Ongoing Research

• Network Virtualization Towards a Future Internet

  Today’s network architectures are stifling innovation, restricting it mostly to the application level while the need for structural change is increasingly evident. The absence
  of adequate facilities to design, optimize and interoperate new networks currently forces a convergence to an architecture that is suboptimal for many applications and that
  cannot support innovations within itself, the Internet.
  
  We therefore introduce a set of radical architectural approaches built on our networking systems background in order to overcome this impasse. We improve our ability
  to design inter-operable and complementary families of network architectures by enabling the co-existence of multiple networks on common platforms through virtualization
  of networking resources.
  
  
  

• Software Router Virtualization on Commodity Hardware

  A crucial prerequisite for any network virtualization architecture is virtual routers with a higher level of flexibility and programmability than today's commercial routers.
  Multi-core chips can turn commodity hardware into a strong platform for software router virtualization by enabling forwarding rates of several Mpps for minimum-sized
  packets.
  
  In this context, we have designed and implemented a platform which leverages multi-core CPUs and hardware multi-queuing to provide high performance and highly
  configurable forwarding planes for advanced programmability. The platform enables the consolidation of virtual data planes onto a single forwarding domain, which
  results in significantly higher performance than forwarding in separate guest domains. We also introduce a forwarding path architecture which allows multiple cores to
  concurrently access a physical network interface and process packets in parallel. At the same time, packets and their associated data structures are typically confined
  to a single cache hierarchy, avoiding costly main memory accesses.

Past Research

• Adaptive Rate Control for Multimedia Applications in the Internet

  Achieving smooth transmission for media-streaming applications is challenging, especially when the provided smooth rate control should not compromise bandwidth
  efficiency and fairness. We observe that although multiplicative decrease is necessary to achieve fairness, it does not necessarily sacrifice the system throughput, as long
  as the system operates between the points knee and cliff.
  
  In this context, we improve the smoothness of AIMD rate control for multimedia applications, within the framework of bandwidth efficiency, fairness, and TCP friendliness.
  We introduce rate control parameters which are adaptable to current network conditions to avoid the damage of static multiplicative decrease on throughput performance
  and smoothness. This rate adjustment strategy relies on RTT estimation and can be easily adapted and incorporated into existing transport protocols.
  
  
  

• Budget Oriented Packet Service

  As the Internet continues to expand in size, diversity and reach, Quality of Service is a fundamental problem for Internet users today. We propose a new pricing mechanism,
  namely Budget Oriented Packet Service (BOPS), for Internet packet services. According to this model, each flow manages its own packets within the confines of some
  budget per flow, allowing service differentiation among packets of the same flow.
  
  BOPS allows for dynamic adjustment of strategies based on current network contention and service requirements of competing packets. A central property of this model is
  that users (instead of network operators) take charge of their individual service, exploiting dynamically the service tradeoffs at their wish. Service is therefore, by nature, adaptive
  and individual.
  
  
  

• Cross-layer Design for Real-time Streaming Applications in Wired/Wireless Networks

  TCP’s insistence on reliable delivery without timing considerations has an adverse effect on the performance of streaming applicattions, especially when their data packets
  bear information with a limited useful lifetime. On the other hand, UDP lacks all basic mechanisms for flow/congestion control. Furthermore, multimedia applications are
  expected to run in physically heterogeneous environments composed of both wired and wireless components.
  
  We propose Scalable Streaming Video Protocol (SSVP) which enables AIMD-oriented congestion control on top of the light-weight UDP achieving smooth transmission
  patterns, while maintaining TCP-friendliness. SSVP incorporates the combined dynamics of sending rates, queuing delay and link losses to enhance media delivery. In the event
  of limited or dynamically changing bandwidth, SSVP is augmented with an application-level layered adaptation mechanism which interacts with the transport protocol adapting
  video quality along with long-term variations in the available bandwidth.