QUALITY OF SERVICE IN HETEROGENEOUS NETWORKS: CURRENT STATUS, EXAMPLES, AND OPEN ISSUES Kostas Pentikousis and Milla Immonen
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
QOS DEFINED: RESEARCH •QoS is a mature, well-researched topic •Packet Classification •Shaping and Policing •Buffer acceptance and queue management •Scheduling algorithms •However, QoS seems to fade as a research topic •The research community is more interested in • Network measurements and analysis • TCP and TCP-friendly protocol performance over multi-gigabit pipes, multi-hop wireless • P2P, Routing, Overlays • Security, Gaming
7.7.2006
2
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
QOS DEFINED: DEPLOYMENT •Several deployment attempts did not succeed in making QoS ubiquitous •Integrated Services (early 90s) •Differentiated Services (mid 90s), QBone •ATM, MPLS, wireless ATM vs. IEEE 802.11, … •In addition, overprovisioning seems to be more widespread, more attractive than deploying QoS •Is that a bad thing? •Is overprovisioning the solution? •Is it enough? Why not?
7.7.2006
3
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
QOS ACCORDING TO NETWORK OPERATORS •The ability of the network to deliver traffic within a certain range of values of •Throughput (b/s) •Delay (s) and Jitter (s) •Packet Loss (%) due to congestion and corruption •Out-of-order delivery •Typically, network operators agree to meet such predetermined ranges X% of the time •5 "9s" = 99.999% = fail to meet QoS 5 min/year •3 "9s" = 99.9% = ~9 hours/year •2 "9s" = 99% = ~3.65 days/year
7.7.2006
4
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
REASONS TO DEPLOY QOS • Huston (2000) notes three main reasons for introducing QoS in the Internet architecture 1. High-quality support for IP voice and video 2. Service response management 3. A differentiated Internet access service, providing a network client with a range of service-quality levels at a range of prices • All three seem to be centered around network operator interests
7.7.2006
5
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
QOS: CAVEATS •QoS does not create any (new) capacity, but •is the means to deal with lack of resources • if capacity is scarce, resource sharing requires frugality • "proper" resource allocation •QoS introduces a certain degree of complexity in the network •Ordinary end users expect something else than "9s" • Ulseth (2004) A network QoS class definition is not sufficient for the user • Networks not belonging to the operator domain such as WLAN are not included, • Media processing (voice coding) and other terminal related characteristics are not included 7.7.2006
6
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
OVERPROVISIONING: IT AIN'T BAD Overprovisioning is not a new idea Factor of safety (a.k.a. factor of ignorance) Eighteenth century iron bridges had a factor of safety of 3-7x the calculated load
The Harilaos Trikoupis bridge connecting Rio-Antirio in SW Greece
7.7.2006
7
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
OVERPROVISIONING (2) Redundancy RAID: increase fault tolerance/reliability and/or performance
Availability A. S. Tanenbaum asks: when was the last time you picked up the phone and got a busy tone?
7.7.2006
8
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
OVERPROVISIONING (3) Ease of use Memory garbage collection Peak performance Do you really need a dual core 64-bit CPU at 3 GHz? Infinitesimal extra cost Ride the Ethernet upgrade wave: 10 102
103 Mb/s
Deploy 802.11a/b/g although either of the 3 would be more than enough
7.7.2006
9
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
OVERPROVISIONING vs. QoS Overprovisioning •"throwing money at a problem" •"inefficient" •"ineffective" •"wasteful" •it simply sounds wrong
7.7.2006
10
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
OVERPROVISIONING vs. QoS (2) •But, then, which of the two figures below do you consider more efficient, effective, or wasteful?
7.7.2006
11
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
OVERPROVISIONING vs. QoS (3) •Do you consider this overprovisioned?
Interstate 105/Interstate 110 Interchange, Los Angeles, California,USA Source: http://www.fhwa.dot.gov/eihd/i105i110.htm
7.7.2006
12
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
OVERPROVISIONING vs. QoS (4) •Well, the I-105/I-110 interchange (completed in 1993) received an Award of Merit in 1996 by the US DOT Federal Highway Administration, for Excellence in Highway Design
•This is an intermodal interchange: "It has three levels of transfer facilities, including direct HOV connectors between the two freeways" •HOV? That's a form of QoS, isn't it? 7.7.2006
13
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
OVERPROVISIONING vs. QoS (5) •So, perhaps, excess capacity may not be a "bad thing" and can coincide with QoS •More seriously, considering total cost of ownership (TCO), can it be that overprovisioning alone is the right thing and no QoS is needed? •Networkers need to determine whether QoS is •deployable? •reliable? •cost-effective? •the only viable solution?
