Topology Switching

This research explores topology switching, a new paradigm for managing datacenter networks.  Topology switching allows individual applications to qualitatively specify the network performance they desire, and admits a new unit of allocation for software-defined networks: the routing task.   We are currently exploring how to efficiently compile, install, and run such tasks using Openflow networking.    

Blender: Mixing multiple isolation models on one fabric

Blender is our current topology-switched network prototype for creating and allocating datacenter routing tasks.  Large data centers host thousands of services and end-user applications.  For efficiency, they multiplex tenants across the physical infrastructure, leveraging server virtualization to carve out isolated units of CPU, memory, and storage.  In contrast, tenants typically receive loose, qualitative descriptions of network performance.   Operators employ one of many possible network isolation models to apportion network resources across tenants.  However, each model strikes its own particular balance between efficiency and predictability.

In reality, there are many useful isolation models that data center operators will want to run.  For instance, while some tenants benefit from multiplexing all traffic over a multi-path fabric, other services call for strict isolation.  Blender is a system for authoring and managing multiple isolation models on one physical network.  It can express and compose isolation schemes that provide both fixed and proportional bandwidth shares.  In addition, Blender provides modular isolation components, such as work conservation and distributed rate limits, that can be combined to extend existing schemes.  We have implemented an OpenFlow-based prototype to explore the expressiveness, scale, and performance of our design and hope to release the software using research-only license soon.

Funding and industry partners 

  • 2009 National Science Foundation award number 0917339
  • 2012 Ericsson grant through the Center for Networked Systems
  • 2012 Broadcom Foundation grant
  • 2012 Google Faculty Research Award
  • Cisco
SSL.UCSD: 2012