NVMe-TCP + VMware Design Questions
Q1: In a medium - large datacenter (ie anticipated growth to 500 hosts), would it be a problem to use large subnets (ie 2 x /23 or /22's) to place the initiators + target IPs? All our existing nvme-tcp deployments (which we're having great success with btw) use 2 x /24's, which provides more than enough capacity for the array + a decent sized cluster. But now we're entering into the realms of multiple flasharrays (ie 3 or 4 in a single site) and hundreds of hosts - we'll quickly exhaust a /24 range. Q2: In this site, where we have say 2 x FA-X's an 2 x FA-C's, is it valid to present datastores from all of these arrays over nvme-tcp to a single cluster? With each array / datastore cluster being accessed using the same initiators on the same layer 2's? At what point does it get silly and I need to draw a line and segregate environments? Q3: Is there a capacity limit to the number of client hosts I can have on an array / mapping storage to? (nvm - I think I found the answer to that - looks like it's 1000 nqn's on pretty much all X and C arrays)NVMe-over-Fabrics (NVMe-oF) with Windows Server Initiator and Everpure FlashArray
Are you actively using or considering to use NVMe-oF with Linux and FlashArray? Do you know that Microsoft recently announced a preview of Windows Server NVMe-oF initiator? For those who are interested in this topic, I tested the initial preview with FlashArray and posted the results into the Everpure blog page here
