ADAPTING GRID SERVICES FOR URGENT COMPUTING

ENVIRONMENTS

Jason Cope and Henry M. Tufo

Department of Computer Science, University of Colorado, 430 UCB, Boulder, CO, 80304-0430, U.S.A.

Keywords:

Data Grids, Urgent Computing, Grid and Web Services, Service-Oriented Architectures.

Abstract:

Emerging urgent computing tools can quickly allocate computational resources for the execution of time criti-

cal jobs. Grid applications and workﬂows often use Grid services and service-oriented architectures. Currently,

urgent computing tools cannot allocate or manage Grid services. In this paper, we evaluate a service-oriented

approach to Grid service access and provisioning for urgent computing environments. Our approach allows

resource providers to deﬁne urgent computing resources and Grid services at a much ﬁner granularity than pre-

viously possible. It accommodates new urgent computing resource types, requires minimum reconﬁguration

of existing services, and provides adaptive Grid service management tools. We evaluate our service-oriented,

urgent computing approach by applying our tools to Grid services commonly used in urgent computing work-

ﬂows and evaluate management policies through our urgent service simulator.

1 INTRODUCTION

Recent research into urgent computing systems has

improved several dynamic data-driven workﬂows that

perform emergency computations. The goal of ur-

gent computing is to provide a cohesive infrastruc-

ture to support time-critical computations. Examples

of these applications and workﬂows include severe

weather forecasting workﬂowssuch as those provided

by Linked Environments for Atmospheric Discov-

ery (LEAD) (Droegemeier et al., 2004), the Southern

California Earthquake Center’s (SCEC) TeraShake

earthquake simulation applications (Cui et al., 2007),

the Southeastern Universities Research Association

(SURA) Coastal Ocean Observing and Prediction

(SCOOP) storm surge modeling applications (Bog-

den et al., 2007), epidemic transmission simulations

using tools like Virginia Tech’s EpiSims application

(Eubank et al., 2006), and the Data Dynamic Sim-

ulation for Disaster Management project’s Coupled

Atmosphere-Fire (CAF) wildﬁre forecasting work-

ﬂow (Mandel et al., 2007).

The LEAD and SCOOP projects use the Spe-

cial PRiority and Urgent Computing Environment

(SPRUCE) to obtain high-priority access to the

shared computing resources available on the TeraGrid

(Catlett et al., 2007), a distributed supercomputingen-

vironment in the US similar to DEISA (Lederer et al.,

2007). SPRUCE provides project users with elevated

and automated access to TeraGrid computational re-

sources so that high-priority applications run imme-

diately or as soon as possible (Beckman et al., 2007).

SPRUCE currently provides capabilities for allocat-

ing computational resources but does not yet support

urgent storage or data management capabilities. Ur-

gent storage and data management capabilities pro-

vide cohesive infrastructure for prioritized usage of

storage resources, such as ﬁle systems, data streams,

and data catalogs. Since several of these applications,

such as LEAD and TeraShake, have signiﬁcant data

requirements, providing urgent storage and data man-

agement is an essential and currently absent capability

for urgent computing workﬂows. Supporting end-to-

end urgent computing workﬂows requires support for

common data capabilities, such as data storage, ac-

cess, search, and manipulation capabilities, required

by these workﬂows.

We are developing an Urgent Data Management

Framework (UDMF) to address the data requirements

of urgent computing workﬂows. This framework will

manage data related tasks and provision storage re-

sources for urgent workﬂow usage. Several urgent

computing workﬂows use service-oriented architec-

tures for resource and application integration. These

application deployments typically operate in multi-

ple modes and share services between project deploy-

135

Cope J. and M. Tufo H. (2008).

ADAPTING GRID SERVICES FOR URGENT COMPUTING ENVIRONMENTS.

In Proceedings of the Third International Conference on Software and Data Technologies - PL/DPS/KE, pages 135-142

DOI: 10.5220/0001893301350142

 SciTePress

ments. While the current SPRUCE infrastructure can

negotiate access to compute resources for these sys-

tems, it does not support similar capabilities for the

Grid services utilized by these workﬂows. In order to

support end-to-end urgent workﬂow scheduling and

data management, the Grid services that provide data

capabilities to these workﬂows should also support

authorization and usage capabilities similar to other

urgent computing resources.

