Study guide · Core notes
Reference notes for racking and storage, operating-system administration, backups and hardening, and the troubleshooting order.
How to read these notes: Items tagged Extra sit outside the SK0-005 objective bullets, such as the retired claim that Server+ never expires. Learn the unmarked items first. New to the exam? Start with the overview for the format, the 750 passing score, and the October 1, 2025 continuing-education change. Hardware names this exam assumes are on the A+ notes.
Domain 1 is 18 percent. The rack, the power path, and the chassis come before the operating system.
| Item | What it decides |
|---|---|
| Rack unit | 1U is 1.75 inches of height. A 2U server is twice that. Check floor load and keep heavy equipment low so the rack does not tip. |
| PDU | Distributes power inside the rack. It is not a battery. A UPS is the battery that bridges a short outage. |
| Separate circuits | Redundant supplies should land on different breakers, and ideally different providers. One tripped breaker should not darken both supplies. |
| Chassis | A tower stands alone. A rack server uses rails. A blade shares power and networking with the enclosure. |
| Fiber | Single-mode runs farther. Multimode is the shorter run. LC is the small connector on an SFP. SC is larger. SFP+ is the common 10 Gigabit module. QSFP carries more lanes. |
Check the hardware compatibility list before you buy a controller the operating system will not drive. CPUs, memory, and expansion cards have to fit the board and the heat budget. A GPU is still a card that needs power and airflow. Cable management is part of the install: a cord in a fan is a cooling failure you caused.
An SSD wears from writes. A write-intensive disk belongs on a log or a database volume. A read-intensive disk belongs on a volume that is mostly read. Spinning disks are named by RPM: 15,000 and 10,000 for faster access, 7,200 for capacity. A hybrid drive caches on flash and stores on a platter. It is not a full SSD.
SAS and SATA are the internal disk links you will see most. eSATA is SATA outside the chassis. USB and SD show up as install or boot media. PCI is the expansion bus family. Do not call a SAN protocol a disk cable.
Name the failure the layout survives before you name the vendor.
| Layout | What you can lose |
|---|---|
| RAID 0 | Striping, no parity. One disk failure loses the set. Capacity is the sum of the disks. |
| RAID 1 | A mirror. One copy can fail. Usable capacity is one disk of the pair. |
| RAID 5 | Distributed single parity. Survives one disk. Needs at least three. A second failure during the rebuild loses the array. |
| RAID 6 | Dual parity. Survives two disks. Needs at least four. The write penalty is higher than RAID 5. |
| RAID 10 | A stripe of mirrors. Survives a failure inside each mirror. Needs at least four disks. It is not dual parity. |
| JBOD | Just a bunch of disks. No RAID redundancy. Losing a disk loses the data on that disk. |
A hardware controller can cache writes. If its cache battery fails, the controller often turns that cache off and writes get slow. The array is not automatically destroyed. Software RAID, such as a volume the operating system manages, does not depend on that battery. A failed hardware controller can hide the array from a different server.
NAS shares files. NFS and CIFS (the SMB family) are the file protocols on the objectives. SAN presents block storage. iSCSI carries SCSI over TCP/IP. Fibre Channel uses its own fabric. FCoE puts Fibre Channel frames on Ethernet. A file share is not a LUN, and a LUN is not a folder permission.
Capacity planning asks whether the volume, the IOPS, and the rack will still fit the workload later. A full disk is an operations problem. It is not a backup. Deduplication and compression save space. They do not replace an off-site copy.
Out-of-band management exists so you can work when the operating system cannot.
iLO, iDRAC, IPMI, and a CIMC reach power, a remote console, and virtual media on a path that does not need the server operating system. An IP KVM does the keyboard and screen over the network. SSH and Remote Desktop need a running OS. If the host will not POST, those sessions are the wrong tool.
A crash cart is a monitor and keyboard you roll to the rack when remote management is down. A serial console is the last path when the network management port is dead. Firmware updates belong in a change window. A bad firmware flash can be the outage.
