Troubleshooting VMware LSI Logic SAS bottlenecks
Troubleshooting VMware VM LSI Logic SAS controller performance bottlenecks starts with separating a virtual SCSI ceiling from a slow disk, congested host or saturated storage array. The lsilogic-sas controller normally allows 64 outstanding commands per virtual controller, which many workloads never reach but busy consolidation, backup and migration jobs can.
Capture latency, IOPS, throughput, queue length and datastore performance before changing anything. This matters in Australia, where hybrid estates often combine on-premises VMware with cloud replication and workloads hosted in the Sydney and Melbourne AWS regions. For background on the authors and the practitioner focus behind these checks, IT Diversified explains how field experience shapes its technical material.
Recognise the real symptom
First classify the complaint. High latency with low IOPS suggests IO starvation or a backend problem; saturated queues with otherwise healthy storage point towards host or virtual-controller limits. Compare several disks and both clusters, and correlate spikes with vMotion, snapshot consolidation, backup, RAID rebuilds or datastore maintenance rather than assuming every delay comes from the LSI adapter.
Useful baselines include average and 95th/99th-percentile latency, read/write latency, commands per second, bandwidth, queue length and CPU ready time. Averages can hide short stalls that damage database or virtual-desktop response times, so retain peak data from the same interval. In Sydney or Melbourne datacentres, also check whether unrelated tenants share the same physical SAN or power circuit.
Confirm controller and path mapping
Confirm what the guest actually sees. In vSphere, inspect the VM’s virtual SCSI controller and disk mapping; in ESXi, use esxcli storage core device list, esxcli storage core adapter list and esxcli storage core path list. Record the display name, backing device, queue limits, path state and active path count before attaching traces or changing the controller type.
Distinguish lsilogic-sas from lsilogic-parallel, pvscsi and virtual NVMe. The guest’s LSI Logic driver usually talks to a VMware virtual controller, not directly to a physical LSI SAS HBA, so replacing an HBA may not change the VM bottleneck. Check ESXi build notes, virtual-hardware compatibility and HBA firmware only where the SAN path actually uses them.
Measure queue depth and latency
Use esxtop and press D for disk counters while the issue is active. Depending on the ESXi release, the view exposes latency, commands per second, read/write latency, queue length and throughput. Compare the selected virtual disk with the underlying datastore device; if both worsen together, the controller is unlikely to be the sole cause.
In the guest, collect iostat -x 1 on Linux or equivalent disk counters in Windows. A consistently full VMware queue with modest backend latency, rising guest wait and no datastore-wide issue suggests virtual-controller saturation. Conversely, high array latency, full active paths or IOPS limits that appear without guest queueing indicates the backend is the constraint.
Trace contention to the storage backend
Inspect the SAN or disk-array telemetry for service time, IOPS, bandwidth, cache hit rate, queue depth and rebuild or scrub state. Thin-provisioned datastores can also suffer when free space runs low because allocation and metadata operations compete with foreground IO. Confirm that multipathing is genuinely active and that APD or NMP is selecting the intended paths rather than merely presenting extra routes.
Account for Australian geography and operating patterns. A Perth or Brisbane branch restoring a large backup over a congested WAN can make datastore latency appear to be a controller fault, while scheduled afternoon maintenance may overlap with an arvo change window. Record WAN throughput, retransmits and traffic priority alongside storage metrics so remote traffic is not misdiagnosed.
Check host and guest interference
Use esxtop CPU and memory views with disk counters. Saturated host CPU, memory pressure, ballooning or swap can increase application wait and storage queueing even when the array is healthy. Check whether Storage vMotion, vMotion, host consolidation, antivirus scanning or backup agents are generating the load, and compare idle hosts with the affected cluster.
Verify VMware Tools, guest drivers, paravirtual components and time synchronisation. Check VMkernel, vSwitch and physical-network MTU where NFS or IP-based storage participates; for Fibre Channel SANs, inspect fabric congestion, CRC errors and link saturation instead. If Network Port Security or storage policies are active, capture a short trace and review drops rather than inferring a controller fault from a single high-latency sample.
Apply targeted remediation
Remediation should match the evidence. If the virtual queue is full while backend performance has headroom, test pvscsi on a non-critical clone with supported VMware Tools and operating-system drivers. Changing a controller usually requires a planned power cycle and proper disk migration, and it will not solve a saturated array or faulty path.
If the backend is constrained, reduce competing work, move hot data to capable SSD or NVMe storage, restore healthy RAID layouts, spread load across hosts and resolve path imbalance. Revisit queue-depth settings only within vendor limits, and validate changes against workload latency rather than IOPS alone. For regulated Australian organisations, keep data-residency and sovereign-hosting constraints in scope when shifting workloads.
Compare controllers and confirm recovery
These controller choices differ materially in scalability and compatibility:
| VMware controller | Main strength | Main limitation | Best fit |
|---|---|---|---|
LSI Logic SAS (lsilogic-sas) |
Broad guest compatibility and a 64-command virtual queue | Can reach its queue ceiling under heavy concurrency | General-purpose VMs with moderate IOPS |
LSI Logic Parallel (lsilogic-parallel) |
Legacy compatibility | Older emulated path with lower scalability | Older operating systems or migrations |
VMware Paravirtual SCSI (pvscsi) |
Large command queue and strong throughput | Requires compatible tools, drivers and planned conversion | Databases, file servers and virtual desktops |
Validate the remedy with the same workload and metrics used for the baseline. The LSI Logic SAS controller is the problem only when its queue remains saturated independently of backend, host and network contention. Recheck latency percentiles, active paths, datastore capacity and guest wait time after each controlled change, then document the final configuration for the next arvo handover.