You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
OpenCHAMI is already a fairly small stack that can handle a constrained failure domain with relatively low resource usage. Through optimizations, the lowest level hardware management functions can be made even less resource intensive. Modern management switches tend to have modern CPUs and memory, but take little advantage of them in steady state operation because the network ASIC handles most of the packet switching and routing functions. Modern switch management software, like the firmware based on SONiC from the Open Compute Platform, supports using the CPU and Memory of the switch to run containers and provides tooling to allow containers to communicate with each other and across the network from switch to switch.
Based on our initial research, it is likely that we could deploy OpenCHAMI, as it exists today, as a set of containers that run as part of the Top-of-Rack(ToR) switches, removing the need for head nodes in small clusters. In addition, a collection of federated failure domains, as defined by the blast radius of a management switch can be orchestrated externally through a federation system that doesn’t need to run close to the hardware.
The text was updated successfully, but these errors were encountered:
OpenCHAMI is already a fairly small stack that can handle a constrained failure domain with relatively low resource usage. Through optimizations, the lowest level hardware management functions can be made even less resource intensive. Modern management switches tend to have modern CPUs and memory, but take little advantage of them in steady state operation because the network ASIC handles most of the packet switching and routing functions. Modern switch management software, like the firmware based on SONiC from the Open Compute Platform, supports using the CPU and Memory of the switch to run containers and provides tooling to allow containers to communicate with each other and across the network from switch to switch.
Based on our initial research, it is likely that we could deploy OpenCHAMI, as it exists today, as a set of containers that run as part of the Top-of-Rack(ToR) switches, removing the need for head nodes in small clusters. In addition, a collection of federated failure domains, as defined by the blast radius of a management switch can be orchestrated externally through a federation system that doesn’t need to run close to the hardware.
The text was updated successfully, but these errors were encountered: