Dube price indices: difficult to incorporate variable priced components
since price is not related to market ratio: if you request more, you will pay more...
partial solution: determine indices for non-variable price components and associate an average of the variable priced components with this
e.g. each instance has an average of 1.6 GB outgoing bandwidth
difficulty: how to scale above/below this average without knowing price relation between e.g. bandwidth and other components
match specific ratio’s (=provider specific)
core problem: not enough information about how price is determined by providers
Similar bid matching
search the billboard for bids that are similar to the bid the consumer wishes to make
define similarity measure
bids are similar if bid constraints are similar, use ratio’s to compare numerical constraints (linear, quadratic, ...)
use a predetermined ratio for (non)matching absolute constraints (e.g.: “ubuntu 10.10”)
aggregate constraint similarity to find bid similarity
price for new bid is then, e.g. (weighed) average price of x most similar bids
problems:
constraint similarity ratio’s influence price: in reality not all constraint’s had the same influence on price
e.g. a difference of 2% in similarity of one constraint can have a larger influence on the final price of a match than a 50% similarity difference of an other constraint
this can also be true within a single constraint! (e.g. a high-end CPU that is 10% faster than an other high-end CPU will have a larger influence on the price than a low-end CPU that is 10% faster than an other low-end CPU).
this is especially difficult for the variable priced components vs the other components
What do we do with bids that do not have (enough) similar bids on the billboard ?
e.g. small instance that consumes a lot of bandwidth
Core difficulties
known: price P and set of variables X that are combined in a (non)-linear way: impossible to determine individual price components
More general question: is price discovery useful in an ecosystem where providers claim to have unlimited capacity and let each consumer pay the same amount?
there is no efficiency gain (for the consumer) if a consumer wants to pay more than the actual cost: the consumer is just paying more than an other consumer who is getting the same service.
(this is also one of the reasons why a double auction is not interesting?)
Of course, in the end, price discovery is always a heuristic.
Dube’s heuristic is also not able to cope with cases that have not occurred yet
price equilibrium nice to have, but probably not feasible in current market
“any function that evolves in the direction of the match price will provide an equilibrium eventually”.
The question is, how good is the heuristic ?
find a metric to measure this (boxplot of difference between discovered price and actual price ?)
As more standardization occurs, evolve to sub market that compete on CRA market (as with the desktop cloud market).