After many years of using Linux vserver as PlanetLab's node provisioning mechanism, we have officially started migrating to LXC - the implementation of container-based virtualization in the Linux kernel. We have been working on this new release for several months and are pleased to announce that it has become stable and mature enough to deploy for users. Here are some of the highlights of the update...
- LXC - a fast and robust virtualization mechanism in the Linux kernel will be used to isolate slices
- We will no longer maintain our own version of the Linux kernel. LXC is supported natively by Linux, therefore the new software stack uses a stock kernel.
- The kernel will be updated to the 3.x series.
- Nodes are now managed using libvirt - a de facto standard for VM management - and an integral part of the OpenStack project.
- We have extended LXC in several ways for it to fit the needs of PlanetLab users - such as by securing the /proc file system and implementing IP address sharing (see below)
From a usability standpoint, we have tried to ensure that the transition will be smooth. PlanetLab users need to be aware of the following points during the update:
- All nodes will be reinstalled with short notice. A message will be sent out a week before the transition begins. Users will be expected to back up their data in this timeframe.
- Instead of seeing a physical interface with a public IP address, users will see a virtual interface with a private IP address, routed via NAT.
- It is however possible to bypass NAT use the physical interface directly using a new mechanism called "transforwad." See here for instructions on how to use it
- When you ssh into a node, you will be logged into your slice as root
... Other than these differences, all other functionality should be backwards compatible.
If you have any questions or concerns, then please don't hesitate to contact us.