7.7.2006
14
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
QoS vs. CHARGING •QoS has been typically associated with tiered, e.g. bronze, silver, gold and platinum services, and policing/charging schemes
•Charging, the argument goes, is an effective means for enforcing QoS •Flat pricing: all packets are marked as platinum
7.7.2006
15
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
QoS vs. CHARGING (2) •QoS is by no means identical to tiered charging; it does not have to be amalgamated with tiered billing, and may have nothing to do with charging per packet
•Instead, QoS can provide the framework to deliver a service in the first place •Case in point? Maxinetti, a triple play service (IPTV + VoIP + Broadband Internet access) offered in the metropolitan Helsinki area in Finland
7.7.2006
16
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
QoS AS A BUSINESS ENABLER: maxinetti •End users pay X euros for a given IPTV channel package, Y euros for VoIP, Z euros for Internet access, or buy the bundle at a discount
•The operator, Maxisat, must differentiate flows from different services 7.7.2006
17
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
PRAGMATIC QoS •Differentiating between classes of traffic is easier, more scalable •More like traffic prioritization •Given 8 Mb/s of downlink capacity, must provide •sufficient & sustained bandwidth (IPTV: 3-5 Mb/s) •low end-to-end delay for VoIP •low jitter for VoIP and IPTV •operational reliability and low packet loss rate •Maxisat could have employed DiffServ, IntServ, or any other more elegant or sophisticated QoS scheme. They didn't.
7.7.2006
18
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
"QoS THAT WORKS"
Gigabit Song ring & residential cabling infrastructure
Use IEEE 802.1P CoS and IP TOS fields to deliver bundled digital IPTV, VoIP and broadband Internet access 7.7.2006
Cope with standard equipment (keep costs low, increase reliability)
DSLAM handles downstream classification
19
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
Maxinetti •It works :) and shows that CoS may be enough and it should be the first step to a tier-service system. •Maxisat opted for rudimentary downlink flow classification using CoS at Layer 2 and ToS at Layer 3 to provide end-to-end QoS •Why? Reliability and cost effectiveness
•Yet this is a closed, homogeneous network infrastructure, under single administrative control •What about end-to-end cross AD QoS? First, let's see what kinds of QoS frameworks exist
7.7.2006
20
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
ATM • ATM allows point to point and point to multipoint virtual circuit to be requested with pre-specified QOS. • Rich set of QoS mechanisms with a wide variety of service categories or QoS descriptors. • Class A (AAL1) - Constant Bit Rate (CBR) service • bit rate is constant • Class B (AAL2) - Variable Bit Rate (VBR) servicet • the bit rate is variable but requires a bounded delay for delivery. • Class C (AAL3/4 or AAL5) - Connection-oriented data service • connection is set up before data is transferred, • variable bit rate • does not require bounded delay for delivery • Class D (AAL3/4 or AAL5) - Connectionless data service • datagram traffic and in general • data network applications where no connection is set up before data is transferred AAL= ATM Adaptation Layer
7.7.2006
21
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
QoS IN CELLULAR NETWORKS G
Frequency (MHz)
1
450/900
2
900/1800/1900
Technology
Data rate (kb/s)
QoS
Analogue voice
1.2 kbits/s
NO
CDMA & TDMA voice
9.6 kbits/s
NO
2.5
same
GPRS, EDGE, and HSCSD (in addition to digital voice)
Up to 76 kbits/s
Available (not used)
3
2000
WCDMA
Up to 384 kbits/s
Available
3.5
same
HSDPA (extension to WCDMA)
Theoretically up to 10,8 Mbits/s Current situation 1 Mbits/s
Available
7.7.2006
22
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
QoS MECHANISMS IN UMTS • Versatile needs of applications lead to traffic prioritising • Traffic can be divided into 4 QoS classes 1. Conversational class 2. Streaming class 3. Interactive class 4. Background class • Biggest difference between these classes is the delay sensitivity
7.7.2006
23
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
UMTS QoS CLASSES
VoIP
Telephony speech
Conversational Video telephony E-mail Podcasts
Real-time video
Background
Streaming
Messaging
Radio
File downloads Web browsing
Interactive Games 7.7.2006
IM
DB & server access 24
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
3G - Universal Mobile Telecommunications Service (UMTS) Architecture
UTRAN = UMTS Terrestrial Radio Access Network Node B = Base station
Applications Content
RNC = Radio Network Controller
Internet / Intranet /ISP
Wap WWW, Application servers Gateway E-mail
GGSN= Gateway GPRS Node SGSN= Serving GPRS Support Node
Ethernet NODE B
Iub
RNS Iu RNC
NODE B
GGSN
Core Network
Iub Iur
NODE B
IP Firewall
Iub
SGSN
RNC NODE B
Iub
Iu RNS UTRAN
7.7.2006
25
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
3G TEST PLATFORM • Provides access for real WCDMA terminals to Core network and Internet • Enables easily the endto-end service testing in 3G environment • Makes optimisation and enhancements of QoSmechanism in UTRAN and Core network possible without intruding upon public network Iub = UMTS interface between radio network controller and base station Gi = Interface between gateway GPRS support node and external network
7.7.2006
26
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
3G AND BEYOND TEST NETWORK
Sensor network
WLAN •Session mobility •Terminal mobility
3G
Internet
7.7.2006
27
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
