Institute of Technology Reading and Special Problems - Final Exam | CS 597 | Exams Computer Science

Using the Access Grid as a Testbed for Network

Management Research

Cynthia S. Hood, Anthony DeSalvatore, Satish Devarapalli, Nirav Gadhia,

Sanjay Hegde, Vikram Mallikarjuna, Shashank Shankar, Srividya

Visweswarababu, Srikanth Yoginath

Illinois Institute

of Technology

The network management problem is a very practical, yet complex and theoretically

interesting problem. Understanding the problem requires exposure to the issues faced in

managing a complex system. Given the lack of theoretical models for

large

scale

networks, we seek to understand the issues through data analysis. Toward this end, we

are using the Access Grid (AG) as a testbed for network management research. The AG

is a large

scale distributed collaboration environment where audio, video

, and data is

exchanged over high

-performance networks using multicasting.[2] It is a research

environment that is actively used for collaboration across geographic distances. Over the

past year, we have installed our own AG node at IIT.

Operational Inf

ormation

There are currently over 100 AG nodes around the world.[7] There is an established

community of AG node operators and technical experts willing to discuss and assist on

topics ranging from operational issues to research problems. There is signif

icant

informal documentation of problems. One of the tools most frequently used by the

community is the AG mud, a text

based chat system. A wide variety of topics related to

the AG are discussed on the mud and the sessions are archived. By searching thr

ough the

archives, we can identify the issues considered most pressing. Additional information on

the network infrastructure can be found through the Quilt [9] and Abilene Network and

Operations Center website [1]. The resources of this community can be d

rawn upon to

answer questions, arrange multi-node experiments and test solutions. An additional

benefit of running our own AG node is being part of the AG community and gaining first

hand operational experience.

Measurements

The AG spans a variety of in

teresting networking technologies involving LANs, WANs,

multicast and multimedia. It is a large

scale networked system that allows us to study

problems in a realistic manner. The ability to collect significant amounts of meaningful

data is key. The data c

ollected may include statistical measurements, events,

configuration information, problem descriptions, and any other information that will

provide insight into the system or network state.

We focus on audio and video applications. Application performan

ce is used to identify

periods of degraded service. There is a measurement infrastructure in place to collect

network, system, and application level measurements across the grid. This infrastructure

takes advantage of existing measurements and tools to cr

eate a unified measurement log.

Partial preview of the text

Download Institute of Technology Reading and Special Problems - Final Exam | CS 597 and more Exams Computer Science in PDF only on Docsity!

Using the Access Grid as a Testbed for Network

Management Research

Cynthia S. Hood, Anthony DeSalvatore, Satish Devarapalli, Nirav Gadhia,

Sanjay Hegde, Vikram Mallikarjuna, Shashank Shankar, Srividya

Visweswarababu, Srikanth Yoginath

Illinois Institute of Technology

The network management problem is a very practical, yet complex and theoretically interesting problem. Understanding the problem requires exposure to the issues faced in managing a complex system. Given the lack of theoretical models for large-scale networks, we seek to understand the issues through data analysis. Toward this end, we are using the Access Grid (AG) as a testbed for network management research. The AG is a large-scale distributed collaboration environment where audio, video, and data is exchanged over high-performance networks using multicasting.[2] It is a research environment that is actively used for collaboration across geographic distances. Over the past year, we have installed our own AG node at IIT.

Operational Information There are currently over 100 AG nodes around the world.[7] There is an established community of AG node operators and technical experts willing to discuss and assist on topics ranging from operational issues to research problems. There is significant informal documentation of problems. One of the tools most frequently used by the community is the AG mud, a text-based chat system. A wide variety of topics related to the AG are discussed on the mud and the sessions are archived. By searching through the archives, we can identify the issues considered most pressing. Additional information on the network infrastructure can be found through the Quilt [9] and Abilene Network and Operations Center website [1]. The resources of this community can be drawn upon to answer questions, arrange multi-node experiments and test solutions. An additional benefit of running our own AG node is being part of the AG community and gaining first hand operational experience.

Measurements The AG spans a variety of interesting networking technologies involving LANs, WANs, multicast and multimedia. It is a large-scale networked system that allows us to study problems in a realistic manner. The ability to collect significant amounts of meaningful data is key. The data collected may include statistical measurements, events, configuration information, problem descriptions, and any other information that will provide insight into the system or network state.

We focus on audio and video applications. Application performance is used to identify periods of degraded service. There is a measurement infrastructure in place to collect network, system, and application level measurements across the grid. This infrastructure takes advantage of existing measurements and tools to create a unified measurement log.

Links to many of the measurement efforts can be found at [6]. CAIDA has comprised a list of monitoring tools at [3]. Using these tools, different types of information can be collected in different ways.

Measurements can be collected using active or passive methods. Active methods add synthetic load to the system or network and observe the resulting performance. Passive methods collect measurements resulting from existing load. Our measurement infrastructure utilizes both of these methods. We consider the infrastructure to be a starting point for data analysis. The goal is to determine the utility of different types of information under different circumstances. We are collecting

SNMP Management Information Base (MIB) information from hosts and routers. MIB data contains both static and dynamic information
End-to-end measurements including Round Trip Time (RTT), packet loss, one- way delay, out of order packets and duplicate packets
Routing histories and changes
System information regarding processor, memory, disks, network and other devices
Application (vic and rat) performance measures. Vic and rat are the video and audio streaming applications on the AG
Multicast beacon measurements [5].

We have been logging many of these measurements for several months. We are in the final stages of enabling measurement collection from Abilene and regional gigapops. Presently, we are getting readings on most of the measurements every fifteen seconds. As the utility of different measurements is better understood, the monitoring frequency can be adjusted. The goal of this measurement infrastructure is to provide a data set for studying the management problem, not for real-time monitoring.

Simulation To complement the “real” data, we also have a simulation of the AG. This allows us to simulate performance problems and test solutions. A simulation model of the AG can be an important tool for understanding how problems occur and propagate, as well as for testing the trigger models that are developed as part of this research. The tool is particularly useful in terms of the network, because the opportunity for inserting faults and testing models is very limited there.

We simulate the AG [4] using the OPNET [8] network simulation tool. OPNET provides models for different types of equipment and protocols across many different technologies. The simulation project is broken into two pieces, (1) the AG node simulation, and (2) the network simulation.

[4] focuses on simulating the AG node. A typical AG node contains four machines; video capture machine, video display machine, audio machine and control machine. Our simulation models the first three machines. The primary function of the control computer is audio control. Since we have a good environment for studying the audio control

Institute of Technology Reading and Special Problems - Final Exam | CS 597, Exams of Computer Science

Related documents

Partial preview of the text

Download Institute of Technology Reading and Special Problems - Final Exam | CS 597 and more Exams Computer Science in PDF only on Docsity!

Using the Access Grid as a Testbed for Network

Management Research

Cynthia S. Hood, Anthony DeSalvatore, Satish Devarapalli, Nirav Gadhia,

Sanjay Hegde, Vikram Mallikarjuna, Shashank Shankar, Srividya

Visweswarababu, Srikanth Yoginath

Illinois Institute of Technology