DCAT cheat sheet

Tier I–IV (Uptime Institute) map to 99.671%, 99.741%, 99.982%, and 99.995% availability; Tier IV adds fault tolerance with concurrently maintainable paths.

Availability % = (Agreed uptime − Downtime) ÷ Agreed uptime × 100; annual minutes = 525,600.

N+1 provides one redundant component; 2N provides two full independent capacity paths; 2(N+1) combines both.

Redundancy classifications: N, N+1, 2N, and 2(N+1); higher tiers typically require N+1 or 2N distribution paths.

Rack unit (U) = 1.75 in (44.45 mm); a standard 42U rack holds 42 U of equipment.

White space = raised-floor equipment area; grey space = supporting mechanical/electrical rooms.

Hot aisle/cold aisle layout separates intake and exhaust; containment prevents hot/cold air mixing.

Cross-connect (MDA/HDA/EDA hierarchy) links distribution areas; main distribution area houses core switches and routers.

PUE = Total facility power ÷ IT equipment power; ideal approaches 1.0, typical enterprise values are ~1.5–2.0.

CPU socket count determines physical processor count; cores × threads per core gives logical processor count.

DDR4 DIMM has 288 pins; DDR5 DIMM has 288 pins but a different keying and higher per-DIMM capacity.

Memory population rules: matched DIMMs per channel, balanced across channels, and follow vendor population order to avoid unbalanced interleaving.

ECC detects and corrects single-bit errors and detects (not always corrects) multi-bit errors; registered/buffered DIMMs add a register between DRAM and controller.

NUMA: each CPU owns local memory; remote memory access is slower, so pin workloads to local NUMA nodes for latency-sensitive apps.

PCIe lane counts: x1, x4, x8, x16; PCIe generations roughly double per-lane bandwidth each generation (Gen3 ~1 GB/s per lane, Gen4 ~2 GB/s).

RAID on server: RAID 0 stripes (no redundancy), RAID 1 mirrors, RAID 5 needs ≥3 drives with one parity, RAID 6 needs ≥4 drives with dual parity, RAID 10 mirrors then stripes.

IPMI/BMC provides out-of-band management independent of the OS; Redfish is the RESTful successor API.

Hot-swap components (drives, fans, PSUs) can be replaced with the system running only if the platform supports redundancy.

HDD access is mechanical (seek + rotational latency); SSD access is electronic, so random IOPS differ by orders of magnitude.

IOPS and throughput (MB/s) are distinct metrics; sequential workloads favor throughput, random workloads favor IOPS.

RAID 5 usable capacity = (N − 1) × drive size; RAID 6 = (N − 2) × drive size; RAID 10 = N/2 × drive size.

RAID is not backup; it protects against drive failure, not deletion, corruption, or site loss.

SAN presents block storage (FC or iSCSI); NAS presents file storage (NFS/SMB); object storage uses HTTP APIs with buckets and keys.

Fibre Channel speeds: 8/16/32/64 GFC; iSCSI runs SCSI over TCP/IP and can use standard Ethernet.

LUN masking and zoning control which hosts see which block devices; zoning applies to FC fabrics, masking to array LUNs.

Thin provisioning allocates capacity on demand; over-subscription risks running out of physical space.

Replication can be synchronous (zero RPO, latency-bound) or asynchronous (nonzero RPO, distance-tolerant).

OSI layers 1–7: physical, data link, network, transport, session, presentation, application.

IPv4 private ranges: 10.0.0.0/8, 172.16.0.0/12, 192.168.0.0/16; link-local 169.254.0.0/16 indicates DHCP failure.

VLANs segment Layer 2 broadcast domains; inter-VLAN routing requires a Layer 3 device.

Spanning Tree Protocol prevents Layer 2 loops; RSTP converges faster than classic STP.

Top-of-rack (ToR) switching connects servers to access layer; spine-leaf provides predictable east-west latency and scale.

Fiber types: single-mode for long distance, multimode for short reach; OM3/OM4 multimode support 10G/40G/100G over limited distances.

