What about storage? Each server must have its own HD? Or is it possible to have a large disk at the head of the cluster and share it with the rest of the nodes? Which is the best practice? I have in mind to run Ray in a local network, not in the cloud.
Thanks in advanced.
Hi! I’m using a single network storage (NFS) which is mounted on all nodes and it works well. For me this is the cleanest solution, since I do not solely work with ray. Of course, whether the performance is sufficient for you depends on how much you write on the drive and how fast your network is.
I can only recommend this, since the solution is quite general and not only limited to ray.
+1 to using NFS or some mounted filesystem. Note that Ray doesn’t really make much use of the filesystem unless your application needs it. If you do need to write to the file system a lot in your application though, you may need a better file system.