A managed instance group must share a large read-only dataset on each VM's filesystem with fast startup, low-latency access, and minimal cost. Which approach do you use?

Attaching a Compute Engine persistent disk in read-only mode to multiple instances is the most efficient solution. This provides low-latency access directly from the VM's filesystem, fast startup as the data is immediately available, and minimal cost since only one disk is provisioned and shared. Cloud Storage FUSE can have higher latency and startup times due to network access and FUSE overhead. Copying data from Cloud Storage via a startup script is slow and increases instance startup time, which is inefficient for large datasets. Creating multiple disks from a snapshot is costly as each VM gets its own dedicated disk, and it doesn't meet the "minimal cost" requirement.