The virtual machine experiences intermittent "hard resets" and performance degradation. Analysis of vmware.log and hostd logs reveals the following indicators:
Error Message: vmkernel.log: PVSCSI: 2682: Failed to issue sync i/o : Busy
Error Message: Hostd.log:No VMkernel vSCSI handler for disk scsi0:X (ctrl:1000, path:...........
Guest Behavior: VM vmware.log: The Guest OS SCSI timeout is reached, leading to a "hard reset" request to the virtual hardware.
ESXi 8.x
The virtual machine's performance is bottlenecked because all high-intensity data disks are routed through a single PVSCSI Controller (SCSI 0). This configuration forces the database and Operating System's concurrent I/O requests to share a limited virtual execution path and queue depth, quickly exhausting the controller's capacity. The resulting synchronous I/O failures are misinterpreted by the Guest OS as a hardware hang, which initiates a system reset for recovery.
Optimizing I/O Performance by Distributing Load with Multiple PVSCSI Controllers
The recommended strategy to eliminate I/O bottlenecks and significantly improve throughput is to distribute the I/O load across several virtual SCSI controllers. This method achieves better performance by creating independent queues and enabling parallel interrupt processing.
1. Add Virtual Controllers
2. Redistribute Virtual Disks
3. Operating System Verification