010-151 DCTECH cheat sheet

UTP categories: Cat5e supports 1 GbE to 100 m; Cat6 supports 1 GbE to 100 m and 10GBASE-T to 55 m; Cat6a supports 10GBASE-T to 100 m.

Fiber types: OM3/OM4 multimode suit 10G-100G short reach (OM4 10GBASE-SR to 400 m); OS2 single-mode carries 10G-400G over kilometers via 10GBASE-LR/ER optics.

Connector pairings: LC duplex for SFP/SFP+ optics, MPO/MTP for 40G/100G parallel multimode (QSFP), SC for legacy single-mode links; APC ferrules are angled and must not mate with UPC.

10GBASE-T SFP+ modules draw roughly 2.5-4 W per port versus under 1 W for 10GBASE-SR optics, so copper SFP+ is limited to short in-rack runs and low port counts.

Twinax DAC cables carry passive 10G/25G to about 5-7 m and active variants to about 10-15 m; they are the default in-rack server-to-switch interconnect.

Power: a 1U server at 450 W on 208 V single-phase draws about 2.2 A; a 30 A 208 V circuit derated to 80% delivers 24 A, or about 5.0 kVA.

Redundant PSU sizing: sum load, then size each feed so one feed carries the full load; N+1 adds one spare unit, 2N duplicates the whole chain.

Rack unit = 44.45 mm (1.75 in); a standard 42U rack is about 1.87 m of usable vertical space, and airflow follows the cold-aisle-front, hot-aisle-rear convention.

Cable management rules: bend radius at least 4x cable diameter for UTP and 10x for fiber; never exceed 100 m channel length for twisted-pair Ethernet.

ECC RDIMM vs LRDIMM: RDIMMs buffer the address/command bus; LRDIMMs buffer data too and allow higher ranks per channel, so LRDIMM wins when capacity per slot matters more than latency.

Memory population rule: populate channels evenly and match DIMMs per channel; unbalanced population drops from quad- to dual- or single-channel mode and cuts bandwidth.

NUMA: each CPU socket owns local memory; a VM or process spanning sockets pays remote-memory latency, so pin latency-sensitive workloads to one node.

RAID levels: RAID 0 stripes with no redundancy; RAID 1 mirrors; RAID 5 tolerates one drive; RAID 6 tolerates two; RAID 10 mirrors then stripes and rebuilds faster than parity RAID.

Usable capacity formulas: RAID 5 = (N-1) x smallest drive; RAID 6 = (N-2) x smallest drive; RAID 10 = N/2 x smallest drive.

SSD vs HDD: SSDs have no seek time and far higher random IOPS but finite program/erase cycles; DWPD x capacity x 365 gives the warranted terabytes written per year.

Storage protocols: SAS is dual-ported and used for internal drives; SATA is single-ported; NVMe runs over PCIe and bypasses the SAS/SATA controller queue limits.

Boot and firmware: UEFI with GPT supports disks over 2 TB and secure boot; legacy BIOS with MBR caps boot volumes at 2 TB.

Virtualization support: Intel VT-x/EPT and AMD-V/NPT must be enabled in BIOS, and SR-IOV or PCIe passthrough is required for direct device access by a VM.

VLANs segment a Layer 2 domain; a trunk port carries multiple VLANs with 802.1Q tags, and the native VLAN is sent untagged, so both ends must agree on it.

Spanning Tree: STP blocks loops; RSTP converges in seconds; use edge/portfast on server ports and BPDU guard to prevent rogue switches.

Link aggregation (LACP) bundles links into one logical channel; all member links must match speed, duplex, and VLAN configuration, and hashing is per flow, not per packet.

MTU: jumbo frames are 9000 bytes; every device in the path, including the vSwitch and VM, must match or large packets silently drop.

vSwitch types: a standard vSwitch is per-host; a distributed switch is managed centrally and keeps port groups consistent across hosts for vMotion compatibility.

NIC teaming policies: route based on originating virtual port ID is default; IP hash requires LACP on the physical switch; failover order determines active/standby uplinks.

Overlay networking: VXLAN encapsulates Layer 2 frames in UDP (port 4789) with a 24-bit VNI, giving about 16 million segments and requiring an IP underlay.

iSCSI and NFS storage traffic belongs on dedicated VLANs or separate physical uplinks; iSCSI needs end-to-end jumbo frame consistency and CHAP for authentication.

IP addressing: /24 gives 254 usable hosts, /25 gives 126, /26 gives 62; private ranges are 10.0.0.0/8, 172.16.0.0/12, and 192.168.0.0/16.

Troubleshooting order: physical layer first (cable, link LED, optic), then link speed/duplex, then VLAN/IP, then routing and firewall, then application.

Link LED dark on both ends points to cable or transceiver; LED lit but no traffic points to VLAN mismatch, duplex mismatch, or a down switchport.

Intermittent CRC or FCS errors on an interface usually indicate a bad cable, dirty or mismated optic, or duplex mismatch rather than a software fault.

Server will not power on: check both PSU LEDs, confirm both feeds are live, verify the power button and BMC/iLO/iDRAC access, then review SEL logs for a fault.

