Storage Performance Tuning Guide – Best Practice
This article provides technical guidance for setting up iSCSI and Fibre Channel (FC) connections on QSAN XCubeFAS, XCubeSAN, XCubeNXT, and XCubeNAS in Windows, Unix-like, and VMware environments to achieve better performance.
Download: Performance Tuning Guide
Click here to download the Storage Performance Tuning Guide – Best Practice.
Key Sections:
- Prerequisites: Storage topology example (XF2026 / XS5226 / XS3226 / XS1226), hosts, management, iSCSI, FC.
- Guidelines for Hosts: Keep HBA/NIC/switch firmware and drivers updated; ensure the host local drive is not a bottleneck for copy tests.
- Guidelines for Configuring Storage Pools:
- Prefer Thick provisioning; enable Disk Write Cache, Read-ahead, Command Queuing.
- Enable Write-back cache on volumes; optional Video Editing Mode for steadier throughput.
- Avoid snapshots during performance tests; consider multiple pools split across controllers when backend exceeds front-end ports.
- Default LUN masking “*” if masking not required.
- VMware: do not use 4K block size.
- Guidelines for Configuring iSCSI Connections:
- In Storage: Configure iSCSI data ports (XEVO/SANOS/QSM); verify link speed; ensure enough host NICs to match controller bandwidth.
- Optional: Jumbo Frame (consistent MTU across host/switch/storage); VLAN as needed; iSCSI Entity Name.
- Trunking/LACP only for large multi-client topologies; otherwise MPIO is sufficient.
- Ethernet Switch: Jumbo Frame, Flow Control (ON/OFF depends on environment), LACP if used; port mirroring + Wireshark for troubleshooting.
- Windows: Use all NICs; specify source IP per session; NIC tuning: set RSS Queue ~2, max Receive/Transmit Buffers, Interrupt Moderation OFF;
netsh int tcp set global autotuninglevel=restricted(orhighlyrestricted). - Unix-like: Use different subnets per NIC; increase RAID read-ahead (e.g., 4096/8192); raise TCP receive buffer (e.g., 524,284+); consider disabling Hyper-Threading.
- VMware: Use different subnets per NIC; avoid 4K block size; MPIO policy Round Robin with IOPS = 1; Delayed ACK notes and ATS heartbeat guidance per VMware KB.
- Guidelines for Configure FC Connections:
- In Storage: FC data ports; default topology P2P; 16Gb FC supports P2P only (Loop requires 8G/4G and storage restart); link speed Auto.
- FC Switch: Auto topology/speed; configure ZONE if needed.
- Windows: Update FC HBA; for Marvell QLogic, optional registry
DriverParameter=qd=255. - Unix-like: Update FC HBA; in
multipath.confsetrr_min_io=1; increase read-ahead; consider disabling Hyper-Threading. - VMware: Same VMware constraints (no 4K; RR IOPS 1).
- Test Results and Use Cases:
- Random Read 4K: ~500K IOPS <1 ms (max ~784K).
- Random Write 4K: ~280K IOPS <1 ms (max ~384K).
- Use cases: High-IOPS/low-latency (VDI, DB), onboard 10GbE usefulness, extreme throughput (>12,000 MB/s possible with combined ports).
- Apply To: XEVO firmware 2.0.3+; SANOS firmware 2.0.1a+; QSM firmware 3.3.0+.
- References: XEVO/SANOS/QSM Software Manuals; related white papers; video tutorial playlists.
Critical Notes:
- CAUTION: Do not set 4K block size for VMware ESXi datastores.
- TIP: With VMware MPIO, use Round Robin and set IOPS to 1 (instead of 1000).
- INFORMATION: 16Gb FC supports P2P topology only; Loop requires 8G/4G and a storage restart.
- TIP: Avoid Trunking/LACP unless serving many clients; standard MPIO is typically sufficient.
JetStor Support
For assistance with performance tuning on JetStor deployments:
๐ง [email protected]
๐ซ Submit Support Ticket