CXL: A Technology of Today That Gets Better Tomorrow #
CXL has been touted as coming soon or just around the corner for years, it’s time to tout it as having arrived.
We have some exciting announcements that we will fully reveal at the OCP Global Summit in San Jose this year starting October 15th.
A lot of CXL development has been focused in the very technical part of “getting it to work”, but where CXL and the overall ecosystem have largely fallen short is “making CXL usable, today”. That’s where our focus has been working closely with many member companies at OCP and in the broader CXL ecosystem.
Our focus has paid off, and there is now a way to use CXL with Kubernetes with no code changes and no special flavors of the linux kernel or the Kubernetes distribution you use. Jackrabbit Labs will introduce the most advanced way to schedule pods that allows systems to leverage memory from a far memory pool dynamically to provision pods and even clean up after when those pods are shut down or re-scheduled.
What does this mean for you as a user of Kubernetes?
You can now have less disruption to pod scheduling due to only violating the memory constraints on a single worker node. Pods can be scheduled on worker nodes that have exhausted memory without having to reschedule pods to free up local memory on any number of worker nodes in a Kubernetes cluster.
You can schedule many more jobs to fully utilize the capacity of the worker nodes with all of the CPU cycles available without having to worry about exhausting DRAM that is attached.
You can schedule low priority, non performance sensitive applications to leverage far memory regardless of worker node freeing up extremely valuable locally attached DRAM for applications that demand the maximum level of performance.
It means by simply using Kubernetes you can unlock the latent potential of disaggregated infrastructure in it’s entirety – CPU, Disk, Network, and now for the first time Memory.
Come see our demonstration in the CMS section of the OCP Innovation Village