In this paper, we describe our use of monitoring,

authorization, and provisioning infrastructure to pro-

vide urgent computing capabilities to Grid services.

We provide these capabilities by leveraging past re-

search on Grid authorization and monitoring infras-

tructure and using Service Level Agreements (SLA)

for establishing and sustaining Quality of Service

(QoS). Providing these urgent Grid service capabili-

ties allows resource providers to directly apply urgent

computing policies to Grid services. These policies

can indirectly provision resources used by a Grid ser-

vice. This feature is especially useful for providing

urgent computing capabilities to software or resources

that can not directly support urgent computing capa-

bilities. Our approach is reusable by other Grid ser-

vices and requires minimal reconﬁguration of existing

services.

The remainder of this paper describes our current

work on urgent Grid services. Section 2 describes

the urgent computing paradigm and the current urgent

computing infrastructure. Section 3 describes related

areas of research. Section 4 describes our urgent ser-

vice provisioningand policy management framework.

Section 5 details our initial evaluation of this frame-

work through its use by Grid resource management

and data services. In the ﬁnal sections of this paper,

we present future work and conclusions.

2 BACKGROUND

2.1 The State of Urgent Computing

The Special PRioirty and Urgent Computing Envi-

ronment (SPRUCE) (Beckman et al., 2007) enables

on-demand resource allocation, authorization, and se-

lection for urgent computing applications and work-

ﬂows. This environment provides on-demand access

to shared Grid computing resources with a token-

based authorization framework. By utilizing shared

resources, SPRUCE allows data centers and virtual

organizations to utilize existing computing infrastruc-

ture for emergency computations instead of allocat-

ing dedicated resources for these tasks. Users sub-

mitting SPRUCE jobs specify a color-coded urgency

parameter with their job description. SPRUCE au-

thorizes the urgent job execution request by verifying

that a user is permitted to execute jobs with the spec-

iﬁed urgency on the target resource. Each resource

provider deﬁnes policies for how the urgent tasks are

handled on a per-resource basis. For example, a re-

source provider may choose to preempt non-urgent

jobs for high-priority tasks or to give the urgent tasks

next-to-run privileges. The infrastructure is currently

deployed on several TeraGrid resources, including re-

sources at NCAR, UC/ANL, SDSC, and TACC, and

is used by several Grid computing workﬂows, such as

LEAD and SCOOP.

While SPRUCE provides access to computational

resources, it does not provision or negotiate access to

other common Grid services and capabilities, such as

data management services, resource management ser-

vices, and Grid credential management services. To

completely support urgent computing in Grids, these

other services must be supported so that urgent work-

ﬂows do not bottleneck against these unmanaged ser-

vices.

2.2 Service Requirements of Urgent

Computing Workﬂows

Recent service-oriented systems have provided many

capabilities for users to integrate into their workﬂows.

For example, LEAD provides a large set of services

hosted across the project partner sites. The available

services include data management services that can

assimilate data, reference metadata catalogs, stream

data from remote sensing equipment, and executedata

mining operations. LEAD leverages the core Globus

Grid services (Foster, 2005), such as the Grid Re-

source Allocation and Management (GRAM) service,

to interact with computational and storage resources.

Similar to LEAD, SCOOP provides a series of data

services exposed as Grid and Web Services. These

services provide capabilities to access catalog data,

transform data into compliant formats, and visualize

data sets. Like LEAD, SCOOP implements these ser-

vices as Web or Grid services.

Since one of the beneﬁts of Web and Grid services

is software reuse, these services are used by projects

and users other than SCOOP or LEAD. SCOOP has

stated that their tools are available for community use.

Furthermore, these systems also provide multiple op-

erational modes and use cases. Not only is LEAD

capable of forecasting severe weather events such as

tornadoes, it is also used as an educational and train-

ing tool. The combination of multimodal use and the

lack of urgent computing infrastructure to provision

these services necessitates infrastructure to manage

ICSOFT 2008 - International Conference on Software and Data Technologies

136

Storage Resource

Data Manager

Data Processing

Task

Urgent Computing

Authorization

Infrastructure

Urgent Multi-

Resource Manager

Urgent Storage Manager

Urgent Data

Manager

Computational Resource

Urgent Computational

Resource Manager

Web

Service

Network

Service

Web

Service

Network

Service

Urgent Service Access

Manager

Data Client Task Request

Figure 1: UDMF architecture.

and provision access to these services. In the event of

an urgent workﬂow, it is not acceptable for the work-

ﬂows performance to be degraded because of conges-

tion at these services and the resources they manage.

2.3 Urgent Data Management

We are developing an Urgent Data Management

Framework (UDMF) that provides the necessary data

management capabilities for urgent computing work-

ﬂows (Cope and Tufo, 2008). This framework pro-

vides users with priority based access to data capa-

bilities used in urgent computing workﬂows. These

tools provide storage and network resource provision-

ing, autonomic resource management, and resource

management policies that address urgent computing

data requirements. Figure 1 illustrates the compo-

nents and interactions of UDMF. UDMF leverages

existing infrastructure (illustrated as the solid compo-

nents in Figure 1) when possible. UDMF contains

several software managers that can execute the re-

source management policies and negotiate urgent user

access to the data resources. The UDMF components

are illustrated as the dashed components in Figure 1.

Many of the existing software components are exten-

sible, such as the Data Storage Interface to GridFTP

(Allcock et al., 2002) or the Authorization and Au-

thentication Framework in the Globus Web service

container.

The focus of this paper is on the Grid service com-

ponentsof UDMF. These Grid service speciﬁc UDMF

components provide monitoring and instrumentation

for Grid service invocations, user authorization con-

sistent with other urgent computing tools, and adap-

tive resource provisioning tools and polices to manage

user access and interaction with Grid services.

3 RELATED WORK

The research we present in this paper draws upon

the experiences from several related areas of work,

including Grid security, Service Level Agreements

(SLA) for Grid services, and Grid Quality of Service

(QoS). Several authorization and authentication tools

exist for use in Grids. The Globus Toolkit provides

the bulk of the authorization capabilities required

(Foster, 2005). We use the Globus Policy Decision

Point (PDP) and Policy Information Point (PIP) au-

thorization chains to negotiate access to Grid services

during the invocation of a service operation. Several

other projects use the same authorization infrastruc-

ture for integration of security policies and authoriza-

tion tools. GridShib integrates Shibboleth within the

Globus authorization frameworks using custom PDP

and PIP infrastructure (Chadwick et al., 2006; Lang

et al., 2006). The Virtual Organization Management

System (VOMS) also provides similar capabilities to

GridShib using the Globus authorization and authen-

tication framework (Alﬁeri et al., 2003).

In this paper we leverage past work on Web

service QoS and Service Level Agreements (SLA)

techniques for supporting urgent computing environ-

ments. In the context of service-oriented architec-

tures, our work on Grid service provisioning is most

similar to past research on SLAs. The use of SLAs is

still an active area of research, but they have not yet

been used to support urgent computing applications

and resources. The infrastructure to support SLAs

in UDMF is similar to other SLA management tools

that use brokers to advertise, negotiate, and establish

SLAs between clients and services (Dan et al., 2004).

Grid QoS research is often associated with SLA

research. QoS monitors typically observe the prop-

erties of systems and SLA brokers use these obser-

vations to determine attainable and sustainable QoS

for a speciﬁc SLA. Several projects have used this

model for Grid systems (Liu et al., 2004; Wang et al.,

2005; Truong et al., 2006; Al-Ali et al., 2004). Our

work is similar to request prioritization and QoS dif-

ferentiation techniques that adapt service executions

to priority-based policies (Sharma et al., 2003; Zhou

et al., 2007; Benkner et al., 2007; Erradi and Ma-

heshwari, 2007). Other tools, such as the Network

Weather Service (NWS), use collected observations

to forecast expected QoS for a system (Wolski et al.,

2005; Nurmi et al., 2007).