Memory errors: correctable ECC errors are logged and scrubbed; uncorrectable errors trigger a machine check; DIMM slot location in the SEL identifies the failing module.

Firmware and driver baseline: apply the vendor support matrix combination of BIOS, BMC, NIC firmware, and OS driver; mismatched versions cause link flaps and boot failures.

Change control: record device, change, rollback plan, and window; capture a configuration backup before any switch or storage change so rollback is possible.

Escalation data to collect: timestamps, affected hosts and ports, switch logs, BMC SEL, and recent changes; without this the vendor case stalls.

Decommissioning: wipe drives per policy, remove host from cluster and monitoring, release IPs and DNS records, and update the asset inventory and cabling records.

UTP categories: Cat5e supports 1 GbE to 100 m; Cat6 supports 1 GbE to 100 m and 10GBASE-T to 55 m; Cat6a supports 10GBASE-T to 100 m.

Fiber types: OM3/OM4 multimode suit 10G-100G short reach (OM4 10GBASE-SR to 400 m); OS2 single-mode carries 10G-400G over kilometers via 10GBASE-LR/ER optics.

Connector pairings: LC duplex for SFP/SFP+ optics, MPO/MTP for 40G/100G parallel multimode (QSFP), SC for legacy single-mode links; APC ferrules are angled and must not mate with UPC.

10GBASE-T SFP+ modules draw roughly 2.5-4 W per port versus under 1 W for 10GBASE-SR optics, so copper SFP+ is limited to short in-rack runs and low port counts.

Twinax DAC cables carry passive 10G/25G to about 5-7 m and active variants to about 10-15 m; they are the default in-rack server-to-switch interconnect.

Power: a 1U server at 450 W on 208 V single-phase draws about 2.2 A; a 30 A 208 V circuit derated to 80% delivers 24 A, or about 5.0 kVA.

Redundant PSU sizing: sum load, then size each feed so one feed carries the full load; N+1 adds one spare unit, 2N duplicates the whole chain.

Rack unit = 44.45 mm (1.75 in); a standard 42U rack is about 1.87 m of usable vertical space, and airflow follows the cold-aisle-front, hot-aisle-rear convention.

Cable management rules: bend radius at least 4x cable diameter for UTP and 10x for fiber; never exceed 100 m channel length for twisted-pair Ethernet.

ECC RDIMM vs LRDIMM: RDIMMs buffer the address/command bus; LRDIMMs buffer data too and allow higher ranks per channel, so LRDIMM wins when capacity per slot matters more than latency.

Memory population rule: populate channels evenly and match DIMMs per channel; unbalanced population drops from quad- to dual- or single-channel mode and cuts bandwidth.

NUMA: each CPU socket owns local memory; a VM or process spanning sockets pays remote-memory latency, so pin latency-sensitive workloads to one node.

RAID levels: RAID 0 stripes with no redundancy; RAID 1 mirrors; RAID 5 tolerates one drive; RAID 6 tolerates two; RAID 10 mirrors then stripes and rebuilds faster than parity RAID.

Usable capacity formulas: RAID 5 = (N-1) x smallest drive; RAID 6 = (N-2) x smallest drive; RAID 10 = N/2 x smallest drive.

SSD vs HDD: SSDs have no seek time and far higher random IOPS but finite program/erase cycles; DWPD x capacity x 365 gives the warranted terabytes written per year.

Storage protocols: SAS is dual-ported and used for internal drives; SATA is single-ported; NVMe runs over PCIe and bypasses the SAS/SATA controller queue limits.

Boot and firmware: UEFI with GPT supports disks over 2 TB and secure boot; legacy BIOS with MBR caps boot volumes at 2 TB.

Virtualization support: Intel VT-x/EPT and AMD-V/NPT must be enabled in BIOS, and SR-IOV or PCIe passthrough is required for direct device access by a VM.

VLANs segment a Layer 2 domain; a trunk port carries multiple VLANs with 802.1Q tags, and the native VLAN is sent untagged, so both ends must agree on it.

Spanning Tree: STP blocks loops; RSTP converges in seconds; use edge/portfast on server ports and BPDU guard to prevent rogue switches.

Link aggregation (LACP) bundles links into one logical channel; all member links must match speed, duplex, and VLAN configuration, and hashing is per flow, not per packet.

MTU: jumbo frames are 9000 bytes; every device in the path, including the vSwitch and VM, must match or large packets silently drop.

vSwitch types: a standard vSwitch is per-host; a distributed switch is managed centrally and keeps port groups consistent across hosts for vMotion compatibility.

NIC teaming policies: route based on originating virtual port ID is default; IP hash requires LACP on the physical switch; failover order determines active/standby uplinks.

Overlay networking: VXLAN encapsulates Layer 2 frames in UDP (port 4789) with a 24-bit VNI, giving about 16 million segments and requiring an IP underlay.

iSCSI and NFS storage traffic belongs on dedicated VLANs or separate physical uplinks; iSCSI needs end-to-end jumbo frame consistency and CHAP for authentication.

IP addressing: /24 gives 254 usable hosts, /25 gives 126, /26 gives 62; private ranges are 10.0.0.0/8, 172.16.0.0/12, and 192.168.0.0/16.

Troubleshooting order: physical layer first (cable, link LED, optic), then link speed/duplex, then VLAN/IP, then routing and firewall, then application.

Link LED dark on both ends points to cable or transceiver; LED lit but no traffic points to VLAN mismatch, duplex mismatch, or a down switchport.

Intermittent CRC or FCS errors on an interface usually indicate a bad cable, dirty or mismated optic, or duplex mismatch rather than a software fault.

Server will not power on: check both PSU LEDs, confirm both feeds are live, verify the power button and BMC/iLO/iDRAC access, then review SEL logs for a fault.

Memory errors: correctable ECC errors are logged and scrubbed; uncorrectable errors trigger a machine check; DIMM slot location in the SEL identifies the failing module.

Firmware and driver baseline: apply the vendor support matrix combination of BIOS, BMC, NIC firmware, and OS driver; mismatched versions cause link flaps and boot failures.

Change control: record device, change, rollback plan, and window; capture a configuration backup before any switch or storage change so rollback is possible.

Escalation data to collect: timestamps, affected hosts and ports, switch logs, BMC SEL, and recent changes; without this the vendor case stalls.

Decommissioning: wipe drives per policy, remove host from cluster and monitoring, release IPs and DNS records, and update the asset inventory and cabling records.

UTP categories: Cat5e supports 1 GbE to 100 m; Cat6 supports 1 GbE to 100 m and 10GBASE-T to 55 m; Cat6a supports 10GBASE-T to 100 m.

Fiber types: OM3/OM4 multimode suit 10G-100G short reach (OM4 10GBASE-SR to 400 m); OS2 single-mode carries 10G-400G over kilometers via 10GBASE-LR/ER optics.

Connector pairings: LC duplex for SFP/SFP+ optics, MPO/MTP for 40G/100G parallel multimode (QSFP), SC for legacy single-mode links; APC ferrules are angled and must not mate with UPC.

10GBASE-T SFP+ modules draw roughly 2.5-4 W per port versus under 1 W for 10GBASE-SR optics, so copper SFP+ is limited to short in-rack runs and low port counts.

Twinax DAC cables carry passive 10G/25G to about 5-7 m and active variants to about 10-15 m; they are the default in-rack server-to-switch interconnect.

Power: a 1U server at 450 W on 208 V single-phase draws about 2.2 A; a 30 A 208 V circuit derated to 80% delivers 24 A, or about 5.0 kVA.

Redundant PSU sizing: sum load, then size each feed so one feed carries the full load; N+1 adds one spare unit, 2N duplicates the whole chain.

Rack unit = 44.45 mm (1.75 in); a standard 42U rack is about 1.87 m of usable vertical space, and airflow follows the cold-aisle-front, hot-aisle-rear convention.

Cable management rules: bend radius at least 4x cable diameter for UTP and 10x for fiber; never exceed 100 m channel length for twisted-pair Ethernet.

ECC RDIMM vs LRDIMM: RDIMMs buffer the address/command bus; LRDIMMs buffer data too and allow higher ranks per channel, so LRDIMM wins when capacity per slot matters more than latency.

Memory population rule: populate channels evenly and match DIMMs per channel; unbalanced population drops from quad- to dual- or single-channel mode and cuts bandwidth.

NUMA: each CPU socket owns local memory; a VM or process spanning sockets pays remote-memory latency, so pin latency-sensitive workloads to one node.

RAID levels: RAID 0 stripes with no redundancy; RAID 1 mirrors; RAID 5 tolerates one drive; RAID 6 tolerates two; RAID 10 mirrors then stripes and rebuilds faster than parity RAID.

Usable capacity formulas: RAID 5 = (N-1) x smallest drive; RAID 6 = (N-2) x smallest drive; RAID 10 = N/2 x smallest drive.

SSD vs HDD: SSDs have no seek time and far higher random IOPS but finite program/erase cycles; DWPD x capacity x 365 gives the warranted terabytes written per year.

Storage protocols: SAS is dual-ported and used for internal drives; SATA is single-ported; NVMe runs over PCIe and bypasses the SAS/SATA controller queue limits.

Boot and firmware: UEFI with GPT supports disks over 2 TB and secure boot; legacy BIOS with MBR caps boot volumes at 2 TB.

Virtualization support: Intel VT-x/EPT and AMD-V/NPT must be enabled in BIOS, and SR-IOV or PCIe passthrough is required for direct device access by a VM.

VLANs segment a Layer 2 domain; a trunk port carries multiple VLANs with 802.1Q tags, and the native VLAN is sent untagged, so both ends must agree on it.

Spanning Tree: STP blocks loops; RSTP converges in seconds; use edge/portfast on server ports and BPDU guard to prevent rogue switches.

Link aggregation (LACP) bundles links into one logical channel; all member links must match speed, duplex, and VLAN configuration, and hashing is per flow, not per packet.

MTU: jumbo frames are 9000 bytes; every device in the path, including the vSwitch and VM, must match or large packets silently drop.

vSwitch types: a standard vSwitch is per-host; a distributed switch is managed centrally and keeps port groups consistent across hosts for vMotion compatibility.

NIC teaming policies: route based on originating virtual port ID is default; IP hash requires LACP on the physical switch; failover order determines active/standby uplinks.

Overlay networking: VXLAN encapsulates Layer 2 frames in UDP (port 4789) with a 24-bit VNI, giving about 16 million segments and requiring an IP underlay.

iSCSI and NFS storage traffic belongs on dedicated VLANs or separate physical uplinks; iSCSI needs end-to-end jumbo frame consistency and CHAP for authentication.

IP addressing: /24 gives 254 usable hosts, /25 gives 126, /26 gives 62; private ranges are 10.0.0.0/8, 172.16.0.0/12, and 192.168.0.0/16.

Troubleshooting order: physical layer first (cable, link LED, optic), then link speed/duplex, then VLAN/IP, then routing and firewall, then application.

Link LED dark on both ends points to cable or transceiver; LED lit but no traffic points to VLAN mismatch, duplex mismatch, or a down switchport.

Intermittent CRC or FCS errors on an interface usually indicate a bad cable, dirty or mismated optic, or duplex mismatch rather than a software fault.

Server will not power on: check both PSU LEDs, confirm both feeds are live, verify the power button and BMC/iLO/iDRAC access, then review SEL logs for a fault.

Memory errors: correctable ECC errors are logged and scrubbed; uncorrectable errors trigger a machine check; DIMM slot location in the SEL identifies the failing module.

Firmware and driver baseline: apply the vendor support matrix combination of BIOS, BMC, NIC firmware, and OS driver; mismatched versions cause link flaps and boot failures.

Change control: record device, change, rollback plan, and window; capture a configuration backup before any switch or storage change so rollback is possible.

Escalation data to collect: timestamps, affected hosts and ports, switch logs, BMC SEL, and recent changes; without this the vendor case stalls.

Decommissioning: wipe drives per policy, remove host from cluster and monitoring, release IPs and DNS records, and update the asset inventory and cabling records.

UTP categories: Cat5e supports 1 GbE to 100 m; Cat6 supports 1 GbE to 100 m and 10GBASE-T to 55 m; Cat6a supports 10GBASE-T to 100 m.

Fiber types: OM3/OM4 multimode suit 10G-100G short reach (OM4 10GBASE-SR to 400 m); OS2 single-mode carries 10G-400G over kilometers via 10GBASE-LR/ER optics.

Connector pairings: LC duplex for SFP/SFP+ optics, MPO/MTP for 40G/100G parallel multimode (QSFP), SC for legacy single-mode links; APC ferrules are angled and must not mate with UPC.

10GBASE-T SFP+ modules draw roughly 2.5-4 W per port versus under 1 W for 10GBASE-SR optics, so copper SFP+ is limited to short in-rack runs and low port counts.

Twinax DAC cables carry passive 10G/25G to about 5-7 m and active variants to about 10-15 m; they are the default in-rack server-to-switch interconnect.

Power: a 1U server at 450 W on 208 V single-phase draws about 2.2 A; a 30 A 208 V circuit derated to 80% delivers 24 A, or about 5.0 kVA.

Redundant PSU sizing: sum load, then size each feed so one feed carries the full load; N+1 adds one spare unit, 2N duplicates the whole chain.

Rack unit = 44.45 mm (1.75 in); a standard 42U rack is about 1.87 m of usable vertical space, and airflow follows the cold-aisle-front, hot-aisle-rear convention.

Cable management rules: bend radius at least 4x cable diameter for UTP and 10x for fiber; never exceed 100 m channel length for twisted-pair Ethernet.

ECC RDIMM vs LRDIMM: RDIMMs buffer the address/command bus; LRDIMMs buffer data too and allow higher ranks per channel, so LRDIMM wins when capacity per slot matters more than latency.

Memory population rule: populate channels evenly and match DIMMs per channel; unbalanced population drops from quad- to dual- or single-channel mode and cuts bandwidth.

NUMA: each CPU socket owns local memory; a VM or process spanning sockets pays remote-memory latency, so pin latency-sensitive workloads to one node.

RAID levels: RAID 0 stripes with no redundancy; RAID 1 mirrors; RAID 5 tolerates one drive; RAID 6 tolerates two; RAID 10 mirrors then stripes and rebuilds faster than parity RAID.

Usable capacity formulas: RAID 5 = (N-1) x smallest drive; RAID 6 = (N-2) x smallest drive; RAID 10 = N/2 x smallest drive.

SSD vs HDD: SSDs have no seek time and far higher random IOPS but finite program/erase cycles; DWPD x capacity x 365 gives the warranted terabytes written per year.

Storage protocols: SAS is dual-ported and used for internal drives; SATA is single-ported; NVMe runs over PCIe and bypasses the SAS/SATA controller queue limits.

Boot and firmware: UEFI with GPT supports disks over 2 TB and secure boot; legacy BIOS with MBR caps boot volumes at 2 TB.

Virtualization support: Intel VT-x/EPT and AMD-V/NPT must be enabled in BIOS, and SR-IOV or PCIe passthrough is required for direct device access by a VM.

VLANs segment a Layer 2 domain; a trunk port carries multiple VLANs with 802.1Q tags, and the native VLAN is sent untagged, so both ends must agree on it.

Spanning Tree: STP blocks loops; RSTP converges in seconds; use edge/portfast on server ports and BPDU guard to prevent rogue switches.

Link aggregation (LACP) bundles links into one logical channel; all member links must match speed, duplex, and VLAN configuration, and hashing is per flow, not per packet.

MTU: jumbo frames are 9000 bytes; every device in the path, including the vSwitch and VM, must match or large packets silently drop.

vSwitch types: a standard vSwitch is per-host; a distributed switch is managed centrally and keeps port groups consistent across hosts for vMotion compatibility.

NIC teaming policies: route based on originating virtual port ID is default; IP hash requires LACP on the physical switch; failover order determines active/standby uplinks.

Overlay networking: VXLAN encapsulates Layer 2 frames in UDP (port 4789) with a 24-bit VNI, giving about 16 million segments and requiring an IP underlay.

iSCSI and NFS storage traffic belongs on dedicated VLANs or separate physical uplinks; iSCSI needs end-to-end jumbo frame consistency and CHAP for authentication.

IP addressing: /24 gives 254 usable hosts, /25 gives 126, /26 gives 62; private ranges are 10.0.0.0/8, 172.16.0.0/12, and 192.168.0.0/16.

Troubleshooting order: physical layer first (cable, link LED, optic), then link speed/duplex, then VLAN/IP, then routing and firewall, then application.

Link LED dark on both ends points to cable or transceiver; LED lit but no traffic points to VLAN mismatch, duplex mismatch, or a down switchport.

Intermittent CRC or FCS errors on an interface usually indicate a bad cable, dirty or mismated optic, or duplex mismatch rather than a software fault.

Server will not power on: check both PSU LEDs, confirm both feeds are live, verify the power button and BMC/iLO/iDRAC access, then review SEL logs for a fault.

Memory errors: correctable ECC errors are logged and scrubbed; uncorrectable errors trigger a machine check; DIMM slot location in the SEL identifies the failing module.

Firmware and driver baseline: apply the vendor support matrix combination of BIOS, BMC, NIC firmware, and OS driver; mismatched versions cause link flaps and boot failures.

Change control: record device, change, rollback plan, and window; capture a configuration backup before any switch or storage change so rollback is possible.

Escalation data to collect: timestamps, affected hosts and ports, switch logs, BMC SEL, and recent changes; without this the vendor case stalls.

Decommissioning: wipe drives per policy, remove host from cluster and monitoring, release IPs and DNS records, and update the asset inventory and cabling records.

UTP categories: Cat5e supports 1 GbE to 100 m; Cat6 supports 1 GbE to 100 m and 10GBASE-T to 55 m; Cat6a supports 10GBASE-T to 100 m.

Fiber types: OM3/OM4 multimode suit 10G-100G short reach (OM4 10GBASE-SR to 400 m); OS2 single-mode carries 10G-400G over kilometers via 10GBASE-LR/ER optics.

Connector pairings: LC duplex for SFP/SFP+ optics, MPO/MTP for 40G/100G parallel multimode (QSFP), SC for legacy single-mode links; APC ferrules are angled and must not mate with UPC.

10GBASE-T SFP+ modules draw roughly 2.5-4 W per port versus under 1 W for 10GBASE-SR optics, so copper SFP+ is limited to short in-rack runs and low port counts.

Twinax DAC cables carry passive 10G/25G to about 5-7 m and active variants to about 10-15 m; they are the default in-rack server-to-switch interconnect.

Power: a 1U server at 450 W on 208 V single-phase draws about 2.2 A; a 30 A 208 V circuit derated to 80% delivers 24 A, or about 5.0 kVA.

Redundant PSU sizing: sum load, then size each feed so one feed carries the full load; N+1 adds one spare unit, 2N duplicates the whole chain.

Rack unit = 44.45 mm (1.75 in); a standard 42U rack is about 1.87 m of usable vertical space, and airflow follows the cold-aisle-front, hot-aisle-rear convention.

Cable management rules: bend radius at least 4x cable diameter for UTP and 10x for fiber; never exceed 100 m channel length for twisted-pair Ethernet.

ECC RDIMM vs LRDIMM: RDIMMs buffer the address/command bus; LRDIMMs buffer data too and allow higher ranks per channel, so LRDIMM wins when capacity per slot matters more than latency.

Memory population rule: populate channels evenly and match DIMMs per channel; unbalanced population drops from quad- to dual- or single-channel mode and cuts bandwidth.

NUMA: each CPU socket owns local memory; a VM or process spanning sockets pays remote-memory latency, so pin latency-sensitive workloads to one node.

RAID levels: RAID 0 stripes with no redundancy; RAID 1 mirrors; RAID 5 tolerates one drive; RAID 6 tolerates two; RAID 10 mirrors then stripes and rebuilds faster than parity RAID.

Usable capacity formulas: RAID 5 = (N-1) x smallest drive; RAID 6 = (N-2) x smallest drive; RAID 10 = N/2 x smallest drive.

SSD vs HDD: SSDs have no seek time and far higher random IOPS but finite program/erase cycles; DWPD x capacity x 365 gives the warranted terabytes written per year.

Storage protocols: SAS is dual-ported and used for internal drives; SATA is single-ported; NVMe runs over PCIe and bypasses the SAS/SATA controller queue limits.

Boot and firmware: UEFI with GPT supports disks over 2 TB and secure boot; legacy BIOS with MBR caps boot volumes at 2 TB.

Virtualization support: Intel VT-x/EPT and AMD-V/NPT must be enabled in BIOS, and SR-IOV or PCIe passthrough is required for direct device access by a VM.

VLANs segment a Layer 2 domain; a trunk port carries multiple VLANs with 802.1Q tags, and the native VLAN is sent untagged, so both ends must agree on it.

Spanning Tree: STP blocks loops; RSTP converges in seconds; use edge/portfast on server ports and BPDU guard to prevent rogue switches.

Link aggregation (LACP) bundles links into one logical channel; all member links must match speed, duplex, and VLAN configuration, and hashing is per flow, not per packet.

MTU: jumbo frames are 9000 bytes; every device in the path, including the vSwitch and VM, must match or large packets silently drop.

vSwitch types: a standard vSwitch is per-host; a distributed switch is managed centrally and keeps port groups consistent across hosts for vMotion compatibility.

NIC teaming policies: route based on originating virtual port ID is default; IP hash requires LACP on the physical switch; failover order determines active/standby uplinks.

Overlay networking: VXLAN encapsulates Layer 2 frames in UDP (port 4789) with a 24-bit VNI, giving about 16 million segments and requiring an IP underlay.

iSCSI and NFS storage traffic belongs on dedicated VLANs or separate physical uplinks; iSCSI needs end-to-end jumbo frame consistency and CHAP for authentication.

IP addressing: /24 gives 254 usable hosts, /25 gives 126, /26 gives 62; private ranges are 10.0.0.0/8, 172.16.0.0/12, and 192.168.0.0/16.

Troubleshooting order: physical layer first (cable, link LED, optic), then link speed/duplex, then VLAN/IP, then routing and firewall, then application.

Link LED dark on both ends points to cable or transceiver; LED lit but no traffic points to VLAN mismatch, duplex mismatch, or a down switchport.

Intermittent CRC or FCS errors on an interface usually indicate a bad cable, dirty or mismated optic, or duplex mismatch rather than a software fault.

Server will not power on: check both PSU LEDs, confirm both feeds are live, verify the power button and BMC/iLO/iDRAC access, then review SEL logs for a fault.

Memory errors: correctable ECC errors are logged and scrubbed; uncorrectable errors trigger a machine check; DIMM slot location in the SEL identifies the failing module.

Firmware and driver baseline: apply the vendor support matrix combination of BIOS, BMC, NIC firmware, and OS driver; mismatched versions cause link flaps and boot failures.

Change control: record device, change, rollback plan, and window; capture a configuration backup before any switch or storage change so rollback is possible.

Escalation data to collect: timestamps, affected hosts and ports, switch logs, BMC SEL, and recent changes; without this the vendor case stalls.

Decommissioning: wipe drives per policy, remove host from cluster and monitoring, release IPs and DNS records, and update the asset inventory and cabling records.

UTP categories: Cat5e supports 1 GbE to 100 m; Cat6 supports 1 GbE to 100 m and 10GBASE-T to 55 m; Cat6a supports 10GBASE-T to 100 m.

Fiber types: OM3/OM4 multimode suit 10G-100G short reach (OM4 10GBASE-SR to 400 m); OS2 single-mode carries 10G-400G over kilometers via 10GBASE-LR/ER optics.

Connector pairings: LC duplex for SFP/SFP+ optics, MPO/MTP for 40G/100G parallel multimode (QSFP), SC for legacy single-mode links; APC ferrules are angled and must not mate with UPC.

10GBASE-T SFP+ modules draw roughly 2.5-4 W per port versus under 1 W for 10GBASE-SR optics, so copper SFP+ is limited to short in-rack runs and low port counts.

Twinax DAC cables carry passive 10G/25G to about 5-7 m and active variants to about 10-15 m; they are the default in-rack server-to-switch interconnect.

Power: a 1U server at 450 W on 208 V single-phase draws about 2.2 A; a 30 A 208 V circuit derated to 80% delivers 24 A, or about 5.0 kVA.

Redundant PSU sizing: sum load, then size each feed so one feed carries the full load; N+1 adds one spare unit, 2N duplicates the whole chain.

Rack unit = 44.45 mm (1.75 in); a standard 42U rack is about 1.87 m of usable vertical space, and airflow follows the cold-aisle-front, hot-aisle-rear convention.

