Virtual network switching across hypervisors
|Time:||11:30 - 12:15|
|Day:||Friday 22 January 2010|
|Location:||Renouf 2 (MFC)|
|Project:||Kernel Based Virtual Machine|
With the increasing introduction of cloud computing and virtualized servers into the everage data center, network management complexity also increases significantly. New questions that appear include:
- How does the network administrator manage switches that are part of the hypervisor instead of a box from his favourite vendor?
- What kind of setup do you need to support multiple virtual machines with 10 gigabit/s throughput on a single box?
- When each virtual machine needs its own random MAC address, how do you guarantee that they are all unique?
- What do you need to do to make packet filtering rules persistent once you migrate a VM to another server? What about switch port statistics then?
For these and other questions, we will look at the development work that is going on in KVM, and compare them to how other hypervisors work on the same issues.
Arnd Bergmann has been hacking on the Linux kernel professionally since he joined IBM in 2001.
At the IBM Linux Technology Center in Böblingen, Germany, Arnd has worked on the s390 architecture, device drivers, virtualization, 64 bit system call emulation, file systems and networking. He is the maintainer for the Cell Broadband Engine Architecture featured in machines from the Playstation to the fastest supercomputer in the world.
Arnd is currently part of a cross-hypervisor team at IBM that is redefining networking for the next generation of virtual servers, working on KVM in particular.