AgentEnhanced EndtoEnd Service Provisioning

1
Agent-Enhanced End-to-End Service Provisioning
Queens University David Chieng Professor Alan
Marshall
University of Ulster Ivan Wing Kong Ho Professor
Gerard Parr
2
Contents

• Overview of End-to-End Services with existing QoS
  architectures
• Agent-Enhanced Service Brokering Architecture
• SLA and Negotiation Process
• Service Broker with OAM Mediator Agent on Phoenix
• Prototype, Simulation Environment and Results
• Conclusions

3
End-to-End Services
Guaranteed End-to-End Services such as Virtual
Leased Line (VLL), Bandwidth Pipe, Minimum Rate
Guaranteed Service, etc can be realized using

• RSVP
• End-to-End BW Reservation (token rate, peak rate,
  etc.)
• Guaranteed or Controlled Load
• Introduced in Window 2000
• DiffServ
• VLL using EF with hard QoS or AF with soft QoS
• MPLS
• VLL using Label Switched Path (LSP) with QoS
• CISCO MPLS DS-TE QoS Router

4
The SLA Tree
Generic SLA
Service Ingredients?
Business SLA
Network SLA/SLS
Application SLA
5
Agent-Enhanced Service Brokering Architecture
Service Broker Agent


CP Agent
Service Layer


User Agent

HTTP(Phoenix)





Network Layer



Resource/Network Manager
Network Agent

Element Layer
Map
Local Agent
RSVP/DiffServ/MPLS
Configurations
Manager






Local Agents


Commands



JVM


End User/ Content Provider
RSVP/DiffServ/MPLS Enabled Router
Legacy Router
End User


Active Router


6
General Interaction Event Sequences(Inter
domains)
VoD Service Content/Movie Title xxxx Guaranteed
BW Premium Quality (10Mbps) Start Time
xxxxxx (hours, mins, secs) Duration xxxx
(mins)
If ASP 1 cant admit anymore traffic, a
Reject will be sent straight away
ASP1 Agent
End User
Network/Connectivity Provider Agent
Video Server A Agent
7
Interactions
8
SLA Form (ACL Format)
(cfp sender (agent-identifier
name UserAgent.phx
address (http//qub8888/localagent))
receiver (set (agent-identifier
name ASPAgent.phx address
(http//fujitsu8888/servicebroker))
ontology FIPA-VPN-Provisioning protocol,
language .. content
((establish (service-description
service-id Service1
service-type VoD
respond by ) (video-description
video-id Video1
format MPEG2
encryption )))
9
Service Broker using OAM Mediator on Phoenix

• Each Phoenix Engine has a mediator servlet .
• Each mediator has information of the agent
  servlets on the same Phoenix Engine.
• Mediators know each other and exchange
  information.

Information of local servlets
Mediator (realized as a Servlet)
Servlets
Phoenix
Domain B
Phoenix
Phoenix
Domain C
HTTP
Domain A
10
Servlet Plug and Play
3. UpdatingServlet information
1. Adding, removing, updatingServlet dynamically
2. AdvertisingServlet information
Servlet
Phoenix
Phoenix
HTTP

• Enabling dynamic extensions of service functions
• Can be unplugged for maintenance

Dynamic plug-in/out of Servlet
Keeping Servlet informationfresh by advertisement

• Selecting the best Servlet at requested time
• Selecting alternative Servlets for a failed
  Servlet (avoiding single-point-failure)

11
Message driven flows
T2
T3
T4
T5
mediator
T1
T6
INPUT type T1 OUTPUT type T6

• Types of messages determine a flow.
• The mediator chooses and pipes servlets according
  to types of input and output messages.
• Flow controls without controllers are possible.

12
Field Reactor Model
T1
T2
T3
T4
T5
T6
Field
mediator

• The mediator sets up the Field and collects
  necessary servlets.
• Each servlet reads a message from the Field if
  its type is matched and writes the result back to
  the Field.
• Flows can be progressed in parallel and can find
  the best flow.
• Flows can be easily modified and extended.

13
Brokering by Mediator
Information of Servlets
Information of Agents
CPA
ASPA
NA
SA
UA
Phoenix
Phoenix
Phoenix
HTTP
2. Mediator selects Agents involved
ASP Agent hides complicity from end users

• Reducing efforts of end users(parallel access,
  analyzing and summarizing results, etc.)

