Resource Provision and Allocation in Shared Network Testbed Infrastructures

1
Resource Provision and Allocation in Shared
Network Testbed Infrastructures

• Panayotis Antoniadis
• Timur Friedman
• Xavier Cuvellier
• Laboratoire LIP6 CNRS
• Université Pierre et Marie Curie Paris 6

2
Objective

• Achieve the desired level (and quality) of
  contributions
• Allocate resources efficiently
• Users sites and private-PLCs
• Direct control over the amount of resources
  contributed
• System planner The manager of the Shared-PLC
• Defines the resource provision/allocation rules
• Users decide to participate or not and their
  exact contribution
• Trusted software
• Monitoring of resource utilization
• Ability to implement global resource allocation
  policies
• Inter-Shared PLCs federation (e.g.
  Onelab-Planetlab)

3
Overview

• Planetlabs solution
• Provision simple contribution rule (Min. 2
  nodes)
• Allocation (almost) unlimited consumption, equal
  sharing
• On-going research on market-based mechanisms
• Practical issues related to the use of virtual
  currencies
• Not suitable economic models
• Our goal
• Design a rule-based incentive mechanism relating
  contribution with consumption
• Bottom-up approach

4
Current situation
5
Uptime
6
Avg. CPU load
7
Effective Contribution

• Computing and network resources
• CPU power, bandwidth, storage
• Available vs. utilized
• Geographical coverage
• Public good but congestible
• Also technology coverage (onelab)
• Uptime/stability
• Services

8
Provision and allocation

• How much each should contribute?
• Who has priority in case of congestion?
• Economics social welfare maximization
• Maximize total value/utility minus total cost
• Cost and utility are hidden information!
• Possible solutions
• Equal treatment (as in Planetlab today)
• Resource exchange/currency (encourages provision)
• Congestion pricing/auctions (efficient
  allocation)
• Differentiated services

9
Resource exchange
10
Fixed Contributions
11
Different levels of service
12
Different levels of service

• Database level
• Max. slices
• Visible nodes
• Max. nodes
• Service provision
• Priorities (e.g. bandwidth, CPU, ports)
• Reservation privileges (max. time slot, max.
  amount of resources, priorities)

13
A rule-based approach

• Sites with higher (effective) contribution should
  be granted a higher service level
• Unlike markets regulation is needed
• Need to define beforehand the different levels of
  service and contribution and their mapping
• This is price we have to pay for reducing
  complexity and capturing the public good aspect
• No control of actual resource utilization
• But due to central control (and possibly more
  sophisticated resource allocation) fixed
  contribution uptime should be enough
• Additional incentive issues arise when there is
  not central control

14
But markets have problems

• Require some sort of virtual currency
• Inflation, deflation, etc.
• Regulation may also needed
• E.g. expire time of currency
• Differentiate between services and experiments
• Complexity
• Multiple resources
• Could harm the community spirit
• In addition to ignoring the public good aspect of
  the shared resources

15
Methodology (Bottom up)

• Study monitoring data from Planetlab
• Congested resources (CPU, storage, bandwidth)
• Patterns of use and effective contribution
• Feedback from the community
• Start with a simple rule
• Classes of service at a database level (taking
  advantage of unique/popular resources uptime
  for contribution
• Add complexity based on observation

16
Future work

• Conclude on-going work -)
• Federation issues
• Economic modelling and game-theoretic analysis
• Backup the overall approach, insights for
  parameter tuning
• Monetary contributions (e.g node hosting)
• Social incentives