Home

Rack OS is a CentOS 6.x based OS running on a whole rack.

Rack level OS will treat the entire RACK as a single PC even this rack has storage-nodes and compute-nodes built-in the rack. Maybe there will be a rack manager who will connect to a smart PDU (Power Distribute) or serial ports connected to BMC like management port. This OS will treat no rack manager, all servers IPMI remote controller was connected to a switch and a module will run on boot node will manage them.

Boot Node: Boot Node is the first booted node in the rack, all kernels and file system will uploaded to boot node's ramdisk, suggest that boot node will have at least 16GB ram. Boot node will have a KVM VM who will provide the PXE service to all others nodes of whole rack. This VM use 8GB ram as ramdisk and 2 GB ram as storage.

Backup boot node: When 1st node booted and detects another node booting. Boot node VM will put a script to 2nd booted nodes and this node will have a almost same VM with PXE service VM. Once 2nd PXE VM can not get the heart beat from primary VM, 2nd VM will take the job.

Working nodes: All others node was named as working nodes, they will boot from PXE and update drivers or update kernels via PXE reboot. No OS HDDs present. Once a working nodes join to RACK OS, he will open a NFS shared folder, the init script will monitor the NFS folder to see what the boot node wants it to do.

Stage 1: 2014.6.1 – 2015.11.15 Alpha phase will finish a PXE environment which can boot all nodes to OS and a how-to will be available on SourceForge Rack OS project:
https://sourceforge.net/projects/rack-os/files/Document/howto/

Stage 2: 2015.12.1 – 2016.12.1 Beta phase, will try to have a script which will load jobs to working nodes and try to management all nodes as 1 PC. This phase the OS will be Shell script based regarding my cool-bee develop skills is not enough to provide a really OS level stuff.

Stage 3: If someone donated or some really new-bee level developers joined, let’s try to switch this OS to a really OS. This is my dream.

The design target is that all nodes are not always available and all hardware such as CPU/Memory/HDD will be managed as a single PC. The client application can run on all racks and treat it as a single PC with enough memory and enough compute cores.