Fault-tolerant MPI
Was rather a day without any productivity except installing Debain stable/mono/mpi on a VM, registering with openPBS.
All this while I had been thinking that PBS is a scheduler that comes with Globus. In fact, it has been there since 90s and very popular in cluster computing paradigm. And teh next thing I learn is that its free only for upto 32 nodes.
Secondly, read this paper on Fault-tolerant MPI from Frank Copello's group. I have a feeling they are doing excellent research in distributed computing. Applying distributed computing concepts (failure-masking, checkpointing) to real applications.
Indeed, if large scale and parallel distributed (LSPD) computing were to materialize, we would run into a situation with volatile and untrusted nodes. This is not so big an issue for file-sharing systems, since there is no accountability.
Not ofcourse when you use the desktop nodes to run some real scinetific computation.
Can't ignore failues, node-compromises, security issues, secure computation.
Thats it...
All this while I had been thinking that PBS is a scheduler that comes with Globus. In fact, it has been there since 90s and very popular in cluster computing paradigm. And teh next thing I learn is that its free only for upto 32 nodes.
Secondly, read this paper on Fault-tolerant MPI from Frank Copello's group. I have a feeling they are doing excellent research in distributed computing. Applying distributed computing concepts (failure-masking, checkpointing) to real applications.
Indeed, if large scale and parallel distributed (LSPD) computing were to materialize, we would run into a situation with volatile and untrusted nodes. This is not so big an issue for file-sharing systems, since there is no accountability.
Not ofcourse when you use the desktop nodes to run some real scinetific computation.
Can't ignore failues, node-compromises, security issues, secure computation.
Thats it...
posted by Hypervisor at 9:14 PM
0 comments
