Quality of Service for new applications of the Internet

1
Quality of Servicefor new applications of the
Internet

• David Billard
• David.Billard_at_cui.unige.ch

2
Outline

• New applications in (and of) the Internet
• Quality of service
• Accounting and charging
• Payment
• Routing with Quality of service
• Voice over IP
• Conclusion

3
New applications in (and of) the Internet

• Telephony over IP
• Voice and Fax communications supported by IP
  networks
• Video on demand
• Continuous flows of Video/Audio sequences
• Virtual Private Networks (extranets)
• Private networks built upon a public
  infrastructure (such as the Internet)

4
New applications in (and of) the Internet
Charging
Accounting
5
Essential need assured quality of service

• Quality of service?
• Assured bandwidth from end-to-end (e.g. 8Kbps for
  voice),
• communication privacy,
• minimal round time for packet delivery,
• regularity of the data flow,
• etc.

6
Quality of Service in the Internet

• Current strategy, implemented into each router
  the  best effort , or minimal service
• every packets are processed in the same way,
• therefore it is not possible to assure a QoS for
  each paquet of the same application
• it is a brake for the deployment of such
  applications

7
Quality of Service in the Internet

• However
• the TOS (Type Of Service) field is present in IP
  since a long time.
• In sept. 81, in le RFC 791, J. Postel wrote the
  TOS provides an indication of the abstract
  parameters of the QoS desired.
• The TOS field has never really been exploited.

8
Quality of Service in the Internet

• What can we do?
• 2 models can be drawn
• Intserv (Integrated services) data flows among
  applications, with flow recording at each router
• DiffServ (Differenciated services) flow
  aggregate packet tagging

9
Integrated Services (IntServ)
ISP 1
ISP 2
10
Integrated Services (IntServ)

• 2 fundamentals
• Resource reservation
• Admission control

Reservation initial agent
Management agent
Routing agent
Router (flow control)
Admission control
DB for traffic management
Routing DB
11
IntServ - RSVP

• RSVP protocol, receiver oriented (it was designed
  with multicast in mind)
• The sender gives its flow specifications in a
  RSVP message
• Each intermediate ISP can modify these
  specifications in regards of his possibilities
• The receiver sends OK or NOK via the same path

12
IntServ - Admission control
Packets coming
Classification (agregation)
Scheduling
Router - Packets processing

• Classification fields port, _at_ source, etc.
• Scheduling simple priority, round robin, WFQ
  (Weighted Fair Queuing), ...

13
IntServ - Drawbacks

• Does not scale
• A router has to keep track of all the flows using
  it at time t
• Applications must be RSVP-aware
• The applications give their flow specifications

14
Differentiated Services (DiffServ)
ISP 1
ISP 2
15
Differentiated Services (DiffServ)

• Each packet is  tagged  (fields DS or TOS)
• The first router (ingress), looking at the tag,
  applies a Per-Hop-Behavior (PHB)
• chooses a route for the packet
• gives a new tag for the exit router (egress)
• ISPs must have peering agreements

16
Differentiated Services (DiffServ)
Measures (traffic profile)
Packets comming
Classification (DS codepoint)
Packet tagging
Profiling / Reject

• DiffServ is more scaleable
• Works even with classical applications

17
Mixing IntServ / DiffServ

• Idea
• IntServ in the stub networks (not backbones)
• DiffServ in the backbones
• The peripheral routers must understand the 2
  protocols and apply a mapping between IntServ and
  DiffServ QoS

18
Mixing IntServ / DiffServ

• IntServ proposes 2 QdS for the flows
• Controlled load (probalistic) and Guaranteed
  Service (time critical)
• DiffServ proposes 2 QdS for its packets
• Assured service (probalistic) et Premium service
  (time critical)
• It is necessary to do a mapping

19
Managing the Quality of Service

• From an ISP, proposing a QoS is not enough
• For example a user reserves all the bandwidth
  available
• The ISP must charge the user for its QoS and must
  adapt very rapidly to the demand

20
Accounting and Charging

• Accounting units of invoicing
• used time,
• bandwidth,
• real traffic,
• etc.
• Charging these units to a user

21
Accounting and Charging

• In the network data information on used
  resources
• secured and certified information
• all the ISP concerned for the same service have
  to cooperate
• establishment of Service Level Agreements among
  ISPs

22
Global architecture
ISP 1
Bandwidth Broker 1
Bandwidth Broker 2
23
Global architecture
Best effort IP
Congestion
24
Global architecture
SLA negociation
25
Global architecture
26
IntServDiffServ architecture