WiMAX Enhancements for VTT's Laboratory Network
Testbed environment •Open development and testing environment •Connectet to VTT laboratory network and to Internet through FUNET and GEANT
Research and implementation work •Video and Voice over IP services
VHO = vertical handover MIP = mobile IP HIP = host identity protocol 7.7.2006 DiffServ = differentiated services
•Fast rate control supporting cross-layer information •Mobility and multi-access enhancements •VHOs between different access networks •Subscriber station (SS) and base station (BS) side solutions to gather and process channel and network state information 28
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
IEEE 802.11 WLAN: FAMILY OF STANDARDS •IEEE Subgroups has standardised •physical layer of OSI • 802.11b: 11 Mbits/s in 2.4 GHz band • 802.11a: 54 Mbits/s in 5 GHz band • 802.11g: 54 Mbits/s in 2.4 GHz band •MAC sub layer •Provide transparent interface for the higher layer users •existing network protocols run over IEEE 802.11 WLAN WLAN can be thought as a wireless version of the Ethernet, which provides best-effort service
7.7.2006
29
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
WLAN 802.11 NEW STANDARDS •IEEE 802.11e, 802.11f and 802.11i under standardisation process •IEEE 802.11e will provide enhanced QoS mechanisms •IEEE 802.11f Inter-Access Point Protocol (IAPP) •IEEE 802.11i will provide security mechanisms
7.7.2006
30
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
IEEE 802.11 MAC SUBLAYER
•Distributed coordination function (DCF) • ”listen before talk” • works based on a Carrier Sense Multiple Access (CSMA) • DIFS = DCF Interframe Space
SEND AFTER DIFS SECONDS
FREE
SENSE THE CHANNEL FREE
BUSY
SENSE THE CHANNEL FOR ADDITIONAL RANDOM TIME
FREE
7.7.2006
BUSY
BACKOFF BUSY
SENSE THE CHANNEL
31
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
WLAN IEEE 802.11e •QoS Standard •Work is Final •Goal: •enhance the access mechanisms of IEEE802.11 •provide service differentiation •Enhanced DCF (EDCF) •extension of DCF •allows traffic to be classified into 8 different traffic classes, by modifying the backoff times
7.7.2006
32
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
MONITORING QoS •Close to network traffic measurements •Main difference: result analysis •in QoS analysis network traffic is used as a tool to reveal the performance characteristics •delay •maximum throughput •jitter, etc. •passive measurement methods •monitoring existing traffic •active methods •traffic is generated for the measurements
7.7.2006
33
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
Mobility Management Mechanisms • During the last years Mobile IP (MIP) has become "de facto" mobility management protocol for Internet • Although MIP is workable, it has several defects • handovers may not be fast and smooth • message overhead can be significat if Home Agent is distant • QoS implementations are problematic due to tunneling • It does not support micromobility • rely on IP addresses hard to identify host which are beyond NATs etc. • Several new protocols and enhancements have been proposed • Hierarchial MIPv6, Cellular IP, HAWAII for micromobility • Host identity protocol (HIP) for security, multihoming, end-host mobility 7.7.2006
34
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
SUBJECTIVE QoS vs. OBJECTIVE QoS • •
7.7.2006
User experience is the one that counts! Subjective QoS is the service quality from the user perspective • measuring subjective QoS is done by user tests • only reliable way • Mean Opinion Score (MOS) tests are often used expensive and time consuming
35
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
SUBJECTIVE QoS vs. OBJECTIVE QoS • User experience is the one that counts! • Subjective QoS is the service quality from the user perspective • measuring subjective QoS is done by user tests • only reliable way
7.7.2006
36
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
SUBJECTIVE QoS vs. OBJECTIVE QoS • User experience is the one that counts! • Subjective QoS is the service quality from the user perspective • measuring subjective QoS is done by user tests • only reliable way • Mean Opinion Score (MOS) tests are often used expensive and time consuming • Objective QoS • can be measured directly • can be used to estimate subjective QoS
7.7.2006
37
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
MONITORING TOOLS •Available to all •Off-the-shelf network analyzers (Ethereal, Tcpdump, WinDump, … ) •Custom software based on standard packet capture libraries (libpcap, WinPcap) •Operator and enterprise level monitoring tools •OSS •RTCP, RMON2, RTFM, … •MRTG •Typically measure round trip, not end-to-end one-way parameters •Network asymmetries dictate a closer look at one-way end-to-end measurements
7.7.2006
38
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
Packet capture
Packet capture
QoSMET –End-to-end QoS Monitoring Tool
7.7.2006
39
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
VERIFYING QoSMeT •VoIP call between the laptops •Network emulator for adjusting packet loss value •Measurements with QoSMeT •packet loss •offered load and •throughput
7.7.2006
40
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
VERIFYING QoSMeT EXAMPLE –packet loss
Time (s)
Emulated packet loss (PlossE)
Measured avg. Ploss + 95 % CI
[0, 30)
0.00
0.000
[30, 90)
0.10
0.100 ± 0.010
[90, 120)
0.00
0.000
[120, 180)
0.20
0.201 ± 0.021
[180, 210)
0.00
0.000
[210, 270)
0.05
0.050 ± 0.008
[270, 300)
0.00
0.000
[300, 360)
0.15
0.152 ± 0.013
[360, 390)
0.00
0.000
[390, 450]
0.03
0.033 ± 0.007
7.7.2006
41
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
