This  issue means that one or more vSAN backend capacity disks do not see any IOs. This issue is delivered only for all-flash vSAN clusters and for benchmarks that have some read IO activity. While some capacity disks may not see read IOs for some intervals of time, the best performance is usually achieved when read IOs are spread evenly across all backend capacity disks.

Here is a list of possible remedies:

1: If you are running a 100% read workload, such as 100% sequential read or 100% random read, then it is possible that the contents of the capacity disk are cached in the write buffer of the cache tier. In this case, the reads are serviced from the cache tier and do not hit the capacity tier. Depending on the specifications of your cache tier and capacity tier storage, and the number of capacity tier disks in your disk group, it may be possible to get better read bandwidth if some read content is in the capacity tier. In this case, please increase the size of the virtual machine disks (VMDKs) to have more than a 400GB working set per disk group. In general, for an all-flash vSAN cluster, the performance of a 100% read workload is better with large working sets.

2: If your benchmark has some write IO component, you may increase the number of VMDKs so that all backend capacity disks have some object components. We recommend that any benchmark should create one VMDK for each capacity disk on the system, and concurrently drive IO to each of these VMDKs. A safe rule of thumb is to create two VMs for every disk group on your system, and create eight VMDKs for each virtual machine. Size VMDKs according to the available size of cache tier across all disk groups

3: Alternatively, it is possible that your benchmark is not issuing IOs to all the VMDKs that were created. Please check if that is the case, and correct it so that IOs are targeted to all VMDKs.

4:If you do not want to increase the number of virtual machines or the number of VMDKs, you may increase the “Number of disk stripes per object” (the default value is 1), in the vSAN storage policy with which the VMDKs were created. This number refers to the number of capacity disks across which each replica of a storage object is striped. You may choose to apply the policy to existing virtual machines, in which case all existing VMDKs will be reconfigured, and you must wait for the reconfiguration traffic to end. Alternatively, you can apply the policy manually to existing or new virtual machines/VMDKs.