Version 8 (modified by, 14 years ago) (diff)


Toward Universal Redundant Traffic Elimination

Aditya Akella, Ashok Anand,University of Wisconsin
Srinivasan Seshan, Carnegie Mellon University

Internet traffic is increasing at a tremendous rate. Recent reports suggest that the total amount of traffic might grow 5X by 2013. This is expected to put tremendous strain on installed network capacity. If left unaddressed, this issue can severely impact user experience, particularly in accessing multimedia content. Unfortunately, the obvious solution of upgrading link speeds is too expensive and cannot keep up with the rapid traffic growth. We propose universal redundancy elimination (RE) to improve network efficiency and support robust user experience. In our RE framework, network elements keep track of recently forwarded packets in a local store, and strip redundant content from packets on the fly by comparing against the recently forwarded packets. The downstream network elements reconstruct full packets by substituting content from their store. Our approach expands the scope of existing duplicate suppression techniques (for example, Web caches) to the entire Internet as well as to all applications, including hitherto unforeseen ones.

We demonstrate the benefits of redundancy elimination using a GENI-based experiment. We emulate a popular on-demand video service and show how our framework allows both the service and the network capacities to scale; note that such services are predicted to be predominant sources of traffic in the future. Our GENI slice spans multiple sites and our experiment composes several different GENI software and hardware resources. We also leverage the flexibility that GENI provides to demonstrate the benefits of modifying other aspects of the Internet, e.g., Internet routing, based on traffic redundancy information.

Implications of Network-level Packet Caching

Brighten Godfrey, Ashish Vulimiri, University of Illinois

Many of the key shortcomings in the Internet---including unreliability, inefficiency, and security against traffic attraction attacks---can be traced to inflexible routing decisions that are fixed within the network. The Internet offers only a single path to each destination, and this path may be broken, congested, or insecure. An alternate approach is to decouple routing decisions from the network, allowing devices at the edge, such as source end-hosts or edge routers, to select paths on a per-packet basis informed by observed performance.

This project is developing and experimenting with pathlet routing, a highly flexible routing architecture that enables policy-compliant, scalable source-controlled routing. Specifically, we are studying how edge devices can effectively use multipath routing flexibility in the face of network dynamics via end-to-end performance observations. The project utilizes GENI's ProtoGENI and OpenFlow resources, whose deep programmability is essential for performance-sensitive measurements of the system, to enable novel experiments within the network. In our demo, we show how a streaming real-time media application can use the pathlet substrate's flexibility to weather "storms" on the network.


Jae Woo Lee, Jan Janak, Roberto Francescangeli, Suman Srinivasan, Eric Liu, Michael Kester, Salman Baset, Wonsang Song, and Henning Schulzrinne, "Internet Real-Time Lab, Columbia University"

NetServ Project Home Page

We present NetServ, a framework for deploying in-network services in the next generation Internet. Traditionally Internet nodes fall into one of two categories. Routers in the network core provide packet processing – forwarding, monitoring, and manipulating packets – but they do not normally provide addressable services. Servers at the network edge provide addressable services at the application layer, but they do not perform network-level services. NetServ blurs the distinction between routers and servers by providing a common platform for both types of network services. Our vision is to provide a common API, virtualized execution environment, and signaling protocol. Network services implemented as NetServ modules are freely installed, removed, and migrated among Internet nodes of all kinds – from backbone routers to set-top boxes.

GENI and NetServ share a mutually beneficial relationship. GENI makes at-scale experiments of NetServ possible. NetServ potentially provides an alternative method for researchers to interact with GENI nodes, allowing rapid and convenient creation of experiments. We demonstrate two network services implemented as NetServ modules at GEC9. ActiveCDN is a dynamic content distribution network (CDN) service. Unlike traditional CDN, ActiveCDN can incorporate content processing logic specific to a content provider. SIP Remote Agent shows how a Voice-over-IP (VoIP) service provider can offload server processing to routers at the network edge. The module can respond to NAT keep-alive messages or throttle incoming traffic during a distributed denial of service (DDoS) attack.

GENI-VIOLIN: In-Network Suspend and Resume for GENI Experiments

Pradeep Padala, Bob Lantz, Ulas C. Kozat, Ken Igarashi, DOCOMO USA Labs
Ardalan Kangarlou, Sahan Gamage, Dongyan Xu, Purdue University

Unexpected failures and outages will continue to affect the operation of cyber infrastructures like Amazon EC2 and network infrastructures like GENI. For many applications running in such infrastructures, such as long-running scientific jobs and networked system emulations, failure recovery means re-running the application from the beginning thus losing (partial) work done and wasting system resources. It is desirable for the infrastructure to provide efficient, application-transparent failure recovery capability that takes live "snapshots" of an infrastructure for future recovery or replay.

With advances in virtualization technologies, live snapshotting is feasible for a single virtual machine. However, the current technique is not adequate for suspending and resuming distributed experiments that run on GENI. GENI-VIOLIN's goal is to provide fast "live snapshotting" that allows suspend and resume of an entire GENI experiment distributed across multiple sites spanning multiple networks. Our contributions are three-fold: (1) A distributed live snapshot algorithm that allows snapshotting entirely in the network with no changes to end-host systems and minimal performance degradation; (2) A distributed live snapshotting system based on Openflow and Xen; (3) Demonstration of in-network live snapshotting over wide area network. GENI-VIOLIN can also be used for debugging distributed experiments and preempting long running experiments. During GEC9, we will demonstrate live snapshotting over two ProtoGENI sites (Utah and BBN) connected through a wide area network.

GENI in the Classroom

Jeannie Albrecht, Williams College

Distributed applications have become a core component of the Internet's infrastructure. However, many undergraduate curriculums do not offer courses that focus on the design and implementation of distributed systems. The courses that are offered often focus on the theoretical aspects of system design, but fail to provide students with the opportunity to develop and evaluate distributed applications in real-world environments, leaving students unprepared for graduate study or careers in industry. Historically, one main reason for this lack of preparation is a lack of computing infrastructure. Moving forward, the availability of GENI resources will address this limitation, making it possible to perform large-scale experimentation even at small colleges.

With this in mind, in this demo we highlight one potential assignment for an undergraduate Distributed Systems or Networks course that has a unique emphasis on giving students hands-on access to distributed systems. The assignment asks students to implement multiple distribution and caching strategies in a simple content distribution network, and evaluate their performance under different load patterns. In this particular case, we will use a combination of PlanetLab and ProtoGENI resources to deploy and evaluate our sample software package.



Nikhil Handigol, Stanford University

Effective load-balancing systems for services hosted in wide-area networks need to take into account both the congestion of the network and the load on the servers. In this demonstration, we show a comprehensive load-balancing system to minimize client response time and reduce system cost for such services. The system we show, called Aster*x, uses the global state of server load and network congestion, and dynamically routes the requests over appropriate (server, path) pairs calculated using the load-balancing algorithms we developed.

Aster*x runs in a GENI slice spanning 9 campuses across the US. We use GENI for the extensive deployment, evaluation, and demonstration of Aster*x. Aster*x exploits OpenFlow’s logically centralized controller to get the global network state and route flows of various granularities. It uses the Linux and PlanetLab-based computation substrate to host the replicated web service and to generate client requests from multiple locations. Besides the base behavior, we show in this demo the performance of different load-balancing algorithms.

Can I use all the diverse wireless capacity around me?

KK Yap, Stanford University

There is a common perception that wireless capacity is scarce and in short-supply. This is not true. Yet we are surrounded by abundant capacity that is off-limits to us. A cell phone can typically "see" 5-6 service providers, and a laptop can often see 10 or more WiFi networks. But because of long-term contracts, and WPA passwords, most wireless networks are not available for us to use. In my research, I am trying to change that, by allowing users to use any, or even many, wireless networks around them. To start with, I have built a platform to try out many techniques, such as fast, seamless handoff between WiFi and WiMAX networks. I discovered that the GENI testbed is a perfect place for me to try out my ideas, in a real network with real users and real applications. In this demo I will show you an experiment I have created on a GENI testbed at Stanford.

NowCasting: UMass/CASA Weather Radar Demonstration

Michael Zink, David Irwin, University of Massachusetts, Amherst

The foundation of better weather forecasting is better data. Scientists in CASA, an NSF Engineering Research Center, are studying experimental radar systems that comprise dense networks of small, controllable radars. These networks supplement and enhance NEXRAD by accurately sensing conditions close to ground where inclement weather often occurs. As a driving example, we show data from CASA's off-the-grid student testbed in Mayaguez, Puerto Rico. Last July, the testbed successfully detected the severe windstorms that delayed the Central American Games earlier than otherwise possible, which also enabled earlier warnings.

As a result of their accuracy, these systems produce vast amounts of streaming data from a multitude of geographically disparate sites. To be useful at scale, especially in time-critical situations, this data must quickly flow to processing centers that merge it to execute complex forecasting algorithms that predict the movements of weather patterns in real-time. Since inclement weather is rare, maintaining dedicated network/computing resources is a significant barrier to deployment at scale. This demonstration highlights an array of GENI technologies to remove this barrier, by experimenting with the execution of radar workflows and forecasting algorithms, developed by CASA scientists, on GENI and cloud networks that also include computing and sensing resources reserved on-demand.

ParkNet: WiMax

Ivan Seskar, Rutgers University
Thanasis Korakis, Polytechnic University of NYU
Max Ott, NICTA

This demonstration will showcase our research on ParkNet and how it can take advantage of GENI testbed resources. ParkNet is a mobile system of vehicles that collect road-side parking space occupancy information while driving by. Each ParkNet vehicle is equipped with a GPS receiver and a passenger-side-facing ultrasonic rangefinder to determine parking spot occupancy. The data is aggregated at a central server, which builds a real-time map of parking availability and could provide this information to navigation systems, or generate statistics for trip and city planning. Our results to date show that this system can generate useful parking data at significantly lower cost than stationary parking sensing systems.

In this demo, several ParkNet vehicles will comb the Brooklyn, NY downtown area and transmit their collected data through the GENI WiMax cell at Brooklyn Polytechnic University to servers located at the ORBIT testbed at WINLAB, Rutgers University. The raw sensor data as well parking availability will be visible on a map view of Brooklyn area that is updated when vehicles are in WiMax coverage and transmit data. The demo will also convey how the OMF framework can support such experiments.