VERTICAL HANDOVER PERFORMANCE – measurement scenario
7.7.2006
42
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
VERTICAL HANDOVER PERFORMANCE – measurement scenario • Mobile IP in use
PUBLIC 3G NETWORK WLAN
WLAN
LAN
LAN
50s
7.7.2006
50s
50s
50s
50s
43
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
VERTICAL HANDOVER PERFORMANCE - delay 0.2 Delay
0.18 0.16 0.14
Delay [s]
0.12 0.1 0.08 0.06 0.04 0.02 0 0.0
50.0
100.0
150.0
200.0
250.0
Time [s] 7.7.2006
44
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
VERTICAL HANDOVER PERFORMANCE – duration of connection loss 10.000
Connection loss length [s]
Loss length
1.000 0.0
50.0
100.0
150.0
200.0
250.0
0.100
0.010 Time [s] 7.7.2006
45
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
TIME FOR QoS THROUGHOUT THE STACK •Intra- and, to some extent, inter-system handovers based on link layer metrics are commonplace in wireless networks •We need to go further: session continuity •VTT demonstrated session continuity for streaming media between different devices (PC and IPAQ running Linux)
AMBIENT NETWORKS DEMO
7.7.2006
46
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
QoS THROUGHOUT THE STACK (2) •Applications will need to incorporate some form of adaptation too (related work: MAGELLAN, PHOENIX) •Example: QoS-Aware Gaming-on-Demand
O
7.7.2006
pe
r
ck ba et n n r te tio In ec r’s onn o c at
n bo
e
47
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
QoS THROUGHOUT THE STACK (3) •Real-time video coding adaptation method for game service •Network monitoring tool •Real-time video encoding parameter optimization
7.7.2006
48
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
QoS THROUGHOUT THE STACK (4) •Moore's Law is favorable to more efficient, but computationally expensive codecs •Pattern of development cycles efficiency gains
Source: European Broadcasting Union
•at least two cycles to come after MPEG-4 Part 10 D. Wood, EBU 7.7.2006
49
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
QoS THROUGHOUT THE STACK (5) •Conjecture: QoS in heterogeneous environments cannot be delivered with network-based QoS alone •We can provide a certain level of QoS or adaptation at the two ends of the protocol stack •What about the rest of the stack? •Underlying mechanisms need further study •Transport protocols, such as TCP, might need some new options. Example: TCP User Timeout Option (draft-ietf-tcpm-tcp-uto-02, October 2005) •Handovers cannot be solely based on link layer metrics (e.g. SNR). Why?
7.7.2006
50
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
3G/UMTS DYNAMIC CAPACITY ALLOCATION
7.7.2006
51
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
250
300
350
3G/UMTS: FIRST CONNECTION GOODPUT
X
200 150
X
100
Goodput (kb/s)
X
X
50
X X X
X
8
16
0
X 4
32
64
128
256
512
1024
MOSET Payload (KB) 7.7.2006
52
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
LAN: FIRST CONNECTION GOODPUT
6000
X
X X
2000
4000
Goodput (kb/s)
8000
10000
X
X
X
32
64
X X
4
8
0
X
16
128
256
512
1024
MOSET Payload (KB) 7.7.2006
53
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
THE "PROPER" IP QoS •When unconditioned TCP-like traffic (i.e., traffic that slows down in the face of congestion) is mixed in with real time traffic (that keeps going despite congestion), both sides lose — Carpenter & Nichols (2002) •Need a QoS framework matching IP principles: •Network services (QoS) should not be designed for, or tied to any particular application •IP designers did not attempt to predict what applications will be using the network — neither should QoS designers •Provide the means to differentiate traffic and allow for network engineering 7.7.2006
54
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
DIFFERENTIATED SERVICES ARCHITECTURE •Scalable: •classification & conditioning only at boundaries •small set of forwarding behaviors •apply per-hop behaviors to aggregates of traffic
•Incrementally deployable •Differentiation is asymmetric, decoupled from apps •A refinement of the original Precedence model
7.7.2006
55
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
IPv4 CLASS-BASED DIFFERENTION •RFC 791 (1981) and RFC 1812 (1995) Precedence
Type of Service
•RFC 2474 (1998) and RFC 3260 (2002) Differentiated Services Field •RFC 3168 (2001) Differentiated Services Field
7.7.2006
ECN
56
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
SERVICE SPECIFICATION & PHBs •Service level specification (SLS): set of parameters and their values which together define the service offered to a traffic stream by a DS domain •Traffic conditioning specification (TCS): set of parameters and their values which together specify a set of classifier rules and a traffic profile •TCS: integral element of an SLS •Per-hop Behaviors (PHB): •Default; best effort •Class selector •Expedited forwarding (EF); "virtual leased line" •Assured forwarding (AF)
7.7.2006
57
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
CLASSIFICATION •Bonaventure defines a flow as a sequence of packets with one common "characteristic", which can be based on any field of the packets •Flows can be defined at different layers providing finer granularity and control at the cost of more state/lookups •Classify once at edge, mark and then use markings •Static vs. Dynamic Classification
7.7.2006
58
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
STATIC CLASSIFICATION •Layer 2: ATM and Frame Relay circuits, switched e2e circuits (GbE, soon XGbE and beyond); L2 VPNs •Layer 3: IP host-to-host, but also all IP traffic with the same next BGP hop; L3 VPNs •Layer 4: All TCP traffic from host-to-host is treated differently from all UDP traffic for the same pair. •Layer 7: HTTP vs. VOIP vs. FTP vs. SMTP
