IT support and hardware, oriented

Start from facts. Record the user's own description of the problem, then try to make it happen in front of you, because a fault you cannot reproduce is one you cannot prove you fixed. Intermittent faults are the hard case. Navy electronics training from 1998 warned that opening a case can be enough to make a heat-related fault vanish, so bench technicians warmed a suspect board gently until the fault returned.

Narrow the cause by swapping a suspect part for one you know works, testing after each swap. Two precautions come first. If the repair could touch data, confirm a current backup exists and actually restores. Before you handle a part, ground yourself with a corded wrist strap: you only feel a static shock at thousands of volts, yet a discharge of tens of volts can damage a chip with no visible sign. A NASA working group tested wireless wrist straps and found they did not stop charge from building up.

The trap is replacing several parts at once to end the call sooner. Say a desktop reboots at random, and you swap the memory, the power supply and the drive cable together. The reboots stop, but you cannot say which part was bad, and you may have spent parts you did not need.

When it works, test the whole system, since one fixed fault can hide a second. Then write the ticket: the symptom, each change and the result, in words the user could follow. Report a recurring complaint to the people above you rather than fixing it quietly each time.

Rule: change one thing, test it, write it down, and touch nothing until the data and the static are handled.

Treat the motherboard as the contract every other part has to sign. Its form factor names its size and layout, which decides the cases it fits. The processor mounts in a socket, and the chipset beside it governs how the processor trades data with memory, drives and cards. The chipset even caps how much memory the board can take, so you choose a processor for a particular board, never on its own.

Three levers set a processor's speed. Cores run separate work side by side, multithreading uses a core's dead time while data is fetched, and clock speed sets how quickly instructions are processed.

Memory has to match twice. Each double data rate (DDR) generation doubled transfer rates and lowered signaling voltage, so modules do not cross generations, and laptops take the compact small outline dual in-line memory module (SO-DIMM). Servers often add error-correcting code (ECC) memory, and a common scheme uses extra check bits to correct a single flipped bit and detect two.

Expansion cards sit in Peripheral Component Interconnect Express (PCIe) slots, and a slot performs only as well as the lanes wired to it. A board can carry more slot pins than lanes it can drive at once.

The trap is judging a slot by its length. Suppose you move a graphics card into a second full-length slot to make room for a cooler, and its performance drops. That slot may be wired with fewer lanes, so the card cannot reach its rated speed. Installing a drive has a matching check: a serial Advanced Technology Attachment (SATA) drive needs both a data cable and a power cable before the system can see it.

Rule: read what the board supports, the socket, chipset, memory generation and lanes, before you trust any part's own rating.

Power-on self-test comes first. The firmware tests the chipset, processor and memory, and only afterward brings up the video card. A fault caught that early happens before anything can be drawn, so a black screen with the fans running can still be memory or the board.

Next the firmware walks its boot list in the order kept in its settings and starts the first device carrying a valid boot record. A legacy Basic Input/Output System (BIOS) looks for a master boot record, while Unified Extensible Firmware Interface (UEFI) firmware also accepts a newer partition format. A "no boot device" error means that search found nothing usable, while a drive that is found but fails to start points at the loader or the operating system. If a machine forgets its date and settings whenever it is unplugged, suspect the coin cell that keeps the firmware's clock running.

Safe Mode loads Windows with only a few services, so a fault that disappears there likely comes from something that loads at startup. Security tools may not run there either, which is why some ransomware reboots machines into it. Rolling back an update fits a fault that began right after updating. For firmware, 2011 guidelines from the National Institute of Standards and Technology (NIST) warn that an older version can carry known vulnerabilities, so authenticated systems may demand a special process.

The trap is condemning the internal drive at the first boot error. Say a desktop starts an old installer instead of its usual system. A USB stick left in a port carries its own boot record and sits ahead of the internal disk in the list, so unplug it and retry.

Rule: name the last link in the chain that worked, then test the next one.

A hard drive reads magnetic platters with moving parts, while an SSD stores data in flash memory with none, which is why swapping one for the other is such a visible speedup. How a drive connects matters too. SATA was designed for spinning disks and cannot keep up with a fast SSD, so non-volatile memory express (NVMe) was created to talk to storage over PCIe. Workstation boards usually carry small M.2 slots for such drives, each capped by the lanes wired to it.

A Redundant Array of Independent Disks (RAID) spreads data across several drives. Striping alone gains speed and keeps no redundancy, so one failed disk loses everything. RAID 1 mirrors the data onto a second disk, RAID 5 stores parity that can rebuild a lost disk, and RAID 6 keeps two sets. A mirror or single-parity array that has lost a disk keeps running in a degraded state, with no safety margin left until the rebuild finishes.

Drives can warn before they die. Self Monitoring, Analysis and Reporting Technology (SMART) tracks attributes, some of which correlate strongly with coming failure. When they climb, treat the drive as failing and copy the data off first. Before you blame a slow drive at all, rule out the processor, since malware or a cryptocurrency miner can make a whole system feel slow.

The trap is treating RAID as a backup. Say a small office server mirrors two disks, and one fails without anyone noticing. Every file now depends on a single disk, and the array was never protection against a mistake, malware, theft or a disaster in that one location.

Rule: when a drive or an array warns you, copy the data somewhere else before you repair anything.

Power in watts is voltage times current, and a rating is a ceiling, not the normal draw. Size a supply with room above the load rather than matching it exactly. A surge protector guards against spikes and dips in line power. An uninterruptible power supply (UPS) adds a battery that carries equipment through a short outage long enough to save work and shut down, so computers and critical network gear go on the UPS.

Inspect before you switch on. Navy power-supply training from 1998, written for older designs, lists what to look for: loose, broken or corroded connections, stray screws or scraps of wire that could short something, and an electrolytic capacitor with a whitish deposit at its seal. Once power is on, smoke or boiling and sputtering sounds mean you remove the power at once.

On the bench, a meter settles what your eyes cannot. A voltmeter connects across the two points you are comparing, in parallel, so you can read a supply's output while it runs. An ammeter measures current, so it goes in series, in the current's path.

The trap is assuming that unplugged means safe. Say you unplug a dead desktop supply and reach in to check a part. Navy electronics training from 1998 warned that some capacitors keep their charge long after power is gone, and that what remains in many circuits can kill. That training has technicians discharge high-voltage capacitors only with an authorized shorting probe, several times over, before any work begins. A lithium-ion battery is the same story in another form: even a used one can hold enough energy to injure you or start a fire.

Rule: treat any stored energy, in a capacitor or a battery, as live until someone trained has made it safe.

Identify the port at each end before you choose a cable or an adapter. Two of the monitors compared in one study shared no video input type at all, so a cable that fit one would not have fit the other. Refresh rate, the number of redraws per second in hertz, helps only if the whole path can deliver it.

Isolate a display fault by substitution. If the defect follows the monitor to another computer, it lives in the monitor. If it stays with the original computer when a different monitor is attached, suspect the graphics hardware or the cable, then swap the cable alone to separate the two.

Laser printing runs charge, expose, develop, transfer, fuse and clean, in that order. A scratched drum leaves streaks. Fusing bonds toner with heat and pressure, so toner that rubs off the page points at the fuser. When a double-sided job jams, note exactly where the sheet stops, because that names the stage to inspect. Other printers fail differently: inkjet nozzles clog as ink dries, and heat-printed receipts can darken if they are stored somewhere warm.

The trap is reaching for a new toner cartridge for every quality problem. Say you rub a fresh page and the toner smears onto your thumb. A new cartridge changes nothing, because the fuser is not bonding the toner.

When jobs stop moving on Windows, check the Print Spooler service first. It is also an attack surface: after a critical flaw in 2021, the Cybersecurity and Infrastructure Security Agency (CISA) advised turning it off on systems that do not print.

Rule: name the stage that could produce the defect, then inspect or swap only that stage.

What you can upgrade depends on what can come out. Adding memory is often the cheapest upgrade, but some small systems build their storage onto the board, and phones keep theirs there by design. Not every battery is meant to be removed by the user. A swollen lithium-ion battery is a potential fire hazard, so cover its terminals, keep it away from anything flammable and take it to a battery recycler, never the trash.

Screens fail in layers. A touchscreen stacks a display on a touch layer, so a clear image that ignores touch points at the touch layer. A liquid crystal panel makes no light of its own; if the image is faintly visible under a strong light but otherwise dark, suspect the backlight. Many liquid crystal displays from 2000 to 2010 were lit by a tube driven by a high-voltage inverter, a now-dated design, and panels whose pixels make their own light need neither. Slowness under load can be cooling too: thermal paste fills the gaps between processor and heat sink, and a hot processor cuts its own clock.

Each USB generation raised the ceiling, and each stays backward compatible, so a fast device still works on an older port, but only at the older speed. USB Power Delivery lets a charger and a device agree on a higher voltage, which is how one charger serves a phone and a laptop.

The trap is judging a port by its shape. Say an external drive copies slowly over USB-C. One hub's list rated a Thunderbolt 3 port at 40 Gbps beside a same-shaped USB-C port at 5 Gbps, and a look-alike cable can limit the link too.

Rule: speed and power fall to the weakest part of the link, so check the port, the cable and the device.

Two devices talk directly only when the subnet mask puts their addresses on the same network, and traffic for anywhere else goes to the default gateway. With an address outside the local range, a device loses even its neighbors. With only the gateway misconfigured, the neighbors still answer and nothing past them does. Dynamic Host Configuration Protocol (DHCP) hands out an address as a lease. A static address typed in by hand must fit the network's range, mask and gateway.

Names work through the Domain Name System (DNS). A 2013 NIST guide, withdrawn in March 2026, describes a resolving server that answers from its cache or asks onward, with each record setting how long a copy may be kept. So after an entry changes, a client can keep using the old address until its copy expires. That guide also notes DHCP can name the resolvers a device uses, and warned of malware that swaps them, so check which servers a device is really using.