Transceivers: SFP+ (10G), QSFP+ (40G), QSFP28 (100G); DAC cables are short-reach copper alternatives to optics.

Jumbo frames (typically 9000 MTU) must be configured end-to-end or packets are dropped or fragmented.

LACP aggregates multiple links into one logical channel for bandwidth and redundancy.

Power formula: Watts = Volts × Amps; for AC, W = V × A × power factor.

UPS types: offline/standby, line-interactive, and double-conversion online (best protection, lowest efficiency).

PDU types: basic, metered, monitored, and switched; per-outlet metering enables capacity tracking.

Redundant power: dual-corded servers should connect to A/B power paths fed from independent sources.

ASHRAE recommended inlet range is roughly 18–27 °C (64–81 °F); allowable envelope extends wider but with reliability trade-offs.

Cooling capacity often expressed in tons; 1 ton of cooling = 12,000 BTU/hr ≈ 3.517 kW.

Airflow: cold air enters the front of the server, hot air exhausts the rear; never reverse server orientation in a rack.

Humidity control: too low risks ESD, too high risks condensation; ASHRAE recommends roughly 8–60% RH depending on class.

Battery strings in a UPS have finite life; valve-regulated lead-acid typically needs replacement every 3–5 years.

Rack installation: mount rails, secure with cage nuts and screws, verify front-to-rear airflow, and leave clearance for cable management.

Cable categories: Cat5e supports 1G, Cat6 supports 1G/10G to 55 m, Cat6a supports 10G to 100 m.

Maximum copper horizontal run is 90 m plus 10 m of patch cords (100 m channel).

Fiber connector types: LC (small form factor, duplex), SC (square), MPO/MTP (multi-fiber for 40G/100G).

Label both ends of every cable and document port assignments before and after changes.

Hot-swap replacement sequence: verify redundancy, identify the failed unit, remove and replace, confirm status returns to healthy.

Anti-static handling: wear ESD wrist strap bonded to chassis ground, keep components in antistatic bags until install.

Before servicing, follow lockout/tagout and confirm the correct server via asset tag and management console.

Firmware and BIOS updates should be applied per vendor compatibility matrix and validated in a test environment first.

Structured troubleshooting: identify the problem, establish a theory, test the theory, establish a plan of action, verify full functionality, document.

Start at Layer 1: check power, cabling, link lights, and transceiver seating before higher-layer diagnosis.

Ping tests reachability; traceroute shows path; both fail if ICMP is filtered, so absence of reply is not proof of outage.

Common server health indicators: PSU status, fan RPM, temperature sensors, memory ECC errors, drive SMART status.

Logs: OS syslog/event logs, BMC/IPMI SEL (system event log), and switch syslog are primary diagnostic sources.

Baseline metrics (CPU, memory, disk latency, network utilization) enable anomaly detection; compare against baseline, not intuition.

Disk latency thresholds: sustained high queue depth with rising latency indicates saturation; check for failing drives via SMART.

Redfish/IPMI out-of-band access works when the OS is down; use it to check sensors and power state remotely.

Change control: correlate incidents with recent changes; roll back if a change precedes the fault.

Electrical safety: de-energize before working on circuits; use lockout/tagout; treat all conductors as live until verified.

Arc flash boundaries and PPE requirements are defined by NFPA 70E; never work alone on energized equipment.

ESD controls: wrist straps, heel straps, grounded mats, and humidity control prevent component damage.

Physical security layers: perimeter, building access control, cage/mantrap, rack locks, and audit logging.

Logical security: least privilege, role-based access, MFA for management interfaces, and disabling unused ports.

Compliance frameworks: PCI DSS for card data, HIPAA for health data, SOC 2 for service controls, ISO 27001 for ISMS.

Data retention and destruction: follow policy for media sanitization (overwrite, crypto-erase, or physical destruction).

Fire suppression: clean agents (e.g., FM-200, Novec) are used in white space because they leave no residue; water systems are avoided near energized IT.

Incident response: detect, contain, eradicate, recover, and document; preserve evidence before remediation.