Drives, cages, some cards, power supplies, and fans are the hot-swappable parts on the objectives. You replace them while the server stays up. A hot-swap bay is not a hot spare until the controller is told to use that disk. The CPU and the motherboard are not hot-swap parts.
UEFI is the firmware that boots a GPT disk. BIOS is the older firmware. A firmware password and the boot order are hardening settings as well as maintenance settings. A dead CMOS battery forgets the clock and the stored settings.
Domain 2 is 30 percent. The largest share of the exam starts here.
| Name | Use it when |
|---|---|
| GUI and Server Core | A GUI install includes the desktop. A core install does not. Core is smaller. It is still a full operating system. |
| Bare metal and virtual | Bare metal is the hardware itself. A virtual install is a guest. P2V converts a physical machine into a guest. A template or a clone copies a known image. |
| Unattended | A slipstreamed or scripted install answers the prompts and can add drivers. It is not a different operating system. |
| GPT and MBR | GPT with UEFI for large disks. MBR stops near 2 TB and has four primary partitions. Dynamic disks and LVM are how Windows and Linux grow or span volumes after that choice. |
| ext4, NTFS, ReFS, ZFS, VMFS | ext4 is Linux. NTFS and ReFS are Windows. ZFS is the checksum-and-pool filesystem. VMFS is a VMware datastore. Do not format a VMware datastore as NTFS and expect the hypervisor to use it. |
Most "the server is down" tickets are a name, a route, or a VLAN.
RFC 1918 is the private IPv4 space: 10.0.0.0/8, 172.16.0.0/12, and 192.168.0.0/16. IPv6 is the other addressing protocol on the objectives. A static address stays put. DHCP assigns one. APIPA, 169.254.0.0/16, means no DHCP server answered. A wrong scope is different: a server did answer, and the options were wrong.
DNS turns a fully qualified domain name into an address. The hosts file can override that on one machine. A stale hosts entry sends the name to an old address even when DNS is right. The default gateway is the way off the subnet. A missing gateway looks like "the Internet is down" while local servers still answer.
A VLAN is a broadcast domain on a switch. The server NIC and the switch port have to agree. A host firewall drops traffic to a port you forgot to open. The service can be running and still be unreachable.
Print, file, web, database, application, and messaging are the server roles on the objectives. A baseline is the normal performance you compare later alerts with. Directory connectivity is how the host joins an identity store. Page, swap, and scratch space are disk used as memory overflow. They are not backups.
Two powered-on servers are not a cluster until they can see each other fail.
| Term | What it does |
|---|---|
| Active-active | More than one node serves the workload at the same time. |
| Active-passive | One node serves. Another waits. Failover moves the service. Failback returns it after the original node is healthy. |
| Heartbeat | The check that tells the cluster a node is gone. Without it, both nodes may try to run, or neither will. |
| Round robin and MRU | Round robin rotates through servers. Most-recently-used sends work back to the server that just had it. |
| NIC teaming | More than one NIC acts as one logical connection, for failover or for combined links. Link aggregation is the combined-bandwidth case. |
| Host and guest | The host runs the hypervisor. The guest is the virtual machine. Bridged networking puts the guest on the physical LAN. NAT hides it behind the host. |
| Overprovisioning | You assign more virtual CPU or memory than the host physically has, because the guests do not all peak together. It fails when they do. |
Cloud models on the objectives are public, private, and hybrid. Public is someone else's infrastructure. Private is yours. Hybrid uses both. Patch one cluster node at a time so the heartbeat still has a survivor.
The exam asks you to recognize the construct, not to write a long program.
Bash, PowerShell, batch, and VBS are the script types. A loop repeats. A conditional chooses. Comments exist so the next person can read the change. Environment variables come from outside the script. Common jobs are startup, shutdown, a service action, logon, account creation, and bootstrap.
Label the asset, record the serial number and the asset tag, and know whether it is leased or owned. Keep architecture diagrams, baselines, and the change record current. Store sensitive runbooks where the whole floor cannot read them. MTBF is the average time between failures. MTTR is the average time to recover. An SLA is the uptime you agreed to deliver. A business impact analysis says what the outage costs.