Slow has two meanings: bandwidth, the data a link moves each second, and latency, the time one packet takes to cross. A link can have plenty of the first and still delay every packet, so test both before you call a connection slow.

The trap is reading every outage as the same fault. Say you move a desktop with a hand-typed address to another office, and now it reaches nothing. The cable and the switch may be fine: its old address no longer fits the new network's range. A reply to ping, the basic reachability test, proves a path works. No reply proves much less, because a host can be up and stay quiet.

Rule: test the path in order, address, then gateway, then name, with each step proven before you move to the next.

Know which box does which job. The modem speaks your provider's medium. The router is where the Wi-Fi, the built-in firewall and the list of DHCP clients usually live. Connection types differ in delay, not just speed. In measurements from a 2024 Federal Communications Commission (FCC) report, fiber had the lowest median latency, cable sat in the middle and digital subscriber line (DSL) service was highest. Geostationary satellites orbit about 22,300 miles up, so the long trip can impair calls.

A new router needs three things first: a changed admin password, newer firmware from the maker and a renamed network. CISA's tips favor the newest Wi-Fi encryption your equipment supports, and the old Wired Equivalent Privacy (WEP) is no longer considered secure. Port forwarding deliberately opens a path to one inside device. CISA's home-network tips advise disabling Universal Plug and Play (UPnP) unless you need it, because malware inside the network can use it to get past the router's firewall. Smart devices get the same care: replace their default passwords during setup, keep their firmware current, and hold them apart from your main devices, a guest network being the simple way, so one compromised gadget stays contained.

Wi-Fi problems call for measurement. Signal strength swings with movement and walls, so take several readings. A device can cling to a weak access point even when a stronger one is nearby. In one study, microwave ovens cut throughput by 28 to 49 percent at 1 to 3 m, though another study found no effect on throughput.

The trap is buying hardware before moving the router. Say a home office in a far bedroom has weak Wi-Fi. A central router may fix it outright, and a cable gives that desk the highest speeds.

Rule: change the defaults first, then measure the radio before you buy anything.

Typical signs of malware are a slow or crashing computer, a home page that changes or redirects, new toolbars or add-ons, and system tools that are suddenly disabled. Sustained high processor use and heat with no obvious cause can mean someone is mining cryptocurrency on the machine. Respond in order: stop signing in to accounts from the device, update the security software and scan, then change passwords and turn on two-factor authentication. Security software recognizes known code, so a brand-new threat can slip past it.

Theft needs its own answer. Encrypting the whole disk makes a lost laptop useless to anyone who cannot supply the password or PIN. That lock makes no exception for you: lose what opens it and the data stays shut, so store a way back in safely, apart from the laptop, before you need it.

Phones have their own controls. CISA's guidance for high-risk mobile users is to keep Android's Google Play Protect scanning on and to be cautious with sideloading, installing an app from outside the official store. Revoke permissions an app does not need, especially location, camera and microphone. Turn on automatic updates and install them only through the device's own updater or the vendor's site, never a link in an email. Anyone enrolling a personal phone in mobile device management should know it allows a remote wipe.

The trap is reading quiet logs as no attack. Lockout policies commonly allow three to five bad attempts, so an attacker sprays one password across many accounts and stays under the limit. Say you see no lockouts at all this week: that proves nothing. A 2018 CISA alert lists multifactor authentication first among its defenses.

Rule: contain first, clean second, then assume the credentials were taken.

A hypervisor runs virtual machines and keeps them isolated while they share the host's hardware. A Type 1 hypervisor installs directly on the hardware. A Type 2 runs on top of an ordinary operating system, which then has to be trusted and patched along with it. Hardware-assisted virtualization gives the hypervisor a more privileged processor mode, which cuts the code it needs and strengthens isolation.

Plan virtual machines against the physical host. NIST's 2018 hypervisor recommendations call for provisioning that stays within a key physical resource such as processor cores, and for memory over-commit, which lets the machines' total memory exceed the host's RAM, to be off by default.

The cloud models split the work. With infrastructure as a service (IaaS), you get processing, storage and networks and run your own systems. With platform as a service (PaaS), you deploy your own applications on the provider's platform. With software as a service (SaaS), you use the provider's applications and manage nothing underneath. Thin clients and virtual desktops push the same idea to the desk: little computing happens locally, so the network and the servers become the real dependency.

The trap is treating a virtual processor as free capacity. Say a host has eight cores and you guarantee each of four virtual machines the full time of four cores. Those guarantees add up to sixteen cores, twice what the hardware offers, so they cannot all be honored once the guests get busy.

Builds for one job follow the same logic. A storage server is built around its disks, with a mirror or parity inside the box and a copy kept somewhere else.

Rule: size a host by what you have promised its guests, and plan for the network whenever the computing lives elsewhere.