Reservation Schemes for Grid Markets - DigiBiz'09


Published on

Published in: Technology, Economy & Finance
  • Be the first to comment

  • Be the first to like this

No Downloads
Total views
On SlideShare
From Embeds
Number of Embeds
Embeds 0
No embeds

No notes for slide

Reservation Schemes for Grid Markets - DigiBiz'09

  1. 1. Reservation Schemes for Grid Markets Digibiz 2009 London, June 18 th , 2009 Marcel Risch Marcel Risch
  2. 2. Introduction <ul><li>One of the major issues that must be addressed in open Grid markets are reservations. </li></ul><ul><li>Since Grid customers will not only wish to buy resources shortly before they are needed but also far in advance of actual usage, reservations are vital to keep such an economy operating. </li></ul><ul><li>The open question is: Which reservation type is best suited to utility computing environments? </li></ul>
  3. 3. Reservation Schemes In Grid Markets <ul><li>The Distributed European Infrastructure for Supercomputing Applications (DEISA) provides computing power mainly to researchers from participating countries. </li></ul><ul><li>In addition to researchers, private enterprises can also use some of the computing power for which they have to pay. </li></ul><ul><ul><li>However, the reservations are scheduled according to the convenience of DEISA members </li></ul></ul><ul><li>The Amazon Elastic Compute Cloud (EC2) was the first grid market in which anybody can buy resources </li></ul><ul><ul><li>Nowadays, subscriptions have been introduced, allowing users to pay a flat fee and a usage fee if they require resources in the long term </li></ul></ul><ul><li>Tsunamic Technologies provides clustering services to paying customers. It offers the following options: </li></ul><ul><ul><li>On-Demand Cluster: A simple clustering scheme without any requirements. </li></ul></ul><ul><ul><li>Subscription Clusters: In this scheme, users agree to use a minimum number of CPU-hours per month </li></ul></ul>
  4. 4. Other Reservation Types <ul><li>Restaurants-Style </li></ul><ul><ul><li>The customer orders the resources but does not pay for them in advance </li></ul></ul><ul><ul><li>The provider then reserves the resources ordered. If the customer is not ready by the starting time, the reservation is canceled </li></ul></ul><ul><li>Airline-Style </li></ul><ul><ul><li>The customer orders the resources and pays for them in full </li></ul></ul><ul><ul><li>In the case of cancellations, there are three different models to consider: </li></ul></ul><ul><ul><ul><li>The customer is refunded the entire payment </li></ul></ul></ul><ul><ul><ul><li>The customer is returned the payment minus a fixed fee </li></ul></ul></ul><ul><ul><ul><li>The customer cannot cancel the ticket at all. </li></ul></ul></ul><ul><ul><li>The risk incurred by a cancellation is not entirely shouldered by the provider in the last two cases </li></ul></ul>
  5. 5. Other Reservation Types <ul><li>Mixed approach </li></ul><ul><ul><li>In this reservation scheme, reservations do not expire if the customer is late. </li></ul></ul><ul><ul><li>Instead, the length of time that the resources will be kept available depends on the non-refundable down payment the client is willing to pay. </li></ul></ul><ul><ul><li>The client can specify how long the resources should be available through the down payment. </li></ul></ul><ul><ul><li>The provider can determine the likelihood that the resources will be actually used by the customer and can plan the cash flow </li></ul></ul>
  6. 6. Refunds <ul><li>We will change the refund scheme be more flexible and to level the playing field between both parties </li></ul><ul><li>Ideally, the size of the refund should be proportional to the remaining time to the beginning of the intended resource usage. </li></ul><ul><ul><li>If customers cancel far ahead of the actual resource usage, they would get a large part of their money refunded. </li></ul></ul><ul><li>This approach naturally leads to the question of how to determine the size of the refund. </li></ul>
  7. 7. Refund Options <ul><li>Linear Model </li></ul><ul><ul><li>In this model, the refund is directly proportional to the time remaining to the start of the reservation </li></ul></ul><ul><ul><li>This approach is easy to understand for buyers and providers </li></ul></ul>
  8. 8. Refund Options <ul><li>Non-Linear Model (I) </li></ul><ul><ul><li>In this model, a concave-up exponential function is used </li></ul></ul><ul><ul><li>This approach will encourage buyers to decide quickly whether they want to cancel a reservation or not </li></ul></ul>
  9. 9. Refund Options <ul><li>Non-Linear Model (II) </li></ul><ul><ul><li>In this model, a concave-down function is used </li></ul></ul><ul><ul><li>This approach will encourage customers to reserve resources, since the cost of cancellations is very low </li></ul></ul>
  10. 10. Refund Scheme Selection <ul><li>The prepaid model has a number of advantages for providers, but which is the best? To answer this question, we have to consider </li></ul><ul><ul><li>The demand </li></ul></ul><ul><ul><li>The willingness of customers to commit to resource purchases </li></ul></ul><ul><ul><li>The refund models used by other providers </li></ul></ul><ul><ul><li>The objectives of the provider </li></ul></ul><ul><li>In the case of high demand and a high willingness to commit to resources, providers can choose the concave-up refund scheme </li></ul><ul><li>To attract more customers, the linear model or concave-down scheme may be more suitable instead </li></ul><ul><li>The concave-down refund scheme can also be used to bind frequent customers </li></ul><ul><li>In the case of high demand and a low willingness to commit to resources, the provider can hope at best to use the linear model and may have to resort to the concave-down refund scheme </li></ul>
  11. 11. Refund Scheme Selection <ul><li>To answer this question, we also have to consider what other providers offer </li></ul><ul><li>If other providers offer mainly the concave-down or linear refund schemes, customers will prefer these </li></ul><ul><li>If most providers use the concave-up refund model, the provider may be able to attract more customers by using the less strict linear model </li></ul><ul><li>If most providers use the linear of concave-down models, using the concave-up model would be counter productive </li></ul><ul><li>Thus, the best refund scheme depends on the market situation, the actions of other providers and consumer expectations </li></ul>
  12. 12. Conclusion <ul><li>We have presented a number of reservation schemes and described their refund schemes </li></ul><ul><li>We have also shown that alternative refund schemes may be invaluable in Grid environments </li></ul><ul><li>Which scheme to select depends on the market environment </li></ul>
  13. 13. Questions?