Cable management rules: bend radius at least 4x cable diameter for UTP and 10x for fiber; never exceed 100 m channel length for twisted-pair Ethernet.

ECC RDIMM vs LRDIMM: RDIMMs buffer the address/command bus; LRDIMMs buffer data too and allow higher ranks per channel, so LRDIMM wins when capacity per slot matters more than latency.

Memory population rule: populate channels evenly and match DIMMs per channel; unbalanced population drops from quad- to dual- or single-channel mode and cuts bandwidth.

NUMA: each CPU socket owns local memory; a VM or process spanning sockets pays remote-memory latency, so pin latency-sensitive workloads to one node.

RAID levels: RAID 0 stripes with no redundancy; RAID 1 mirrors; RAID 5 tolerates one drive; RAID 6 tolerates two; RAID 10 mirrors then stripes and rebuilds faster than parity RAID.

Usable capacity formulas: RAID 5 = (N-1) x smallest drive; RAID 6 = (N-2) x smallest drive; RAID 10 = N/2 x smallest drive.

SSD vs HDD: SSDs have no seek time and far higher random IOPS but finite program/erase cycles; DWPD x capacity x 365 gives the warranted terabytes written per year.

Storage protocols: SAS is dual-ported and used for internal drives; SATA is single-ported; NVMe runs over PCIe and bypasses the SAS/SATA controller queue limits.

Boot and firmware: UEFI with GPT supports disks over 2 TB and secure boot; legacy BIOS with MBR caps boot volumes at 2 TB.

Virtualization support: Intel VT-x/EPT and AMD-V/NPT must be enabled in BIOS, and SR-IOV or PCIe passthrough is required for direct device access by a VM.

VLANs segment a Layer 2 domain; a trunk port carries multiple VLANs with 802.1Q tags, and the native VLAN is sent untagged, so both ends must agree on it.

Spanning Tree: STP blocks loops; RSTP converges in seconds; use edge/portfast on server ports and BPDU guard to prevent rogue switches.

Link aggregation (LACP) bundles links into one logical channel; all member links must match speed, duplex, and VLAN configuration, and hashing is per flow, not per packet.

MTU: jumbo frames are 9000 bytes; every device in the path, including the vSwitch and VM, must match or large packets silently drop.

vSwitch types: a standard vSwitch is per-host; a distributed switch is managed centrally and keeps port groups consistent across hosts for vMotion compatibility.

NIC teaming policies: route based on originating virtual port ID is default; IP hash requires LACP on the physical switch; failover order determines active/standby uplinks.

Overlay networking: VXLAN encapsulates Layer 2 frames in UDP (port 4789) with a 24-bit VNI, giving about 16 million segments and requiring an IP underlay.

iSCSI and NFS storage traffic belongs on dedicated VLANs or separate physical uplinks; iSCSI needs end-to-end jumbo frame consistency and CHAP for authentication.

IP addressing: /24 gives 254 usable hosts, /25 gives 126, /26 gives 62; private ranges are 10.0.0.0/8, 172.16.0.0/12, and 192.168.0.0/16.

Troubleshooting order: physical layer first (cable, link LED, optic), then link speed/duplex, then VLAN/IP, then routing and firewall, then application.

Link LED dark on both ends points to cable or transceiver; LED lit but no traffic points to VLAN mismatch, duplex mismatch, or a down switchport.

Intermittent CRC or FCS errors on an interface usually indicate a bad cable, dirty or mismated optic, or duplex mismatch rather than a software fault.

Server will not power on: check both PSU LEDs, confirm both feeds are live, verify the power button and BMC/iLO/iDRAC access, then review SEL logs for a fault.

Memory errors: correctable ECC errors are logged and scrubbed; uncorrectable errors trigger a machine check; DIMM slot location in the SEL identifies the failing module.

Firmware and driver baseline: apply the vendor support matrix combination of BIOS, BMC, NIC firmware, and OS driver; mismatched versions cause link flaps and boot failures.

Change control: record device, change, rollback plan, and window; capture a configuration backup before any switch or storage change so rollback is possible.

Escalation data to collect: timestamps, affected hosts and ports, switch logs, BMC SEL, and recent changes; without this the vendor case stalls.

Decommissioning: wipe drives per policy, remove host from cluster and monitoring, release IPs and DNS records, and update the asset inventory and cabling records.

UTP categories: Cat5e supports 1 GbE to 100 m; Cat6 supports 1 GbE to 100 m and 10GBASE-T to 55 m; Cat6a supports 10GBASE-T to 100 m.

Fiber types: OM3/OM4 multimode suit 10G-100G short reach (OM4 10GBASE-SR to 400 m); OS2 single-mode carries 10G-400G over kilometers via 10GBASE-LR/ER optics.

Connector pairings: LC duplex for SFP/SFP+ optics, MPO/MTP for 40G/100G parallel multimode (QSFP), SC for legacy single-mode links; APC ferrules are angled and must not mate with UPC.