7.7.2006
59
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
TRAFFIC CLASSIFICATION & CONDITIONING
Measure the temporal properties of the packet stream
Meter
Packets
Classifier Multi-field classification
Marker Set DSCP
Differentiated Services Field
7.7.2006
Shaper/Dropper
ECN
Delay/discard some or all of the packets in a traffic stream in order to bring the stream into compliance with a traffic profile
60
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
DiffServ ARCHITECTURE •Minimalist — sophisticated simplicity •Separation of control and forwarding (like in IP) •Supported by all major vendors in mid- and high-end routers •Inter-domain, bilateral agreements •For inter-AD traffic, perhaps the only pragmatic, standardized framework in actual deployment •Nevertheless, deployment is not widespread •Non-technical obstacles
7.7.2006
61
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
DiffServ: STILL RELEVANT? •By the time RFCs 2474 & 2475 were released in December 1998 •Asia: the financial crisis was in full swing •USA: the major issue was the Monica Lewinsky scandal •Europe: the euro did not exist •Wall Street: irrational exuberance ruled •In mid-June, crude oil set a 12-year low: it averaged $10.11 per barrel— half of the official OPEC target of $21 •1998 birthdays: Windows 98, iMac, Celeron, and Google
7.7.2006
62
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
SLOW DEPLOYMENT •The Maxinetti case shows that class-based differentiation is deployable, allows for new services, and can be profitable •That is exactly what DiffServ was all about •So why is public deployment of DiffServ soooo slow? •Need inter-provider agreements (cf. VPN) •Need to demonstrate the benefits(?) of QoS •Need to enforce consistent policies •Overprovisioned backbones •QoS is costly and can lead to operational overhead for providers •No common, well-understood service definitions •Your reason here :) 7.7.2006
63
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
OPEN ISSUE: WHO NEEDS QoS? •L3 virtual private networks (VPN)? •Most of the DiffServ deployments •Network games? Henderson & Bhatti (2003): •Many and successful net games… using best effort only •Throughput not an issue, delay is •Reported delays deter users from joining a server •Delay increases while playing do not force users to leave in droves despite the noticeable degradation in their gaming performance •Would gamers pay for QoS? •Yes, if included in the price of the game •No, if it was offered as a "premium" service 7.7.2006
64
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
OPEN ISSUE: WHO NEEDS QoS? (2) •VoIP •Skype is already making VoIP reality without any QoS and you only need a dialup connection •Why would a user pay more for her VoIP packets? She wouldn't. But she would go for a Maxinetti kind of service which is cheap and hip :) •And that is our view: QoS frameworks should be seen as enablers, not as cash cows •IPTV •Video gaming servers
7.7.2006
65
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
MULTIMEDIA APPLICATIONS AND QOS •Application QoS metrics (Bhargava, 2002) •Timeliness (meet play-out deadlines) •Accuracy (play the right data) •Precision (receive what was sent) •For instance, in a multimedia presentation •play all frames in order, without delays and discards •on time and in sync with the rest of the content • do so at the bit rate of the encoded stream, without any discounts on quality. •The network can provide guaranties in drop rate, but an SVC receiver may discard half the received frames •Can meet throughput requirements (averaged over a certain period) but break timeliness for real-time video 7.7.2006
66
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
MULTIMEDIA ADAPTATION LAYER: MOTIVATION •FP6 IST PHOENIX (2004-2006) •Tag line: jointly optimizing multimedia transmissions in IP-based wireless networks •develops solutions that exploit available bandwidth on wireless links efficiently • source coding • MAC, channel coding •targets multimedia transmission over wireless IP networks
7.7.2006
67
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
MULTIMEDIA ADAPTATION LAYER: MOTIVATION (2) •Layered video has a hierarchical structure: •not a "flat" byte stream •neither a series of independent datagrams •but a stratified, interdependent "set of streams" associated with a different end-user "value" and play out deadlines •Are typical transports and rudimentary traffic treatment sufficient? •Need •a solution for different access networks •to evaluate the real benefits from H.264/SVC while it's being standardized
7.7.2006
68
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
A SCALABLE EXTENSION TO H.264/AVC •SVC is a scalable extension to H.264/AVC •jointly developed by MPEG (ISO) and VCEG (ITU) expert groups •aims at offering scalability with comparable coding efficiency versus current state-of-the-art non-scalable coding schemes (H.264) •standardization process is still ongoing •scalability is three-dimensional • Temporal (frames/s) • Spatial (image resolution) • SNR (signal-to-noise ratio)
7.7.2006
69
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
H.264/SVC MAIN GOAL •Video is encoded only once •The encoded version can be scaled easily to several user equipment (from HDTV to mobile phone)
7.7.2006
70
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
H.264/SVC CODING EFFICIENCY
7.7.2006
71
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