• Advantages for the ISP
• Dynamic management of its connections
• Fine granularity of its connections
• Drawbacks
• scheduling of aggregates and moreover
  provisioning, even overprovisioning (all the ISPs
  concerned in a new connection have to deliver the
  QoS)

27
And the payment?

• Several schemes are possible
• classical (bank account transfer, cheques, )
• long delay, heavy to use
• credit card, kleboxes, etc.
• reduced delay, still heavy to use
• real-time payments, thanks to micro-cheques
• very short delay, easy to use and flexible

28
Payment
ISP 1
ISP 1 manages everything One must trust ISP1
29
Payment
ISP 1
One must trust Isp1..ISPn-1 Advantage only
bilateral contracts
30
Payment
ISP 1
The client must know all the ISPs in the path
31
Integrated payment (micro-cheques)
ISP 1
Addressed and signed cheques Periodically
sent (piggybacked with the signaling, PATH, RSVP)
32
Integrated payment

• Possibility for cost sharing
• Example with IP-Telephony the caller pays 60
  and the callee 40
• Possibility for using a javacard and to call from
  a public phone
• No way to get reimbursed in case of failure

33
RSVP-like messages for selecting QoS and payment

• Extending RSVP by creating protocol objects
  dedicated for the payment

34
How to determine the price of a connection
(pricing)?

• The price of a service can be a fixed data (e.g.
  monthly fee)
• The price can be a function of the usage
  (duration, data volume, bandwidth,...) which is a
  dynamic data
• The associated price dramatically changes in
  function of the network usage (peak hours,)

35
How to determine the price of a connection
(pricing)?

• The marginal cost to transport a packet 0, when
  there is NO congestion
• The pricing policy can help to control the
  congestion (rare resources allocation)
• Two policies  smart market  and  profile 

36
Smart market (auctions)

• Each packet has a bid field
• Each bid is compared to the marginal cost of
  transportation (MC)
• If bid lt MC, packet rejected
• If bid gt MC, packet accepted, the price to pay is
  MC, independently of the bid

37
Smart market - problems

• If all the packets propose a high bid, when the
  marginal cost is low
• No packet is rejected
• The congestion control is inefficient
• The management overcost is important
• The sudden variations (bursts) in the bandwidth
  are not well managed

38
Smart market

• Adjunction of an additional level
• auction on the price per packet
• auction on the price per quantity (or by flow)
• Progressive Second Price auction
• The Smart Market model evaluates the price to pay
  in function of the networks load

39
Profils and classes

• Goal predicting the expected QoS in a
  best-effort Internet
• The user precises its parameters in a  service
  profile 
• The network tag the packes in or out profile
• In case of congestion, the out profile packets
  are rejected in first

40
Profils and classes

• Services with a higher class are priced higher
• The Profil and Classes evaluates the price to pay
  in function of the users profile

41
Routing and Quality of Service

• How to route packets?
• In function of the destination, yes.
• In funcion of the nodes load, yes.
• Also in function of the requested QoS by the user
  and the delivered QoS by intermediate ISPs.

42
Routing and Quality of Service
X
43
Routing and Quality of Service

• The normal routing of packets is based on the
  hierarchical (and quite static) structure of the
  Internet
• The QoS-based routing is directly dependant of
  the resources usage
• e.g. the available bandwidth is a dynamic data.

44
Routing and Quality of Service
O
X
45
Routing and Quality of Service

• The QoS-based routing is not trivial
• Recursive routing algorithms, in the Internet,
  are not supposed to succeed
• In particularly for real-time applications
  (telephony)

46
Quality of service and Transactions
ISP4
ISP2
ISP7
ISP5
X
ISP1
ISP3
ISP6
O
?
?
?
?
2Mbps available
1Mbps available
47
Peering agreements between ISPs

• Commercial agreements among ISPs are peering
  agreements
• two ISPs agree on an amount of traffic exchange
• the agreement is always unbalanced
• they need a compensation
• the agreement is static, renegotiated
  periodically

48
Service Level Agreement (SLA)

• ISPs are not used to real-time negotiation
• e.g. to establish an ATM connection with Swisscom
  (the Swiss telephony public company), one must
  senda fax!
• SLAs may become a solution once a standard will
  be adopted by everyone
• How to know is the neighbor-ISP allows SLAs?

49
Definition and publication of services

• Problem how an ISP can know that his
  neighbor-ISP propose a service x?
• When an ISP creates a service (e.g. support of
  IPSec in tunneling mode), he publishes it
• The problem is to propagate the information to
  the other ISPs

50
IP-Telephony - Current architecture
51
IP-Telephony - Intermediate architecture
IP network (for data, voice and fax)
52
IP-Telephony - Final architecture
IP network (for data, voice and fax)
53
Why a single network for voice, fax and data?