10GBASE-T SFP+ modules draw roughly 2.5-4 W per port versus under 1 W for 10GBASE-SR optics, so copper SFP+ is limited to short in-rack runs and low port counts.

Twinax DAC cables carry passive 10G/25G to about 5-7 m and active variants to about 10-15 m; they are the default in-rack server-to-switch interconnect.

Power: a 1U server at 450 W on 208 V single-phase draws about 2.2 A; a 30 A 208 V circuit derated to 80% delivers 24 A, or about 5.0 kVA.

Redundant PSU sizing: sum load, then size each feed so one feed carries the full load; N+1 adds one spare unit, 2N duplicates the whole chain.

Rack unit = 44.45 mm (1.75 in); a standard 42U rack is about 1.87 m of usable vertical space, and airflow follows the cold-aisle-front, hot-aisle-rear convention.

Cable management rules: bend radius at least 4x cable diameter for UTP and 10x for fiber; never exceed 100 m channel length for twisted-pair Ethernet.

ECC RDIMM vs LRDIMM: RDIMMs buffer the address/command bus; LRDIMMs buffer data too and allow higher ranks per channel, so LRDIMM wins when capacity per slot matters more than latency.

Memory population rule: populate channels evenly and match DIMMs per channel; unbalanced population drops from quad- to dual- or single-channel mode and cuts bandwidth.

NUMA: each CPU socket owns local memory; a VM or process spanning sockets pays remote-memory latency, so pin latency-sensitive workloads to one node.

RAID levels: RAID 0 stripes with no redundancy; RAID 1 mirrors; RAID 5 tolerates one drive; RAID 6 tolerates two; RAID 10 mirrors then stripes and rebuilds faster than parity RAID.

Usable capacity formulas: RAID 5 = (N-1) x smallest drive; RAID 6 = (N-2) x smallest drive; RAID 10 = N/2 x smallest drive.

SSD vs HDD: SSDs have no seek time and far higher random IOPS but finite program/erase cycles; DWPD x capacity x 365 gives the warranted terabytes written per year.

Storage protocols: SAS is dual-ported and used for internal drives; SATA is single-ported; NVMe runs over PCIe and bypasses the SAS/SATA controller queue limits.

Boot and firmware: UEFI with GPT supports disks over 2 TB and secure boot; legacy BIOS with MBR caps boot volumes at 2 TB.

Virtualization support: Intel VT-x/EPT and AMD-V/NPT must be enabled in BIOS, and SR-IOV or PCIe passthrough is required for direct device access by a VM.

VLANs segment a Layer 2 domain; a trunk port carries multiple VLANs with 802.1Q tags, and the native VLAN is sent untagged, so both ends must agree on it.

Spanning Tree: STP blocks loops; RSTP converges in seconds; use edge/portfast on server ports and BPDU guard to prevent rogue switches.

Link aggregation (LACP) bundles links into one logical channel; all member links must match speed, duplex, and VLAN configuration, and hashing is per flow, not per packet.

MTU: jumbo frames are 9000 bytes; every device in the path, including the vSwitch and VM, must match or large packets silently drop.

vSwitch types: a standard vSwitch is per-host; a distributed switch is managed centrally and keeps port groups consistent across hosts for vMotion compatibility.

NIC teaming policies: route based on originating virtual port ID is default; IP hash requires LACP on the physical switch; failover order determines active/standby uplinks.

Overlay networking: VXLAN encapsulates Layer 2 frames in UDP (port 4789) with a 24-bit VNI, giving about 16 million segments and requiring an IP underlay.

iSCSI and NFS storage traffic belongs on dedicated VLANs or separate physical uplinks; iSCSI needs end-to-end jumbo frame consistency and CHAP for authentication.

IP addressing: /24 gives 254 usable hosts, /25 gives 126, /26 gives 62; private ranges are 10.0.0.0/8, 172.16.0.0/12, and 192.168.0.0/16.

Troubleshooting order: physical layer first (cable, link LED, optic), then link speed/duplex, then VLAN/IP, then routing and firewall, then application.

Link LED dark on both ends points to cable or transceiver; LED lit but no traffic points to VLAN mismatch, duplex mismatch, or a down switchport.

Intermittent CRC or FCS errors on an interface usually indicate a bad cable, dirty or mismated optic, or duplex mismatch rather than a software fault.

Server will not power on: check both PSU LEDs, confirm both feeds are live, verify the power button and BMC/iLO/iDRAC access, then review SEL logs for a fault.

Memory errors: correctable ECC errors are logged and scrubbed; uncorrectable errors trigger a machine check; DIMM slot location in the SEL identifies the failing module.

Firmware and driver baseline: apply the vendor support matrix combination of BIOS, BMC, NIC firmware, and OS driver; mismatched versions cause link flaps and boot failures.

Change control: record device, change, rollback plan, and window; capture a configuration backup before any switch or storage change so rollback is possible.

Escalation data to collect: timestamps, affected hosts and ports, switch logs, BMC SEL, and recent changes; without this the vendor case stalls.

Decommissioning: wipe drives per policy, remove host from cluster and monitoring, release IPs and DNS records, and update the asset inventory and cabling records.