ADAPTING GRID SERVICES FOR URGENT COMPUTING ENVIRONMENTS

137

Grid Service Client

Grid Service Container

Urgent

PDP

Urgent

PIP

Grid Service

Implementation

Managed Resource

Urgent

Authenticator

Urgent Grid

Service Manager

Urgent Request /

Response Handler

Instrumentation

Database

Figure 2: Implemented urgent service authorization frame-

work.

4 IMPLEMENTATION OF THE

URGENT SERVICES

FRAMEWORK

Our urgent Grid service framework consists of in-

formation services to communicate the state, capa-

bilities, and policies associated with available urgent

services. It also monitors the usage of urgent Grid

services and other services collocated with an urgent

Grid service. We provide authorization infrastructure

to determine the level access to a service a user is

permitted. Figure 2 illustrates these components and

their interactions.

4.1 Advertising and Monitoring Urgent

Grid Services

Information services are a critical component of

service-oriented architectures so that end-users or

other software components can recognize the prop-

erties and capabilities of deployed services. To help

identify available urgent computing services hosted

on a resource, we developed a mechanism to reg-

ister and describe urgent computing services with a

local Globus Monitoring and Discovery System v4

(MDS4) (Schopf et al., 2006). A Perl script pro-

duces XML descriptions of supported urgent services

that describe service usage policies. This information

can be used in conjunction with QoS information to

model the behavior of the service under current or ex-

pected conditions. These tools provide valuable in-

formation for time critical applications. It is essential

for end users to determine their authorized urgency

level so they can select the service that best accommo-

dates their needs. The registration of a Grid service’s

urgency parameters with information services allows

end users to verify service properties before a service

is used.

Grid service invocations are monitored through

our tools. All Grid service interactions, regardless

of their urgency, are monitored. We extended a Grid

service message handler distributed with Globus WS

CORE to log Web service communication into our

instrumentation database. We capture several ﬁelds

form each SOAP header sent or received by the ser-

vice container, including the messages UUID, the tar-

get service address, the service operation, the caller’s

Distinguished Name (DN), the UUID of a related

message, and the time the message was delivered to

our message handler. This handler is installed on the

service containers input and output message handler

chain so that every SOAP message is logged. This in-

formation is used by our policy and service manage-

ment tools to adapt policies based on current service

usage characteristics.

4.2 Urgent Grid Service Authorization

The framework that authorizes urgent access to Grid

services consists of custom Globus Grid service au-

thorization framework that integrates with urgent

computing authorization tools, such as SPRUCE. Our

authorization infrastructure consists of a custom Pol-

icy Decision Point (PDP) and a custom Policy In-

formation Point (PIP) that supports urgent access to

these services. The PDP is responsible for authoriz-

ing client access to Grid services and communicates

with our service management tools. When the PDP

evaluates the incoming service request, the DN for

the user invoking the service is veriﬁed and the ap-

propriate urgency is applied to their invocation of the

service. This veriﬁcation occurs through execution

of the Urgent Authenticator, which is similar to the

SPRUCE Token Authentication script. Once the ur-

gency level has been veriﬁed, the PDP contacts the

Urgent Service Manager. The Urgent Service Man-

ager determines which policy to execute for the ser-

vice and returns this information to the PDP. The PDP

then executes the appropriate policy and permits the

service to execute.

ICSOFT 2008 - International Conference on Software and Data Technologies

138

4.3 Policy Enforcement and Adaptation

The Urgent Service Manager is responsible for deter-

mining which service management policy associates

with a particular user and urgency level. Once the ap-

propriate policy is identiﬁed and adapted to the cur-

rent operating environment, the Urgent Service Man-

ager communicates the policy to the Urgent PDP. The

Urgent Service Manager can simulate the executionof

Web services. We integrated the SimPy discrete event

simulation package into the Urgent Service Manager

client so that we can quickly simulate and experiment

with anticipated Web service requests ﬂows. We en-

vision that the service management and simulation

components can integrate with other autonomic com-

puting tools to foster an automated, self-managing,

urgent computing environment.

We have currently deﬁned and implemented three

service management policies:

• Unmanaged. Requests are permitted as they are

received regardless of the urgency of the request.

• Exclusive. Only the most urgent requests are al-

lowed. This policy is similar to strict priority

scheduling.

• Shared. A mix of requests are permitted at vari-

able rates per urgency level.

The unmanaged policy allows all requests to com-

plete. The exclusive access policy only permits the

most urgent requests and stalls all other requests un-

til the urgent requests complete. The stalled requests

are not denied so that concurrent, less urgent work-

ﬂows do not fail. The shared policy allows multiple

levels of urgent requests to execute and arbitrates the

frequency that the requests are permitted to execute.

Like the exclusive access policy, requests are not out-

right denied but are stalled to achieve allowable ser-

vice execution frequencies. Our PDP throttles service

requests at the authorization point and shapes the Grid

service request trafﬁc.

Each of the deﬁned policies has several conﬁgura-

tion parameters. For example, all policies have an ex-

ecution window parameter to deﬁne how long the pol-

icy should remain active since the last request of that

policy occurred. This can block non-urgentrequests if

more urgent requests are expected. The shared policy

deﬁnes trafﬁc percentages and rates for each urgency

level. This parameter prevents non-urgent trafﬁc from

starving urgent trafﬁc. Multiple policies can be mixed

per service. For example, the most urgent policy for a

service can implement an exclusive policy for a spe-

ciﬁc user while the other policies can be applied to

other users.

Trace of Job Submissions to SDSC DataStar

Time (min)

104500

104700 104900 105100 105300 105500

User ID

105

120

135

150

Normal Request

High Priority Request

Figure 3: SDSC DataStar trace illustrating jobs submitted

per user over a 16 hour interval.

5 EVALUATION OF UDMF GRID

SERVICES

Our evaluation of this service provisioning infrastruc-

ture occurred in two phases. In the ﬁrst phase, we

evaluated service provisioning and request prioriti-

zation in our service management simulator. In the

second phase, we deployed our service provisioning

and management tools into the Globus Web service

container and evaluated the capabilities of the service

management tools with several services.

5.1 Policy Simulation

We performed more detailed analyses of our provi-

sioning software using our simulator and data from

the Parallel Workloads Archive (Feitelson, 2008a).

We used the trace of jobs submitted to SDSC’s DataS-

tar TeraGrid computational resource from March

2004 to April 2005. For our evaluation, we assume

that all jobs submitted to DataStar were using the

Globus WS GRAM Grid service so that we could ap-

proximate a workload trace for a Grid resource using

Grid services. WS GRAM manages Grid applications

running on Grid computational resources and invokes

other Grid services to transfer data between Grid re-

sources. Figure 3 illustrates job submission times for

a segment of the trace. Over the 16 hour period of

the trace, 1010 jobs were submitted to DataStar from

40 different users. From this trace, we studied the

performance and impact of our software by varying

the urgency of users within the trace. As an example,

when user 17 has high priority and exclusive access to

the resource while all other users have low priority ac-

cess, our software limits the low priority requests as

illustrated in Figure 4. In this example, the window

parameter of the high priority policy was set to one

hour and prevented low priority tasks from executing

a service within one hour of the last high priority re-

quest. Low priority tasks were also limited to four

ADAPTING GRID SERVICES FOR URGENT COMPUTING ENVIRONMENTS

139

Trace of Rate Limited Job Requests for SDSC DataStar

Time (min)

104500

104700 104900 105100 105300 105500

Request rate (Job requests / min)

High Priority Request

High Priority Response

Normal Request

Normal Response

Figure 4: SDSC DataStar trace illustrating the effects of

urgent request throttling on job throughput over a 16 hour

interval.

service accesses per hour for all users.

The parameters that deﬁne higher priority poli-

cies impact the trafﬁc throughput of the lower priority

policies. Figure 5 illustrates how the window param-

eter of the high priority policy affects all lower pri-

ority Grid service trafﬁc. Expanding or contracting

the high priority window affects what trafﬁc is per-

mitted to execute. Smaller windows allow lower pri-

ority requests to execute closer to the original service

request. For example, a non-urgenttrafﬁc stream exe-

cuted concurrently with a high priority request con-

ﬁgured with a 30 minute window exclusive access

policy. Lower priority stream interleaved the higher

priority request stream. Larger windows delay non-

urgent requests longer and can prevent the interleav-

ing behavior of the shorter windows.

While the window parameter can reserve access

to the service if more requests are expected, it de-

grades the response time of the lower priority re-

quests. Figure 6 illustrates the response time of low

priority request over the hourly intervals in the sim-

ulation. Without a window, small delays of less than

one second still occur since the high-priority service

requests have exclusive access to the service until the

request completes. As the window values increase

in size for this simulation, the response time for the

lower priority requests increases as the requests ex-

ecute later in the simulation. Adapting the request

scheduling parameters, such as the access window, to

urgent workﬂows is necessary so that the services are

still highly utilized by all priorities.

Conﬁguring the policies is left to the service host-

ing site. Since workloads and service usage patterns

can change as the workﬂows change, we are develop-

ing tools to automatically conﬁgure these policy pa-

rameters. We have begun to implement several ser-

vice request rate, load, and usage pattern forecasters

within the Urgent Service Manager. These forecasters

use time-series analysis techniques (Feitelson, 2008b)

and more recent forecasting techniques (Nurmi et al.,

2007) to predict the expected usage or response of a

Grid service for urgent computing workﬂows. These

Response Time of Non−Urgent Requests

Time (min)

104500

104700 104900 105100 105300 105500

Average Response Time (s)

100

1000

10000

30000

Response rate (0 min window)

Response rate (30 min window)

Response rate (60 min window)

Response rate (90 min window)

Response rate (120 min window)

Figure 6: Response time of normal priority SDSC DataStar

trafﬁc delayed by higher priority trafﬁc.

prediction tools are still under development, but the

initial results from these tools for tuning policy pa-

rameters are promising.

5.2 Deployment Experiences

To evaluate our urgent service provisioning frame-

work, we integrated our tools into several services,

evaluated our service models and provisioning poli-

cies with our simulator, and performed end-to-end

tests using our authorization and provisioning tools

on several deployed services. We ﬁrst integrated our

authorization framework into several services, includ-

ing the Globus DefaultIndexService, a data catalog

service that utilizes the Globus Replica Location Ser-

vice (RLS), and a database query Grid service. With

all of these services, we were able to control access

to the Grid service with only minimal conﬁguration.

The reconﬁguration of the services included adding

the PDP and PIP to the service security conﬁguration

so that the priority of the request could be validated.

The service manager was also conﬁgured so that the

priority validation requests from services running in

the container could check and schedule access to the

service.

6 FUTURE WORK

We are currently developing forecasting tools to au-

tomatically tune the service policies. This automatic

tuning and conﬁguration work leverages techniques

that mine for patterns in Grid service usage and fore-

cast expected Grid service behavior or load. The work

we present in this paper does not address supporting

non-service oriented access to urgent data resources

and we leave this topic for future work. Addition-

ally, we are developing an autonomic data manage-

ment system to adapt and reconﬁgure storage man-

agers for urgent data requirements.

ICSOFT 2008 - International Conference on Software and Data Technologies

140

Impact of Varying Exclusive Access Policy Parameters on Non−Urgent Requests

Time (min)

104500 104700 104900 105100 105300 105500

Request and Response Rates (1 / min)

Request rate

Response rate (30 min window)

Response rate (60 min window)

Response rate (90 min window)

Response rate (120 min window)

Figure 5: SDSC DataStar trace illustrating the effects of urgent policy parameters on non-urgent request trafﬁc.

7 CONCLUSIONS

We presented our approach to provisioning Grid ser-

vices for urgent computing applications and work-

ﬂows. Our approach requires minimal service recon-

ﬁguration for urgent use cases and can support a va-

riety of resources types managed by Grid services.

We evaluated our service management tools using the

core Globus data and resource management Grid ser-

vices. We demonstrated that these services can adapt

and respond to dynamic urgent computing require-

ments using our service and policy simulator. The

current scheduling parameters for urgent services can

provide near immediate access to these services at the

cost of increased response times for lowerpriority ser-

vice requests.

