Account for cache misses against both Azure VM storage limits

Source facts

Azure VMs with premium storage and host caching have separate cached and uncached storage limits. A read satisfied by host cache counts against the cached limit. A cache miss must reach the disk and counts against both cached and uncached VM limits. Attached disks also have their own IOPS and throughput limits. Enabling caching therefore does not make every read independent of the remote disk path. Microsoft Learn.

Applicability

Use this distinction when a read-heavy workload has host caching enabled but still appears constrained by uncached storage. Identify the VM size, disk configuration, actual cache mode, and whether the observed workload is reading data already present in cache.

DSE recommendation

Map the measured read path before deciding which storage limit needs attention. Ask the performance owner to distinguish cache-hit and cache-miss behavior and compare VM-level evidence with the affected disks. Keep observations from an already warmed workload separate from first-read behavior. Do not assume a higher disk tier or a cache setting alone resolves an unexplained limit.

Verification

Use a representative nonproduction workload to compare the relevant VM and disk metrics under controlled read conditions. Record the workload shape, cache configuration, measured latency, and which limit approaches saturation. Change one relevant condition at a time and compare results with the proposed explanation. If the observed bottleneck moves between cached and uncached paths, preserve that finding rather than combining the results into one unexplained throughput number.

Official references

Microsoft Learn: Virtual machine and disk performance. Source reviewed September 9, 2026.