Thanks for the communication on the forum. You information has been very helpful!
This is definitely an interesting project I’m working on. The biggest challenge that we have now is getting a parallel file system that can run as fast as we need. The underlying hardware and the application can run at speed, but now we have to share the data between the nodes for the “Read only” node to do the analysis on the written data. Since both nodes can see all of the NVMe SSDs, a parallel file system would be the fastest way for the analysis node to read the data.
We’ve even played with the idea of just using XFS and mounting one of the nodes as “ReadOnly” for the analysis node to get at the data. When the write node gets done with a file, that triggers the ReadOnly node to unmount and re-mount the volume to pick up the changes. Not the most elegant solution, but it basically works.
I’d be happy to give you a summary of the hardware if you are interested.