> Yahoo! is not affiliated with the authors of this page or responsible for its content.
Adaptive Information Flow Mechanisms and Management for Power Grid Contingencies Adaptive Information Flow Mechanisms and Management for
Power Grid Contingencies

Stian F. Abelsen, Erlend S. Viddal, K. Harald Gjermundr�d, David E. Bakken

and
Carl H. Hauser

Technical Report EECS-GS-012
Washington State University
School of EECS
Pullman, WA 99163

December 2007
Adaptive Information Flow Mechanisms and Management for Power Grid
Contingencies
Stian F. Abelsen , Erlend S. Viddal , K. Harald Gjermundr�d , David E. Bakken
�
and Carl H. Hauser
Washington State University
School of Electrical Engineering and Computer Science
Pullman, Washington, USA
{stian.abelsen, eviddal, haraldg}@gmail.com, {bakken, chauser}@wsu.edu
Abstract
GridStat is a QoS-managed publish-subscribe frame-
work for data delivery for the electric power grid. Grid-
Stats Data Plane delivers data updates through a network
of middleware-level Status Routers. Subscriptions are man-
aged by GridStats hierarchical QoS Management Plane.
The path allocation computations are typically done of-
ine and beforehand, but are complex, not only due to the
multiple QoS constraints but the number of status routers
that would be involved an entire power grid.
In a cri-
sis, many entities may suddenly wish to add a large num-
ber of subscription requests, which would in practice over-
whelm the subscription allocation mechanisms. In this pa-
per we present a mechanism called modes, which lets Grid-
Stat change routing tables quickly. Modes can be either
global, or only active at a given scope within the hierarchy.
We present the design and experimental evaluation of Grid-
Stat modes and of two different mode change algorithms
that provide different tradeoffs of performance and consis-
tency.
1
Introduction
The electrical power grid is highly dependent on data
monitoring and control capabilities in order to better under-
stand and manage power transmissions over a highly com-
plex network of transmission lines and substations. SCADA
(Supervisory Control and Data Access) has in the last 40
years served as the electrical power grids communication
system and incorporates the requirements and network tech-
nologies from the time that it was developed. The require- Current afliation: Eltek Valere, Drammen Norway Current afliation: Simula Innovation, Oslo Norway Current afliation: University of Cyprus, Nicosia Cyprus
�
Contact author
ments for communication in the electrical power grid are
changing [7]. Growing concerns about terrorist attacks,
changes in the power ow structure after the regulatory re-
structuring in the 1990s, new uses of technologies (Intelli-
Grid [3]) and an increased overall load to capacity ratio of
the transmission line system demand a more exible and
adaptive communication network. The SCADA communi-
cation system features a centralized star-topology, point to
point communication, lack of multicast, severe bandwidth
constraints and proprietary protocols which are not suf-
cient to meet the requirements of todays grid. Further dis-
cussion of the limitations of SCADA can be found in [6]
and [2].
GridStat is designed to address the need for a exible and
robust communication system for the electrical power grid
using a specialized publish-subscribe paradigm [5]. Grid-
Stat middleware manages network resources, enables reli-
able delivery of data to any point and provides QoS (Qual-
ity of Service) for data streams. GridStat hides the details
of lower-level network capabilities from application devel-
opers in order to enable the communication system to be
deployed across different network technologies, operating
systems, programming languages and device types. Grid-
Stat is divided into two planes; the management plane and
the data plane. The management plane consists of a hier-
archy of QoS brokers that collectively manage resources
and subscriptions in the data plane. The data plane is a
virtual message bus that lets publishers provide data to the
network and enables subscribers to establish subscriptions
to status data through a status router network. The use of
QoS, on a per-subscription basis, allows subscribers to spec-
ify multiple redundant delivery paths (spatial redundancy),
subscription interval and delay. Furthermore, GridStat pro-
vides status data delivery to multiple recipients at different
rates through the multicast property and the ability to con-
trol and switch routing tables in the status router network at
run-time through the use of modes. A mode contains the necessary forwarding rules for a set
of subscriptions. Using pre-congured modes allows the
status router network to quickly switch between bundles of
subscriptions, an action called a mode change. The pro-
cess of establishing individual subscriptions is a resource-
intensive operation requiring deallocation and allocation of
subscriptions. Doing these computations at run-time is ex-
pensive and may result in unsatisfactory subscription delays
[8]. Modes allow subscription bundles to be allocated and
pre-loaded into the status routers routing tables and then
control which routing tables the status router network uti-
lizes at any given time.
The mode change mechanism will help utility compa-
nies control centers, regional control centers, ISOs and
nation-wide monitoring centers to perform pre-contingency
planning for communication needs and to switch subscrip-
tion bundles when contingencies do occur in the electrical
power grid. For example, modes are one way to exploit
GridStats capability for data load shedding in the commu-
nication infrastructure in analogy with the power load shed-
ding in the electrical power grid. For example, subscribers
could specify two QoS sets: the desired QoS and minimal
acceptable QoS, and the management plane can automati-
cally switch between them when the network is congested.
The research contributions of this paper are: Global and hierarchical modes: QoS brokers dene
and use modes to adapt communication in their respec-
tive administrative domains. Multiple simultaneously active routing tables in the
data plane and the ability to switch between routing
tables at run-time. Design and implementation of two mode change algo-
rithms with different tradeoffs. Experimental evaluation comparing the mode change
algorithms in terms of performance, resource usage
and variance (time) in the presence of various tempo-
rary network conditions.
2
Status Dissemination and GridStat
GridStat is a publisher-subscriber framework that targets
application domains where the majority of data are made
available at periodic time intervals. It is mainly designed to
serve as a exible and robust communication system for the
electrical power grid. Figure 1 shows a small scale GridStat
deployment subdivided in a management plane and a data
plane. The management plane consists of QoS broker mod-
ules that collectively control and manage resources in the
data plane. The data plane is populated by status routers,
publishers
and subscribers, where publishers provide data
and subscribers can subscribe to data. The management hi-
erarchy handles subscription requests and establishes paths
from the publisher to the subscriber through a sequence of
status routers. More detailed information about GridStat
and other baseline mechanisms can be found in [4] and [6].
Figure 1. Status dissemination architecture
2.1
Management Plane
The lowest level of the management plane consists of
leaf QoS brokers
. A leaf QoS broker manages and pro-
vides services to a set of status routers, publishers and sub-
scribers. The leaf QoS broker manages a at collection
of status routers, called a cloud. A leaf QoS broker has
complete control over all available resources and the corre-
sponding resource usage in the cloud. The resources include
status routers, publishers, subscribers and the links connect-
ing them. Leaf QoS brokers must control and make sure
no allocated subscriptions exceed an event channels band-
width constraints. Additionally, the leaf QoS brokers must
ensure that routing tables and computational resources are
not overloaded in the status routers. The main responsibility
of a leaf QoS broker is to control the allocation or deallo-
cation of subscription paths between publisher-subscriber
pairs in its cloud, and to ensure that the allocated path satis-
es the QoS requirements specied by the subscriber.
Interior QoS brokers
denote all non-leaf QoS brokers in
the management hierarchy. Interior QoS brokers manage
multiple clouds and offer services to lower-level QoS bro-
kers, and whose main responsibility is to allocate and deal-
locate inter-cloud subscriptions.
2 2.2
Data Plane
The data plane is a virtual message bus where subscrip-
tion data ows between publishers and subscribers. The vir-
tual message bus consists of interconnected status routers,
whose main purpose is to forward status events from pub-
lishers to the subscriber applications that requested the data.
A status router is in effect a router with additional function-
ality to provide forwarding of status events when subscribed
to a