• Telephony communication costs should be
  dramatically reduced
• A single physical network is easier to maintain
  than 2 (moderation additional equipment must be
  deployed)
• Cabling of new buildings (or extensions) should
  be easier and faster
• The future or PABX is uncertain

54
Why a single network for voice, fax and data?

• (PABX Private Automatic Branch Exchange)
• Drawbacks
• PABX use proprietary software
• the number of connected devices to a PABX is
  physically limited
• the cost of a PABX is not peanuts
• old PABX must be replaced

55
Why a single network for voice, fax and data?

• Emerging applications using the integration of
  services (voice, video, data)
• electronic commerce
• on-site virtual technician
• help desk
• conferencing
• ...

56
Components of an IP-Telephony solution

• Terminals provide real-time voice communications
  (IP-phones, netmeeting)
• Gateways provide PSTN/IP connectivity
• Gatekeepers provide service facilities
• MCU (Multipoint Control Units) provide support
  for teleconferences between 3 to n endpoints.

57
Problem - 1 - PSTN/IP interface

• PSTN is a circuit-switched network
• IP is a packet-switched network
• two different technologies that cannot
  interoperate easily
• An intermediate equipment, the gateway, must be
  deployed

Voice call
H.323 call
58
Component - Gateway

• Objective brings connectivity between IP and
  PSTN networks.
• Functions
• translating protocols and audio codecs
• converting information formats
• transferring information
• gateways are often H.323 compliant
• Usually works closely with gatekeepers.

59
Problem - 2 - Telephony service consistency

• Calling a person

022 705 7644 129.194.20.10billard
022 705 7644
022 705 7644
022 705 7644
129.194.20.10billard
60
Component - Gatekeeper

• Objective management of the (H.323) calls.
• Functions
• address translation (e.g. phone to IP address)
• admission control through RAS (Registration,
  Admission, Status) protocol
• bandwidth management
• Zone management
• Call authorization and management

61
Problem - 3 - Directories

• User information stored using one (several)
  directory managers
• NDS (Novell Directory Server)
• LDAP (Lightweight Directory Access Protocol)
  compatible
• X500 ...
• Which directory system will use the gatekeepers?
• How to cooperate with legacy systems?

62
Problem - 4 - Reliability

• Users are used to an always functioning telephone
  set
• I pick up the phone, it works.
• The reliability of the PSTN is 99.9999 (lt 1
  minute of unavailability per year)
• Reliability of nowaday computer systems is around
  99.9 (8h45mn per year) - 99.995 (27 mn per year)

63
Problem - 4 - Reliability

• The reliability is directly connected to the
  Quality of Service in the network
• e.g. a phone call, with Best Effort on a
  overloaded network the call does not succeed
• unacceptable for a user (emergency numbers, etc.)

64
Problem - 5 - Quality of Service

• Without mechanisms as DiffServ or IntServ in the
  world Internet
• possible to specify the QoS in a LAN or a MAN
• only if the hardware AND software are homogeneous
  
• under the condition to use DiffServ, IntServ, or
  to use lower layers (Sonet, ATM) to manage the
  traffic in function of the clients

65
Problem - 5 - Quality of Service

• jitter (voice and fax are very sensible to delay
  variation)
• Packets are sent at intervals i
• Packets are received at intervals i?, i?, etc.

Sender
Receiver
i?
i?
i
i
If ? or ? are too high (e.g. gt 40ms), voice will
be of bad quality.
66
Conclusion

• The development of mechanisms to assure the
  Quality of Service is imperative
• increasing only the bandwidth (e.g. TEN-34 to
  TEN-155) is not enough
• The new integrated applications in the Internet
  (voice, data, video, etc.) must integrate notions
  of QoS and payment

67
Conclusion

• ISPs will have to
• collaborate to establish standard SLAs
• put in place mechanisms to publish their services
  
• Mechanisms are still needed for
• QoS-based routing
• usage of atomic transactions
• Telephony on the Internet (for the average user)
  is not for tomorrow

68
Disgression IP-Telephony / Voice over IP /
Internet Telephony

• IP-Telephony applying the standard telephony
  business model to the Internet (paying for
  (duration, distance))
• Internet Telephony telephony is an additional
  application on top of the Internet (paying for
  QoS, bandwidth, etc.)
• Voice over IP merge(IP-Telephony, Internet
  Telephony)