This paper is an extended version of our first KSI paper [1], which I had previously discussed.
As the previous paper had earned an award for best paper at the conference, we were invited to prepare an extended version to submit to the IARIA journals.
So we set out to address the short comings of our first implementation of KSI and developed KSI 2.

With KSI 2, we significantly improved the network throughput and enabled support for multi-node clusters such that it can now provide feature complete rootless Kubernetes clusters within Slurm jobs, without significant performance overhead.
We achieved this by replacing the previous rootless network driver, Slirp4netns, with Bypass4netns.
While this requires a more recent Linux kernel to be available on the host system, in our test we achieved throughput with 95% of the root network of the nodes.
The other change we made was to integrate Liqo, a tool for dynamically setting up multi-cluster Kubernetes environments.
So when deploying a multi-node Kubernetes cluster with KSI 2, it actually deploys a single node cluster on every node and then employs Liqo to aggregate them into a combined cluster.

The full details of our evaluation and the design can be found in the paper, which is available here: https://www.iariajournals.org/intelligent_systems/tocv18n34.html