Around 60% of our virtual machines currently run on older B2ms or B2s sizes with temporary disks, and those sizes are no longer available for reservations. We don't need high performance and are primarily trying to keep costs under control, so I've started moving them to B2s_v2 and comparable sizes.
These newer sizes don't support NVMe, which is fine because most of the machines are lightly used web front ends. My concern is the longer-term impact: could choosing SCSI-based VM sizes force us to delete and recreate the machines in a year or two if newer Azure features or VM families require NVMe? Would there be a better low-cost option that avoids another migration later?
2 Answers
I wouldn't choose a more expensive or complicated design solely to prepare for a hypothetical NVMe requirement. For ordinary IIS or Nginx servers that mostly serve requests and connect to a separate backend database, B-series v2 should be fine. Monitor CPU credits, memory, disk latency, and network usage after the move; those measurements will tell you more than the storage interface about whether the size is appropriate.
For lightly loaded web front ends, moving to B-series v2 sounds reasonable. NVMe isn't necessary for that workload, and using SCSI-based storage shouldn't create a near-term problem. Azure is very unlikely to drop SCSI support while customers still depend on it. You'd mainly need to reconsider if the workload becomes storage-intensive or if you later need a VM family with NVMe-specific performance features.
That matches the workload pretty well. My main concern was avoiding another rebuild if SCSI support eventually disappears. Since the only reason for moving is that the older B-series sizes can no longer be reserved, I'm leaning toward making the smaller change now and accepting a modest cost increase.