K12LTSP generally involves 1 high end workstation or server with lots of memory, cpu, and storage. Then thin clients are booted via the private network (pxe or etherboot) off of the K12LTSP server. Thick clients may also be used, such as Pentium 120 with 32MB of memory. The clients load the kernel and a section of X11 locally, everything else is run on the K12LTSP server so the clients don't need a fast processor nor a lot of memory.
OpenSSI Cluster is one or more primary nodes directly attached to a shared storage and optionally one or more other nodes not directly attached to shared storage. Each node is booted via the private network using PXE or etherboot. The clients locally load the kernel and join the cluster, providing its cpu, memory, and optionally other components such as network and storage. The root filesystem is provided by the primary node and each member node access that as their own root filesystem. If the primary node no longer runs then another primary node will take over and provide the root filesystem. If there are no other primary nodes to serve up the root filesystem then the cluster dies.
Lustre comes from the words "Linux Cluster". Lustre provides shared storage access via the network.
Idea:
My idea involves combining K12LTSP with OpenSSI and with a shared storage via TCP/IP.
Details of idea:
Install and setup a K12LTSP server with thin and thick clients.
Method 1:
Have a 2nd server that will assist the primary K12LTSP server in its workload and availability. This will focus mostly on high availability of a K12LTSP server, but will also benefit it by sharing in the workload.
This would involve in having either a directly attached shared storage or getting a network based shared storage solution to work.
Method 2:
Allow the thick clients (and disallow the thin and fanless clients) to provide their own cpu power and memory to the OpenSSI-K12LTSP cluster. This way the thick clients will help out in the overall classroom workload.
Optionally add on a TCP/IP based shared storage solution to work on these thick clients so they would provide failover as needed.
Method 3:
I would like to see an OpenSSI cluster with a TCP/IP based shared storage solution, so that every node is directly attached to the shared storage (via the network) making them a primary node type. This would truely be a robust OpenSSI solution.
Ideas for the network based shared storage are
1. Lustre - http://www.lustre.org/
2. drdb - http://www.drbd.org/
3. customize OpenSSI root filesystem so that we provide the necessary pieces through scripts?
4. Sistina's Global File System (GFS)
Search this Site!: