay greater than deadline, … • Best-effort do not have such timing constraints
  • 8. Why end-to-end control not enough? • Problem: with common FCFS schedulers at routers, delay and delay variance increase very rapidly with load • For an M/M/1 model: average delay = 1 / [ServiceRate - ArrivalRate] = 1 / [ServiceRate (1 - Load)] delay variance = 1 / [ (1 - Load)] • As load increases, buffer overflows and router starts dropping packets • Solution: TCP reduces load (slow start and congestion avoidance algorithm) • 2 TCP users on different hosts sharing the same bottleneck may get different share of the bandwidth (uncontrolled unfairness) ==> users should not trust network • Some users may not “play by the rules” and reduce their sending rates upon congestion, i.e. not TCP-friendly sources like a voice or video UDP-based application ==> network should not trust users 2 e ServiceRat
  • 9. How to provide guarantees? • Some form of resource reservation within the network; hosts can’t “hide” delays • Challenge: do it asynchronously (i.e., as needed), don’t reserve peak rate for a voice connection/flow active 30% of the time • Want to support as many real-time flows as possible to increase revenue • Key question: Can the network admit a new flow at its requested QoS, without violating QoS of existing flows? • A flow has to specify its QoS (Rspec) and also its traffic pattern (Tspec)
  • 10. Contract or SLA • Service Level Agreement between client (subscriber) and network (provider): the network keeps its promise as long as the flow conforms to the traffic specification • The network must monitor/police/shape incoming traffic • The shape is important: E.g. a gigabit network contracting with a 100Mbps flow. A big difference between sending one 100Mb packet every second and sending 1Kb packet every 10 microsec.
  • 11. Traffic Shaping • Leaky bucket: - Data packets leak from a bucket of depth “sigma” at a rate “rho”. - Network knows that the flow will never inject traffic at a rate higher than “rho” --- incoming traffic is bounded - Easy to implement - Easy for the network to police - Accommodates well fixed-rate flows (“rho” set to rate), but does not accommodate variable-rate (bursty) flows unless “rho” is set to peak rate, which is wasteful
  • 12. Token Bucket • Allows “bounded” burstiness • Tokens generated at rate “rho” are placed in bucket with depth “sigma” • An arriving packet has to remove token(s) to be allowed into the network • A flow can never send more than [sigma + rho * t] over an interval of time t • Thus, the long-term average rate will not exceed rho • More accurate characterization of bursty flows means better management of network resources • Easy to implement, but harder to police • Can add a peak-rate leaky bucket after a token bucket
  • 13. Token Bucket • Example: Flow A always send at 1 MBps ==> rho = 1 MBps, sigma = 1 byte Flow B sends at 0.5 MBps for 2 sec, then at 2 MBps for 1sec, then repeats ==> rho = 1 MBps, sigma = 1 MB We can also this Tspec for Flow A, but that would be an inaccurate characterization
  • 14. Link Scheduling • Challenge: Tspec may change at exit of scheduler (e.g., FCFS) because of interactions among flows • FCFS increases worst-case delay and jitter, so we admit less flows • Solution: non-FCFS scheduler that isolates different flows or classes of flows (hard, soft, best-effort) • Emulate TDM without wasting bandwidth • Virtual Clock provides flows with throughput guarantees and isolation • Idea: use logical (rather than real) time • AR: average data rate required by a flow • When a packet of size P arrives, VC = VC + P/AR • Stamp packet with VC • Transmit packets in increasing order of time stamps • If a flow has twice AR, it will get double a double rate
  • 15. Virtual Clock • If buffer is full, the packet with largest timestamp is dropped • Problem: A flow can save up credits and use them to bump other flows in the future • Fix: when a packet arrives, catch up with real time first VC = MAX (real time, VC) VC = VC + P/AR • Also, if AI is averaging interval, upon receiving AR*AI bytes on this flow, if VC > real time + Threshold, then send advisory to source to reduce its rate
  • 16. WFQ • WFQ provides isolation and delay guarantees • FQ simulates fair bit-by-bit RR by assigning packets priority based on finishing times under bit-by-bit RR - E.g. Flow 1 packets of size 5 and 8, Flow 2 packet of size 10: size 5 first, then size 10, then size 8 • Round number increases at a rate inversely proportional to number of currently active flows • On packet arrival: recompute round number, compute finish number, insert in priority queue, if no space drop packet with largest finish number (max-min fairness) • Approximation error bounded by max_pkt_size / capacity • WFQ can assign different weights to different flows
  • 17. Computing Deterministic Delay Bounds • If flow is shaped by a token bucket (sigma, rho), all routers along the “h”-hop path employ WFQ schedulers, and the flow is assigned a rate of rho at each hop, then end-to-end delay of a packet is bounded by: (sigma / rho) + (h * max_pkt_size / rho) + total approximation error + total propagation delay • A flow is totally isolated, even if other traffic is not shaped at all • Cons: bandwidth and delay are coupled • WFQ does not bound jitter • A non-work-conserving scheduler (e.g., Stop-and-Go, jitter- EDD) may be used to bound jitter
  • 18. Earliest Due Date (EDD) • Unlike WFQ, delay bound is independent of bandwidth requirement • Packet with earliest deadline selected • EDD prescribes how to assign deadlines to packets • Deadline = expected arrival time + delay bound • “Expected” arrival time is the time the packet should arrive according to traffic specification (to deal with bunching of packets downstream) • Delay bound is the smallest feasible bound, computed assuming worst-case arrival pattern from all flows Xmin1 = 10 Xmin2 = 4 3 3 2
  • 19. Bounding jitter • Assume we want to eliminate all jitter • We can achieve this by making the network look like a constant delay line • Idea: At the entry of each switch/router, have a regulator that absorbs delay variations by delaying a packet that arrived ahead of its local deadline at previous switch • Traffic characteristics are preserved as traffic passes through the network • Schedulers with regulators are called “rate-controlled” schedulers • Reduce burstiness within network, thus less buffers needed • Average packet delay is higher than with work-conserving schedulers, but that’s fine for hard real-time traffic
  • 20. Statistical/soft/predictive QoS • Goal: bound the tail of the measure distribution • Not a good idea to use worst-case delay bounds since very few packets (or none!) will actually experience this worst-case delay • Computing statistical bounds (e.g., using effective bandwidths) is usually approximate and often conservative • FIFO+ attempts to reduce worst-case delay and jitter using minimal isolation (and maximal statistical gain) • At each router, a packet is assigned lower priority if it left previous routers ahead of measured average delay, and higher priority if behind average delay
  • 21. Admission Control with Statistical Guarantees • Key idea (law of large numbers): as capacity increases, number of flows that can be supported increases, the probability that a significant number of flows burst simultaneously decreases, so the sum of peak rates can be higher than the available capacity • As the number of flows increases, the capacity allocated to each flow is “effectively” its average rate • Put in enough buffers to make probability of loss low
  • 22. Measurement-based Admission • Key assumption: past behavior is a good indicator of the future • User tells peak rate • If peak rate + measured average rate < capacity, admit • Over time, new call becomes part of average • Can afford to make some errors for predictive (or controlled load) service • Obviously, can admit more calls than admission at peak rate
  • 23. Summary • Performance guarantees can be achieved by combining traffic shaping and scheduling schemes • How good the bounds are? Loose or tight? • How easy to implement these schemes? • What kind of guarantees they provide? • How good is the utilization of the network? • How do clients signal their QoS requirements? • What is the best path to route a flow? • How to achieve QoS for multicast flows and with mobility?
  • 24. RSVP ReSerVation Protocol • Designed to handled multicast flows, where multiple receivers may have different capabilities • RSVP is receiver-initiated, i.e. receiver generates reservation • PATH message tentatively reserves resources, RESV makes reservation (travels as far back up as necessary) • RSVP supports merging of reservations • RSVP is decoupled from routing • PATH and RESV messages periodically sent to refresh state before timeout (i.e. soft state) • Multiple senders can share the same reservation • In contrast, ATM signaling (Q.2931) is sender-initiated, coupled with routing, uses hard state (i.e. explicit delete), and handles only uniform QoS
  • 25. Where to Implement Policies in a Network Topology? Or, QoS Architectures • Key to scaling is to maintain levels of hierarchy: core, distribution, and access • Link speeds should be faster toward the core of the network • Access routers generally do not have to handle high packet switching rates and thus can do complex traffic identification, classification, policing and marking • The overhead of implementing QoS policies in the core would affect a large amount of traffic • Access routers can mark the IP precedence field in the packet header • Core routers simply map precedence to queuing or drop actions • This is similar to the IETF DiffServ architecture (as opposed to the “less scalable” per-flow IETF IntServ architecture)
  • 26. Tension: Scalability versus QoS • Scalability calls for aggregating flows into precedence classes • Also, aggregating destinations results in fewer routing entries, but less granularity of routing decisions • Performing disaggregation gives finer granularity at a higher cost • E.g. more detailed routing advertisments can allow more granular choice of paths that better satisfy QoS requirements • OSPF can advertise multiple costs per link, and compute multiple shortest paths • Or, loose source route through a particular service provider, or multiple addresses/names per host
  • 27. Other QoS Requirements/Issues • Pricing/Billing - can shift demand to off-peak times - charge more during peak hours - rational users help the network by helping themselves • Privacy • Reliability and availability • Operating systems support - reduce costs of data copying, interrupts, … - real-time CPU scheduling, ...