Hi All,
I think Your answer is not really clear - same as Azure documentation is not clear in this topic. I spent all day on testing VM data (persistent) disks performance and comparing it with the documentation.
Facts:
1) There is a transfer performance limit ~60MB/s - in my opinion 62.5MB/s (will answer later why) but it is not a very hard limit - sometimes it can be a bit bigger (like 70MB/s) but it can depend on way we measure, so let say 62.5MB/s.
2) There is IOPS limit, globally for all standard blobs (including VM data disks): 500IOPS
3) In premium storage with 256kB I/O unit everything seems to be mathematically clear (500x 256kB = 125MB/s, with throttling we have 100MB/s per P10 disk). So limits are clear - 500IOPS or 100MB/s and so on with P20 and P30.
4) I/O unit for standard storage is not clear. With 8kB unit and 500IOPS limit we have 500x 8kB = 3.9MB/s. Transfer limit is ~60MB/s so mathematically (in my opinion - I can be wrong) it should be not possible to reach 60MB/s but... there is no problem with
it. Disks are running with no problem at ~60MB/s and in RAID0 configuration we have transfers equal to "performance = ~60MB/s x # of disks" so in RAID0 with 8 disks I have ~480MB/s.
From those facts we can calculate, that IOPS unit for standard storage is equal to 128kB. 500x 128kB = 62.5MB/s.
Can You clarify the mathematics of 8kB I/O unit size and prove that I'm wrong or check if maybe I'm right and change the documentation? It is very confusing in example during RAID configuration and optimization.
Best regards,
Michal