ACKNOWLEDGEMENTS

Support for this research and University of Colorado

computer time was provided by National Science

Foundation (NSF) MRI Grant #CNS-0421498, NSF

MRI Grant #CNS-0420873, NSF MRI Grant #CNS-

0420985, and a grant from the IBM Shared Univer-

sity Research (SUR) program. This research was sup-

ported in part by the NSF through support for the

National Center for Atmospheric Research and Ter-

aGrid resources provided by UC/ANL and NCAR.

We would like to thank the members of the SPRUCE

project, including Pete Beckman and Suman Nadella,

for their guidance and support of this research.

REFERENCES

Al-Ali, R., Haﬁd, A., Rana, O., and Walker, D. (2004). An

approach for quality of service adaptation in service-

oriented Grids. Concurrency and Computation: Prac-

tice and Experience, 16(5):401–412.

Alﬁeri, R., Cecchini, R., Ciaschini, V., Dell’Agnello, L.,

Frohner, A., A. Gianoli, K. L., and Spataro, F. (2003).

VOMS, an Authorization System for Virtual Organi-

zations.

Allcock, B., Bester, J., Bresnahan, J., Chervenak, A.,

Foster, I., Kesselman, C., Meder, S., Nefedova, V.,

Quesnal, D., and Tuecke, S. (2002). Data Manage-

ment and Transfer in High Performance Computa-

tional Grid Environments. Parallel Computing Jour-

nal, 28(5):749 – 771.

Beckman, P., Beschatnikh, I., Nadella, S., and Trebon, N.

(2007). Building an Infrastructure for Urgent Com-

puting. High Performance Computing and Grids in

Action.

Benkner, S., Engelbrecht, G., Middleton, S., Brandic, I.,

and Schmidt, R. (2007). End-to-end qos support for

a medical grid service infrastucture. New Genera-

tion Computing, Computing Paradigms and Computa-

tional Intelligence, Special Issue on Life Science Grid

Computing, 25(4):355–372.

Bogden, P., Gale, T., Allen, G., MacLaren, J., Almes,

G., Creager, G., Bintz, J., Wright, L., Graber, H.,

Williams, N., Graves, S., Conover, H., Galluppi, K.,

Luettich, R., Perrie, W., Toulany, B., Sheng, Y., Davis,

J., Wang, H., and Forrest, D. (2007). Architecture of a

Community Infrastructure for Predicting and Analyz-

ing Coastal Inundation. Marine Technology Society

Journal, 41(1):53–71.

Catlett, C., Andrews, P., Bair, R., and et al. (2007). Ter-

aGrid: Analysis of Organization, System Architec-

ture, and Middleware Enabling New Types of Appli-

cations. High Performance Computing and Grids in

Action.

Chadwick, D., Novikov, A., and Otenko, A. (2006). Grid-

Shib and PERMIS Integration. Campus-Wide Infor-

mation Systems, 23(4):297–308.

ADAPTING GRID SERVICES FOR URGENT COMPUTING ENVIRONMENTS

141

Cope, J. and Tufo, H. (2008). A Data Management Frame-

work for Urgent Geoscience Workﬂows. In Proceed-

ings of the International Conference on Computa-

tional Science (ICCS 2008).

Cui, Y., Moore, R., Olsen, K., Chourasia, A., Maechling,

P., Minster, B., Day, S., Hu, Y., Zhu, J., Majumdar,

A., and Jordan, T. (2007). Enabling Very–Large Scale

Earthquake Simulations on Parallel Machines. In Pro-

ceedings of the International Conference on Computa-

tional Science (ICCS) 2007, Beijing, China. Springer.

Dan, A., Davis, D., Kearney, R., Keller, A., King, R., Kue-

bler, D., Ludwig, H., Polan, M., Spreitzer, M., and

Youssef, A. (2004). Web services on demand: WSLA-

driven automated management. IBM Systems Journal,

43(1):136–158.

Droegemeier, K., Chandrasekar, V., Clark, R., Gannon, D.,

Graves, S., Joesph, E., Ramamurthy, M., Wilhelmson,

R., Brewster, K., Domenico, B., Leyton, T., Morris,

V., Murray, D., Pale, B., Ramachandran, R., Reed,

D., Rushing, J., Weber, D., Wilson, A., Xue, M.,

and Yalda, S. (2004). Linked Environments for at-

mospheric discovery (LEAD): A Cyberinfrastructure

for Mesoscale Meteorology Research and Education.