RELATED WORK •Lots of work on MPEG-4 FGS (with and without crosslayer optimizations) •FGS coding efficiency is not so good •Lots of work on MAC-based (802.11e, for example) cross-layer design and optimizations •we are interested in a solution that can work based on established standards, independent of MAC if possible •No prior network simulation studies with SVC
7.7.2006
72
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
MULTIMEDIA ADAPTATION LAYER: ARCHITECTURE •Source rate adaptation •Traffic differentiation •Packet prioritization •Rate adaptation
7.7.2006
73
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
MULTIMEDIA ADAPTATION LAYER: ARCHITECTURE (2)
7.7.2006
74
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
SIMULATION METHODOLOGY 1. Virtual video streams
Video
H.264/SVC Encoder/ Packetizer
<100, 1500, 0> <139, 1500, 1> <600, 1500, 2> <60, 300, 2>
D1 Trace
Network
Video receiver Video Server Direction of video packet stream
2. ns-2 simulation
ns-2 trace
Packet arrival time, size, layer, … ? Video quality metrics
3. Trace post-processing 7.7.2006
75
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
RESULTS (1/3): THROUGPUT 600000
500000
Throughput (bytes/s)
400000
300000
200000
100000
0 10
11
12
13
14
15
16
17
18
Time (s) TCP 7.7.2006
H.264/SVC
SVC/PriQ
SVC/MAL 76
19
20
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
RESULTS (2/3): PACKET ARRIVALS 4/5
Packets Received: H.264/SVC/MAL
5.3 3 2 1 0 0 1 2 3 4/5
4.8
Packets Dropped
4/5
4.33
SVC Layer
2 1 0
H.264/SVC/PriQ 0 1 2 3 4/5
3.8
4/5
3.3 3
H.264/SVC
2 1 0 0 1 2 3 4/5
2.8
2.3 Video-on-demand over TCP New Reno (no layers, no priorities) 1.8 10
11
12
13
14
15
16
17
18
Packet Arrival Time (s) 7.7.2006
77
19
20
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
RESULTS (3/3) : PACKET ARRIVALS 4/5
Packets Received: H.264/SVC/MAL
5.3 3 2 1 0
4.8 Packets Dropped
0 1 2 3 4/5
4/5
4.33
SVC Layer
2 1 0
H.264/SVC/PriQ 0 1 2 3 4/5
3.8
4/5
3.3 3
H.264/SVC
2 1 0
0 1 2 3 4/5
2.8
2.3
1.8 16
16.1
16.2
16.3
16.4
16.5
16.6
16.7
16.8
Packet Arrival Time (s) 7.7.2006
78
16.9
17
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
MAL: SUMMARY •The Multimedia Adaptation Layer (MAL) •builds on recent advances in scalable, layered video standards •employs standards-based network and MAClayer traffic prioritization mechanisms •is necessary for scalable video over wireless networks •Our evaluation methodology •capitalizes on • cutting-edge, prototypical H.264/SVC video encoding software • the most widely-used network simulator •provides important insights for future development 7.7.2006
79
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
VOIP WITHOUT QOS: SKYPE •Skype bundles •VoIP (the first p2p-based client) • Free PC-PC, with the best quality compared to Yahoo!, AIM, Google Talk • Very cheap (0.02 EUR/min) to/from most of the world supporting both PC-PSTN (SkypeOut) and PSTN-PC (SkypeIn) • Teleconferencing (up to 5 people) • Recently, video calling too •Instant messaging (IM) •File transfer (p2p-based, of course)
7.7.2006
80
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
WHY SKYPE THRIVES •Skype works •Seamlessly behind NATs and firewalls • Implements TURN and STUN (or some variant) at the client • In contrast SIP-based VoIP requires explicit server configuration in applications •Availability (WinOS, Linux, MacOS, PocketPC) •Easy installation, same interface and functionality •Does not require lots of resources (not even bandwidth) •Gives a certain feeling of privacy to users by encrypting all of its traffic -- other IMs do not
7.7.2006
81
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
SKYPE: NETWORK ARCHITECTURE •Skype is KaZaa-based •Supernode-based hierarchical p2p network •Can detect NATs and firewalls •Skype Uses •TCP for signaling •UDP (preferably) and TCP for VoIP traffic (if firewall/NAT-restricted) •No fixed-ports •Encryption on all but a few initialization packets • 256-bit AES for calls and IMs • 1024-bit RSA to negotiate symmetric AES keys
7.7.2006
82
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
SKYPE: NETWORK ARCHITECTURE (2) •Supernodes are elected based on •Network availability (open Internet access) •Bandwidth availability •CPU, memory, play a smaller (if any) role •Users cannot prevent their node from becoming a supernode (unlike other p2p) •May be able to influence the process, though
7.7.2006
83
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
SKYPE: NETWORK CHARACTERIZATION •Supernodes •Consume less than 205 b/s (!) 50% of the time • Negligible CPU, memory consumption •Relay data only 9.6% of the time • Data sessions are less frequent than VoIP ones • File sizes tend to be considerably smaller than in other p2p networks (photos, docs, slide sets-- not mp3's and videos) •Relay NAT-restricted calls: • Median/mean call duration--2m50s/12m53s • (PSTN calls average 3m) • Max call duration--3h26m 7.7.2006
84
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
QoS WITH FLAT PRICING??? •QoS is about allowing the user to select between quantitative performance guarantees — Crowcroft et al. (2003) •Personal opinion •QoS as a service enabler which brings new products in the market •Unchain QoS from "cost linked to quality" •Marketing should be about a service not the technology •Those familiar with "all-you-can-eat" buffets most certainly appreciate the simplicity in pricing •Yet, when one starts talking to me about QoS I check that my wallet is in place…
7.7.2006
85
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
QoS WITH FLAT PRICING!!! •Free nights and weekends has been quite a common offering from US cellular operators for years now •Vonage, Cablevision offer unlimited US & Canada calls •Do these schemes hurt revenues? Decrease profits? •How much can one "eat" anyway?
