Perimeta SBC Gateway software upgrade fails with SCSI device layer aborts and high DAVG latency

Products

• VMware vSphere ESXi

Issue/Introduction

Symptoms:

• Virtual machines hosting Perimeta SBC gateway software experience severe disk read/write speed inconsistencies.
  
  
• When attempting to upgrade the Perimeta software, the process either takes an excessive amount of time or fails entirely.
  
  
• The ESXi /var/log/vmkernel.log reports SCSI Device I/O aborts and cancellations, indicating the device layer is dropping commands: 

2026-03-02T04:30:37.270Z In(182) vmkernel: cpu24:2097271)ScsiDeviceIO: 4605: Cmd(0x45b985ed5800) 0x2a, cmdId.initiator=0x430d40908ac0 CmdSN 0xbf from world 2099840 to dev "naa.6#####################4" failed H:0x8 D:0x0 P:0x0 Cancelled from device layer
2026-03-02T04:30:37.354Z In(182) vmkernel: cpu10:2098207)ScsiDeviceIO: 4633: Cmd(0x45b98722f000) 0x2a, CmdSN 0xea from world 2099840 to dev "naa.6#####################4" failed H:0x5 D:0x0 P:0x0

• Performance metrics show DAVG spiking more than 50ms or higher, significant I/O queuing with write throughput of ~4.31MB/s and less.  
  
  

Environment

VMware ESXi 8.x
VMware ESX 9.x

Cause

• The storage infrastructure backing the affected datastore consists of mechanical SAS HDDs. While these disks can theoretically achieve up to 200MB/s for purely sequential workloads, the specific I/O volume and profile generated by the Perimeta software upgrade overwhelm the mechanical spindles.
  
  
• This hardware contention causes deep I/O queuing, driving device latency (DAVG) well above acceptable thresholds (>75ms). When latency remains this high, the ESXi storage stack or the underlying storage controller begins aborting the starved I/O requests (H:0x8 aborts and H:0x5 path failures).

naa.6#####################4:
   Display Name: Local Cisco Disk (naa.6#####################4)
   Vendor: Cisco
   Model: UCSC-RAID12G-2GB
   Is Local: true
   Is Removable: false
   Is SSD: false
   Is SAS: true
   Device Max Queue Depth: 128
   No of outstanding IOs with competing worlds: 32
   Drive Type: logical
   RAID Level: RAID1
   Number of Physical Drives: 2

Resolution

To resolve the I/O aborts and meet the application throughput requirements, the physical storage hardware must be aligned with the workload's IOPS and latency profile.

• Provision a datastore backed by Solid State Drives (SSDs) or NVMe storage capable of handling random I/O and sustaining 500MB/s or higher sequential throughput without queuing.
  
  
• Perform a Storage vMotion to migrate the affected virtual machines from the mechanical HDD-backed datastore to the newly provisioned high-performance datastore.

Additional Information

• Hardware vendor analysis confirms that the application I/O profile exceeds the physical capabilities of the currently deployed SAS HDDs.