In Proceedings of the 20th Conference on Interac-

tive Information Processing Systems for Meteorology,

Oceanography, and Hydrology, Seattle, WA. Ameri-

can Meteorological Society.

Erradi, A. and Maheshwari, P. (2007). Enhancing web ser-

vices performance using adaptive quality of service

management. In Proceedings of the 8th International

Conference on Web Information Systems Engineering

(WISE 2007).

Eubank, S., Kumar, V. A., Marathe, M., Srinivasan, A., and

Wang, N. (2006). Structure of Social Contact Net-

works and Their Impact on Epidemics. AMS-DIMACS

Special Volume on Epidemiology, 70:181–213.

Feitelson, D. (2008a). Parallel workloads archive,

http://www.cs.huji.ac.il/labs/parallel/workload/.

Feitelson, D. (2008b). Workload modeling for computer

systems performance evaluations.

Foster, I. (2005). Globus Toolkit Version 4: Soft-

ware for Service-Oriented systems. In IFIP Interna-

tional Conference on Network and Parallel Comput-

ing. Springer-Verlag.

Lang, B., Foster, I., Siebenlist, F., Ananthakrishnan, R.,

and Freeman, T. (2006). A Multipolicy Authorization

Framework for Grid Security. In Proceedings of the

Fifth IEEE Symposium on Network Computing and

Application, Cambridge, MA, USA.

Lederer, H., Pringle, G. J., Girou, D., Hermanns, M. A.,

and Erbacci, G. (2007). Deisa: Extreme computing in

an advanced supercomputing environment. Parallel

Computing: Architectures, Algorithms and Applica-

tions, 38:687–688.

Liu, Y., Ngu, A., and Zeng, L. (2004). QoS Computation

and Policing in Dynamic Web Service Selection. In

Proceedings of the 13th International Conference on

World Wide Web 2004 (WWW2004).

Mandel, J., Beezley, J., Bennethum, L., Chakraborty, S.,

Coen, J., Douglas, C., Hatcher, J., Kim, M., and Vo-

dacek, A. (2007). A Dynamic Data Driven Wildland

Fire Model. In Proceedings of the International Con-

ference on Computational Science (ICCS) 2007, pages

1024–1049, Beijing, China.

Nurmi, D., Brevik, J., and Wolski, R. (2007). QBETS:

Queue Bounds Estimation from Time Series. In Pro-

ceedings of the 13th Workshop on Job Scheduling

Strategies for Parallel Processing.

Schopf, J., Pearlman, L., Miller, N., Kesselman, C., Foster,

I., D’Arcy, M., and Chervenak, A. (2006). Monitoring

the Grid with the Globus Toolkit MDS4. In Proceed-

ings of SciDAC 2006.

Sharma, A., Adarkar, H., and Sengupta, S. (2003). Manag-

ing qos through prioritization in web services.

Truong, H., Samborski, R., and Fahringer, T. (2006). To-

wards a Framework for Monitoring and Analyzing

QoS Metrics of Grid Services. In Proceedings of the

Second IEEE International Conference on e-Science

and Grid Computing.

Wang, G., Wang, C., Chen, A., Wang, H., Fung, C.,

Uczekaj, S., Chen, Y., Guthmiller, W., and Lee, J.

(2005). Service level managment using QoS moni-

toring, diagnostics, and adaptation for networked en-

terprise systems. In Proceedings of the Ninth IEEE In-

ternational EDOC Enterprise Computing Conference.

Wolski, R., Obertelli, G., Allen, M., Numri, D., and Brevik,

J. (2005). Predicting grid resource performance on–

line. Handbook of Innovative Computing: Models,

Enabling Technologies, and Applications.

Zhou, X., Wei, J., and Xu, C. (2007). Quality-of-service

differentiation on the internet: a taxonomy. Journal of

Network and Computer Applications, 30(1):354–383.

ICSOFT 2008 - International Conference on Software and Data Technologies

142