An Approach to Auction-Based Web Server Admission Control - Developing Concepts in Applied Intelligence - page 102

- the performance index defining an income of the web service (sum of submitted
offers),
a scheduling admission strategy should be found to select the set of offers S d for proc-
essing, which increases the performance index and assures the client satisfaction si-
multaneously in every time instant ( T ). In this paper only the situation, in which
server resources are not sufficient to satisfy all users, because of the fact that the total
sum of expected bandwidth exceeds the total volume of bandwidth offered by the web
service, is considered.
4 Algorithms
The managing process should increase the income of the web service (the sum of ser-
viced offers) and should satisfy the web clients. Two scheduling algorithms will be
discussed and examined. The first one - class-based auction algorithm is more justi-
fied and the second one - profitability based auction algorithm based only on pro-
posed offers.
The first algorithm guarantees, that some offers from every request for every class
will be selected in respect of the just treatment and customer satisfaction . The later
policy does not consider class based selection and can lead to postponement of large
files because of the same price both for small and large files. In both cases also a
modified criterion could be defined, taking in account the client satisfaction. It could
be e.g. the price paid by users (charge). When the client will pay finally the lower
price than declared, he will be satisfied [1], [8], [9], [10]. The scheduling algorithm
will run in following periods, which are defined by the periodic instances of the sche-
duling interval T (i.e. every 30 min).
4.1 Class-Based Auction Algorithm
In this approach, it is assumed, that every resource belongs to one class [3]. The
classes are determined by the data size, i.e. three classes are there distinguished:
- the bronze for data size less than 10MB,
- the silver for data size between10MB and 50 MB,
- the gold for data greater than 50MB.
The next assumption is that the defined piece of total bandwidth is assigned justify for
all classes, it means that from every class some offers will be served.
The rest will be assigned to offers containing the maximal value of profitability in-
dex. The approach algorithm consists of three phases.
Phase 1. The first phase of the algorithm based on the clear rules - to every class
some bandwidth will be assigned. Let's assume that 60% of the total bandwidth is
assigned for all the requests classes.
We consider three classes, so that, to every class 20% of total bandwidth will be
assigned. For every class the offers with the highest price will be scheduled until the
bandwidth for class is available.