On Windows Server, you create multiple disks over MSISCSI against the storage array, then do the stressing test for all of the disks. When the test loads fully saturates the I/O storage path (on either network or physical storage end), Windows Server will start to queue the I/O requests in to a pending queue at storport level. If you shut down one of the physical disk on storage array, it could take more than 10 minutes for Windows to report the disk remove event.
here are no guarantees on how fast Windows will handle a device removal event. When a device removal occurs, MSISCSI driver detects this event and sends a BusChangeDetected notification to Storport. Storport then alerts PnP that there are device changes and PnP needs to re-enumerate the devices. PnP sends a request to Storport to re-enumerate its devices. To determine the current devices, Storport issues SCSI commands to each of the supported Buses/Targets/Luns on MSISCSI adapter. Each MSISCSI adapter has a maximum outstanding IO limit and if the number of IOs requested is larger than the maximum outstanding IO limit of the adapter, the IOs will be queued. If the outstanding I/Os against the active disks is at the adapter outstanding IO limit, there will be some latency for storport to issue the request to MSISCSI which causes the delay of reporting disk removal event. This is a by design behavior.
Above behavior could also occur to FC HBA adapter that is managed by storport.
Article ID: 3073577 - Last Review: 17 Jun 2015 - Revision: 1