In enterprise Fibre Channel and iSCSI SAN deployments, connecting redundant HBAs to dual fabric switches does not guarantee high throughput. By default, VMware ESXi may send thousands of sequential I/O commands over a single physical cable before switching to the alternative path.
This article explores the Pluggable Storage Architecture (PSA), the difference between Storage Array Type Plugins (SATP) and Path Selection Plugins (PSP), and how tuning Round Robin limits transforms all-flash SAN performance.
1. PSA Architecture: Native Multipathing Plugin (NMP)
When ESXi discovers physical LUN paths, the Native Multipathing Plugin coordinates two modular components:
- SATP (Storage Array Type Plugin): Handles array-specific hardware failure detection,
active-active ALUA path state transitions, and failover/failback commands (e.g.
VMW_SATP_ALUA). - PSP (Path Selection Plugin): Dictates which active physical path routes the next SCSI
read/write request (e.g.
VMW_PSP_RR,VMW_PSP_MRU,VMW_PSP_FIXED).
2. Understanding SATP Claim Rules
Claim rules instruct ESXi which Storage Array Type Plugin to associate with storage targets upon bus
rescanning. You can create vendor-specific claim rules that bind arrays directly to VMW_PSP_RR.
3. Why the Default 1000 IOPS Limit Fails Modern Storage
When a LUN is assigned the Round Robin policy (VMW_PSP_RR), ESXi defaults to sending
1,000 I/O operations down a single path before switching to the next available path.
On modern all-flash arrays (Pure Storage, NetApp AFF, Dell PowerStore) where latencies are under 200 microseconds, sending 1,000 IOPS down one port causes queue contention on one HBA while the secondary fabric sits completely idle.
4. Tuning to iops=1 for True Dynamic Load Distribution
By lowering the Round Robin threshold from 1,000 to 1, ESXi alternates physical HBA paths on every single I/O request. This effectively doubles or quadruples SAN throughput and cuts tail latencies in half:
iops=1 as their recommended host configuration.
Create a custom SATP claim rule so new LUNs automatically inherit this policy upon discovery.
5. Cluster-Wide PowerCLI Automation
Instead of manually running esxcli per host and LUN, you can apply the Round Robin policy and
1-IOPS limit across an entire cluster using PowerCLI: