/mcp© 2026 Cisco and/or its affiliates. All rights reserved. Page 21 of 34 o "KAVG/cmd" (VM Kernel Average Latency per command) - an indicator of CPU resources/performance. Expected to be 0ms in an ideal environment. Values greater than 2ms may cause performance problems. o "DAVG/cmd" (Device Average Latency per command) - an indicator of disk subsystem performance. Values consistently greater than 20-30ms are often indicative of performance problems in typical applications. o Latency values over time are expected to fluctuate. Ephemeral spikes above the application's indicated maximum with tiny duration and tiny frequency are expected, but spikes above the application's indicated maximum that are frequent, sustained, and/or high-magnitude indicate a potential problem and are likely contributors to application symptoms. o See VMware by Broadcom documentation for monitoring, performance, and esxtop for more details on viewing and interpreting latency values. ● Usable space: The storage system must provide usable space in GB at least equal to the sum of all virtual machines' vdisks. Non-appliance hardware may also use thin provisioning, with the caveat that disk space must be available to the VM as needed; running out of disk space due to thin provisioning will cause application instability and data corruption, including preventing restore from backup. ● IOPS capacity: Some storage system design documentation will request information on the application workloads' I/O operations, including read/write, sequential, and random. For most applications, this varies by deployment model and capacity; refer to each application's Installation Guide for details. ● Best Practices by Storage Option o Local DAS: o Make sure the RAID controller and all disks are listed by Broadcom as compatible for the ESXi release being used, and listed as supported by the hardware vendor. o See hardware providers for guidance on meeting application latency and IOPS requirements. Below are guidelines only; your results may vary. RAID1 o HDD pairs are not recommended unless the environment is very small (fewer than 100 devices or fewer). o Consider SSD pairs with sufficient usable space. High- endurance models are recommended, as many Cisco Collaboration applications are write-intensive. RAID5 o For most Collaboration deployments, RAID5 has been the best tradeoff among competing factors of maximizing usable space, IO performance, fault tolerance, and ease of failed disk replacement while minimizing total storage system price and complexity. o One HDD per physical CPU core, with 4-6 disks per RAID5 array (more disks per volume are discouraged as it increases the risk of long rebuild times if there is ever a multiple disk failure).