Information Technology Reference
In-Depth Information
custom mechanisms and have multiple active mechanisms. Mechanism manage-
ment here refers to the programming constructs to transparently include any
arbitrary mechanism by exposing a standard interface. A mechanism manager
controls how bids and asks are passed to a mechanism and when instances are
created and destroyed; a
Goal Observatory
for defining goals and keeping track
of their adherence via the monitoring framework; a
exchange management
that
keeps track of all incoming asks and bids, matches, as well as one or more active
mechanisms; and finally,
adaption management
as an instantiation of a market
adaption component.
The
Simulator Layer
is the basis for the simulator. It includes a singleton
event handler
, as this enables a simple programming model without the need
for complex thread or concurrency management, and a
tick manager
to control
“time” in the simulator as a sequence of discrete periods. In each tick, we invoke
participants in a renewed random order, and give them the option to “act”, i.e.
do something in the market. We also define a
scenario controller
, which through
the event handler can instigate new scenarios for observation, based upon the
current time. The scenario controller permits us to create issues of instability
or change specific settings in the market to study how the market changes, and
later how adaption actions have improved or worsened the situation. We can
also layer (simple) scenarios to create more complex compound scenarios.
We also define key
utilities
to assist in market simulation. These include:
readers for trace data from existing markets to “stimulate” market events or
scenarios as well as writers to store monitoring data; a
participant factory
to
facilitate the generation of multiple participant types based on a set of input
parameters; and as a key premise for all simulators, a
random number generator
which can simply be the inclusion of the Colt Library
1
or similar.
6 Summary
In this paper, we have argued that existing electronic market platforms are in-
sucient for immature domains like Cloud computing. Therefore, we proposed
the concept of an autonomic marketplace platform: the automatic adaption of
the economic models of the platform and its underpinning infrastructure based
upon a given concept of “market performance”. We described how the auto-
nomic MAPE-K loop can be applied to an electronic market platform. Finally,
we presented our experiences in trying to build a simulation tool as a premise
for the study and evaluation of an autonomic market using GridSim as a case
study. However, we encountered too many scenario specific obstacles that mer-
ited a bespoke simulation framework. Building on top of the lessons learned
from GridSim, we defined a conceptual framework for a market simulator that
can facilitate different aspects of study for an autonomic market. These include
the definition of destructive scenarios, stochastic events and extended user types.
Our future work is the continued investigation of our simulation tool, its on-going
development as well as the development of scenarios for its calibration.
