Charging for GRID Access Uli Harder Imperial College London

Charging for GRID Access
Uli Harder
Work before progress together with Peter Harrison, Maya Paczuski
Imperial College London
• What are GRIDs?
• What are they used for?
• Why charging for access?
• How can you charge for computer access?
• A simple example
• El Farol
• Modelling larger systems
What are GRIDs?
• collection of computing and storage resources
• alomst like distributed computing
• BUT, it is easier for the user to find and use services, using middleware
• similar to www: server + client + Google
• ideally the GRIDs should be heterogeneous
Who uses them?
• There are two main reasons to use a GRID: computation and storage intensive
• right now: particle theorists, they expect to produce 1Pb/sec when LHC comes
online, data
• Astronomers, data
• Meterologists, computation
• Human Genome project
• Medical imaging (mammography)
• future users: banks, insurance companies, env. agencies, games?
• corporate problem: there is no equivalent of VPN for computing and storage
resources - virtualization
What can you charge for and why would you want
to do it?
• resources: CPU, memory, data storage, information on resources
• this has all been done before for main frames and network usage
• for GRIDs this is more complicated because the resources are decentralised
• ultimately charging is about fixing the price for contracts with users
• if you “own” a GRID there are many reasons why you want to charge people,
even if it is only nominal money
– fairness, management
– maximise utilisation
– cost recovery
– profit maximization
– spare capacity
– selling computing power to corporations
How to charge
• the idea of the GRID contravenes having a complicated human negatiation for
each computation
• resources become commodities
• auctions
• stock market of resource commodities
• bartering (even here there needs to be a kind currency to compare resources)
• problems of charging
• phone-like contracts
• fixing the price
• different contracts
• closely related: security
• trust in correctness of results
Some approaches
• Sutherland, 1968, describes a scheme to max. utilisation of a PDP-1 using
• mainframes
• cpu charging (SPAWN)
• Popcorn
• G-commerce
• Cumulus
El Farol
• Brian Arthur, 1994
• customers judge the popularity of a bar (El Farol) by the number of other
customers, they will try to avoid a crowded bar
• each have got different rules to decide whether they expect the bar to be
croweded, depending on their previous visits, and whether it’s worth going to
the bar
• customer do not exchange their information
• the system aranges itslef to be at the desired mean number of customers in the
• extended to the Minority Game.
Our approach
• agent-based simulation of minority game
• incorporate features of computers (in contrast to planes computers actually
cope with ovebooking, briefly)
• do I max. utilisation with a particular charging scheme
• is the “economy stable”
The research was funded by EPSRC (research grant QUAINT, GR/M80826).