14
Simulation Environment
15
Mediator Advertisement Table
/servicebroker/VodAgentServlet.phx
logical.(classvod) /servicebroker/VodAgentServlet
2.phx logical.(classvod) /servicebroker/MusicAge
ntServlet.phx logical.(classmusic) /servicebroke
r/DbaseAgentServlet.phx logical.(classdbase)
/servicebroker/GameAgentServlet.phx
logical.(classgame)
Agent/Server Name
Service Name
Rules
destshttp//hostport/service/PeerMediator
16
Merits of the enhancement with Mediation

• Flexible, scalable, robust, and dynamically
  extensible service brokerage,
• Just-in-time finding of the best services
• Dynamic load balancing
• Avoiding a single point of failure
• Dynamic updates of services
• Preferences for services such as the
  geographically closest servers and availability
  of services at the requested time determines the
  best arrangement of flows.
• Loading, unloading, and reloading servlets
  changes the arrangement of flows without
  affecting flow controllers.
• Mediation performs the best anycast in a flexible
  and robust way.

17
Service Brokering Process
Two Phases

• Content Brokering
• Get desired Services
• Network Resources Brokering
• Get connection with desired QoS
• Ignore servers side resources (BW, CPU)

18
Resource Reservation Table
Reserved BW
bandwidth
Link i
Available BW (for other traffic)
Ri
Time(t)
19
End-to-End Resource Brokering
Network Resource Manager
Domain End-to-End Path Tables
Link i
Link 1
Link 2
20
User Subscription Database
Service Session Manager
21
Scenario and Parameters
C

• General Parameters
• Capacity, C 100Mbps
• Mean request arrival rate, ?
• Exponential mean session duration, µ-1300s
• Uniform random requested BW, Br 1..156units
• Where 1 unit is assumed 64kbps, hence Br
  64..10000kbps
• The mean is therefore 78.5 units (5024kbps)

22
Pricing Analysis
The goal is to Maximize
Revenue (1- ?)(meanBg. p . µ-1 pc) ?
Advance Reservation Charge?
Where ? Blocking Probability meanBg Mean BW
granted per request p Price per unit BW per
unit time pc Price per call µ-1 Mean session
duration per connection ? Mean request arrival
rate
23
Request/Reservation Load Profile
Link Capacity
Mean Request Load approx 80
24
BW Negotiation Policy
Policy 1 At Resource Limit IF (requested BW gt
min available BW for Ti) propose min available
BW IF (proposed min BW gt users tolerance
limit) reject
25
BW Negotiation at Resource Limit
Blocking Prob Rejected Request/ Total Request
26
BW Negotiation at Resource Limit
BWI BW granted/BW requested
27
BW Negotiation at Resource Limit
BW Volume Sold BW Granted Session Granted
28
BW Negotiation at Resource Limit
If BW Price is 0.03333p per sec or 2p per min for
guaranteed 64kbps connection
Extra with negotiation Over 25 hours
29
BW Negotiation Policies

• Policy 2At Different Load Levels
• IF (mean reservation load for Ti gt level X (e.g.
  60))
• propose (requested BW Y of requested BW)
• IF (proposed BW not accepted)
• give the requested BW anyway
• IF (requested BW gt min available BW for Ti)
• propose min available BW (same as policy 1)

30
Comparing with Neg at 60 Load
Proposed BW reduction(Y) 10, User BW
tolerance 20
31
Comparing with Neg at 60 Load
Over 25 hours
32
Comparing with Neg at 60 Load
Proposed BW reduction (Y) 10, user BW
tolerance 20
33
Comparing with Neg at 60 Load
34
What else can be negotiated?

• We can do adaptive negotiation schemes
• At different load levels based on prediction
• Be selective on different type of users based on
  BW usage such as reject short sessions, low BW
  reservation when the load is high
• We can also negotiate
• Reservation Start Time
• Reservation Session
• etc

35
Conclusion

• Agent-Enhanced Service Brokering Architecture
• OAM on Phoenix and Prototype
• Resource Reservation Table User Subscription
  Database to allow BW reservation in advance
• Evaluation of different BW negotiation policies
• Convergence of the best of both models
• (Circuit Switching guarantee and IP
  flexibility)

36
Issue 1 Why Agents?

• Below are the main arguments
• Autonomous
• Which most applications do
• Intelligent
• Can incorporate some of algorithms to perform
  negotiations
• Social Ability
• Therefore can interact among themselves via
  pre-defined protocol
• Java plays an important role within the agent
  community as it is object oriented,
  multi-platforms, scalable by nature

37
VLL Services
VLL via DS-TE QoS
ATM VC
DLCI
ATM Network
Frame Relay
MPLS Backbone
PE
PE
PPP, HDLC, etc
Ethernet
RSVP Enabled
RSVP Enabled
Resv Path
DiffServ Domain
Point B
Point A
ER
ER
VLL via EF(from A to B)
WAN
38
Advance Reservation
BA Book Ahead Service Start Time (Min 600s,
Max 86400s) Advance Reservation Users 20
39
Advance Reservation
40
Advance Reservation
41
Prototype Architecture (ref to TINA)