7.7.2006
86
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
OPEN ISSUE: OPERATIONAL COMPLEXITY •Based on his operational experience Bell (2003) argues that •Network Operation Center personnel have come to believe that complex protocols destabilize a network, mainly due to buggy implementations •Case in point: introducing multicast in the LBNL network led to difficult to trace bugs •Amplification and Coupling principles •IP multicast as a limit-case: Any QoS framework should be less complex than multicast in order to gain wide adoption •As such, IntServ is pretty much done
7.7.2006
87
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
OPEN ISSUE: OPERATIONAL COMPLEXITY (2) •Overprovisioning to the rescue: simple and economical •The "10% rule" •Deal with network congestion Throw bandwidth at the problem or Throw protocols at the problem •There are cases, though, that bandwidth simply cannot be thrown at the problem (regulatory and CAPEX issues, spectrum licenses,… )
7.7.2006
88
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
OPEN ISSUE: TRAFFIC CLASSIFICATION •Traffic classification •End hosts are the natural points, but due to lack of trust and maintaining administrative control, gateways are preferred by NOCs •Dynamic classification of packets into different classes is not a trivial task •Inhibits QoS deployment •M. Roughan, et al. (2004): •Framework for scalable, dynamic traffic classification based on statistical application signature •Obtain signatures insensitive to the particular application protocol
7.7.2006
89
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
END to END QoS •User experienced Quality of Service is the one that counts! •Certain QoS mechanisms in some network technologies (DiffServ, MPLS, SLA, 802.11e, … .) •no method for end-to-end QoS from service to user over various network technologies •End-to-end QoS can be supported in a multivendor environment with standards-based adherence.
7.7.2006
90
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
END to END QoS •Total QoS is composed of several mechanisms on different protocol layers (MAC scheduling, retransmissions, packet sizes, routing decisions, priorization, flow control, congestion control, … ) •mapping of QoS parameters between protocol layers and optimisation within single technology need to be done but is not enough •User experience depends on the performance of the whole chain of technologies between him and the service -> interoperable QoS mechanisms are needed •Number of users and QoS needs of their applications need to be fitted together with the restricted resources available
7.7.2006
91
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
END-TO-END QoS IN HETEROGENEOUS NETWORKS • Network heterogeneity =>Quality of Service has to be deployed end-to-end • QoS schemes in IP Networks • Best Effort • Integrated Services (IntServ) • Differentiated Services (DiffServ) • WLAN QoS • IEEE 802.11e being finalized • Service Level Agreements (SLA) • adjusting QoS classes of different networks • No End-to-End method standardised yet • Application used by the User Equipment should be able to specify its QoS needs
7.7.2006
WLAN 2G 3G PAN
92
LAN
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
EUREKA/ITEA EASY WIRELESS PROJECT Factory WLAN Network
Office WLAN Network
IP NETWORK Wide Services & Interactions
AdHoc Mobile Net Community PAN Network
Local Services & Interactions WLAN H/2 WLAN 802.11 GPRS/UMTS 7.7.2006
93
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
Easy Wireless Allow seamless roaming between wireless networks while maintaining Quality of Service • EUREKA/ITEA project • ITEA is a project clustering organisation • funding from each country • 16 partners from 5 countries • Sept. 2004-Sept. 2007 • Total budget: 12 Million € • Partners • Thales Communications • Telefónica • 4 Universities • 5 SME’ s • 4 Research Centres
7.7.2006
Belgium Finland Netherlands Norway Spain
94
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
SYNOPSIS •QoS is a well-researched issue •Mature frameworks developed for LANs, WANs, and inter-AD •No e2e QoS framework • Mappings are not standardized • Deployment is still slow •QoS used as an enabler for new services, not as a cash cow. •QoS-awareness needs to be diffused throughout the stack •Overprovisioning not a bad thing, not antithetic to QoS
7.7.2006
95
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
ACKNOWLEDGMENTS •Sari Järvinen, Jukka Mäkelä, Jyrki Huusko (VTT) •Stephen Sykes (Maxisat)
7.7.2006
96
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
FURTHER READING • G. Armitage, Quality of service in IP networks: Foundations for a multi-service Internet, Indianapolis, IN: Macmillan Technical Publishing, 2000. • G. Bell, "Failure to thrive: QoS and the culture of operational networking", Proc. ACM SIGCOMM 2003 Workshops, Karlsruhe, Germany, August 2003, pp. 115-119. • B. K. Bhargava, "Guest Editorial: Quality of Service in Multimedia Networks", Multimedia Tools and Applications, 17(2-3), 151-156. • S. Blake, D. Black, M. Carlson, et al., An Architecture for Differentiated Service, Internet RFC 2475, December 1998. • B. Carpenter, & K. Nichols, "Differentiated Services in the Internet", IEEE Proceedings, vol. 90, no. 9, 2002, pp. 1479-1494. • K.G. Coffman & A.M. Odlyzko. "Internet growth: Is there a "Moore's Law" for data traffic?," In: J. Abello, et al. (eds.), Handbook of Massive Data Sets, Boston, MA: Kluwer, 2001. • J. Crowcroft, S. Hand, R. Mortieret, et al., "QoS's downfall: at the bottom, or not at all!", Proc. ACM SIGCOMM 2003 Workshops, Karlsruhe, Germany, August 2003, pp. 109-114.
7.7.2006
97
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
FURTHER READING (2) • S. A. Baset and H. Schulzrinne, "An Analysis of the Skype Peer-toPeer Internet Telephony Protocol" Proc. INFOCOM 2006, Barcelona, Spain, April 2006. • B. Davie, A. Charny, J.C.R. Bennett, et al., An Expedited Forwarding PHB (Per-Hop Behavior), Internet RFC 3246, March 2002. • B. Davie, "Deployment Experience with Differentiated Services," Proc. ACM SIGCOMM 2003 Workshops, Karlsruhe, Germany, August 2003, pp. 131-136. • D. Grossman, New Terminology and Clarifications for Diffserv, Internet RFC 3260, April 2002. • S. Guha, N. Daswani, and R. Jain. "An Experimental Study of the Skype Peer-to-Peer VoIP System," Proc. 5th International Workshop on Peer-to-Peer Systems (IPTPS '06), Santa Barbara, CA, February 2006. • J. Heinanen, F. Baker, W. Weiss, J. Wroclawski, Assured Forwarding PHB Group, Internet RFC 2597, June 1999.
7.7.2006
98
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
FURTHER READING (3) • W. Hardy, QoS measurement and evaluation of telecommunications quality of service, West Sussex, England: John Wiley & Sons, 2001 • T. Henderson & S. Bhatti, "Networked games: a QoS-sensitive application for QoS-insensitive users?", Proc. ACM SIGCOMM 2003 Workshops, Karlsruhe, Germany, August 2003, pp. 141-147. • G. Huston, "Quality of Service--Fact or Fiction?", The Internet Protocol Journal, 3(1), 27-34. • R. Lloyd-Evans, QoS in Integrated 3G Networks, Norwood, MA: Artech House, 2002. • K. Nichols, et al., Definition of the Differentiated Services Field (DS Field) in the IPv4 and IPv6 Headers, Internet RFC 2474, December 1998. • K. Pentikousis, et al., “Active goodput measurements from a public 3G/UMTS network”, IEEE Communications Letters, 9(9), 802-804. • H. Petroski, To engineer is human— the role of failure in successful design, New York: Vintage Books,1992.
7.7.2006
99
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
FURTHER READING (4) • M. Roughan, et al., "Class-of-service mapping for QoS: a statistical signature-based approach to IP traffic classification", Proc. ACM SIGCOMM IMC 2004, Taormina, Italy, October 2004 pp. 135-148. • K. Singh and H. Schulzrinne, "Peer-to-peer internet telephony using SIP". Proc of International Workshop on Network and Operating Systems Support For Digital Audio and Video (NOSSDAV '05), Stevenson, Washington, USA, June 2005, pp. 63-68.
• T. Ulseth, " A path towards common quality assessment of narrowband and wideband voice", presented at Workshop on Wideband Speech Quality in Terminals and Networks: Assessment and Prediction, Mainz, Germany, June 2004. • Z. Wang, Internet QoS architectures and mechanisms for quality of service, San Francisco, CA: Morgan Kaufmann, 2001. • D. Wood, "Everything you wanted to know about video codecs— but were too afraid to ask", EBU TECHNICAL REVIEW, July 2003.
7.7.2006
100
VTT TECHNICAL RESEARCH CENTRE OF FINLAND
RELATED WEB SITES •Easy Wireless http://ew.thales.no •Ambient Networks www.ambient-networks.org •MAGELLAN— Multimedia Application Gateway for Enterprise Level LANs www.magellan-itea.org •PHOENIX— Jointly optimizing multimedia transmissions in IP based wireless networks www.ist-phoenix.org 7.7.2006
101