Per-socket counts populated CPU sockets. Per-core counts cores. Per-server and per-instance count machines or installations. Per-concurrent-user counts people connected at once. A node-locked license is tied to one machine. A true-up reconciles what you run with what you bought. A subscription is not the same thing as support on a license you already own. Open source still has license terms.
Robocopy is the Windows copy tool that can retry and mirror a tree. SCP copies over SSH. Infiltration and exfiltration, in this domain, mean data moving into or out of the environment during a migration. They are not, by themselves, an attack story.
Domain 3 is 24 percent. It shares that weight with backups and disaster recovery.
Encryption at rest protects stored data. Encryption in transit protects the path. A BIOS or UEFI password and a bootloader password stop someone who has the console. Boot order stops a USB stick from becoming the boot device. Retention says how long you keep data. Off-site storage is a different building, not a different shelf in the same room.
Bollards stop vehicles. A mantrap (an access-control vestibule) admits one person between two doors. Badges, biometrics, and RFID readers decide who passes. Cameras and guards detect. They do not replace a lock. HVAC and fire suppression are environmental controls. A temperature alarm is a sensor, not a backup.
Role-based access follows the job. Rule-based access follows conditions. Segregation of duties means the person who requests a change should not be the only person who approves it. Delegation hands out a limited admin right. Multifactor authentication needs two different factor types: something you know, have, or are. Two passwords are still one factor. Single sign-on is one authentication for many applications. Audit logons, group changes, and deletions.
Disable unused services, close unused ports, patch through change management, and run anti-malware plus a host intrusion system. Disable unused USB ports and set the boot order. Before disposal, confirm the server is unused, update the asset record, and destroy the media. Wiping overwrites. Degaussing clears magnetic disks and tape. Shredding, crushing, or incineration destroys the device. Degaussing is the wrong sole method for an SSD. Remove power and network cables as part of cable remediation. Recycling is not the same as handing the disks to a buyer.
The restore is the point. A copy you have never restored is a hope.
| Method | What the restore needs |
|---|---|
| Full | Everything selected, in one set. The restore can start from that set alone. |
| Incremental | Changes since the last backup of any type. Restore the full, then every incremental since, in order. |
| Differential | Changes since the last full. Restore the full and the latest differential. |
| Synthetic full | A new full built from an older full plus later incrementals, without reading production again. |
| Snapshot | A point in time, often on the same storage. If that array dies, the snapshot dies with it. It is not an off-site backup by itself. |
| System state | Boot files, the registry, and, on a domain controller, the directory database. A copy of user documents is a file-level backup, not system state. |
Media on the objectives include tape, disk, cloud, and print. Open-file backup captures files that are in use. Test restores on a schedule. Restore options are overwrite, side by side, and an alternate path. Inventory the media before you start a restore so you are not missing the tape in the middle.
Pick the site and the replication mode from the objective the business wrote down.
| Term | Meaning |
|---|---|
| RPO | Recovery point objective. How much data, in time, you can afford to lose. Synchronous replication aims at a very small RPO. |
| RTO | Recovery time objective. How long you can wait until the service is back. A hot site supports a short RTO and costs the most. |
| Hot, warm, cold | Hot is ready to run. Warm has some equipment and still needs work. Cold is space, power, and a plan. Cloud can be the alternate site. Geography should not be the same flood plain. |
| Synchronous and asynchronous | Synchronous waits until both sides have the write. Asynchronous acknowledges locally, then replicates. Async can lose more data and can span a longer distance. |
| Application consistent | The application is quiesced so a database backup can be restored cleanly. A crash-consistent copy is closer to pulling the power cord. |
| Tests | A tabletop is a discussion. A simulated failover stops short of moving production. A live failover actually moves it. Test in non-production before you promise a live cutover. |
Domain 4 is 28 percent. The order is the question, even when the symptom is dramatic.
| Step | What you do |
|---|---|
| 1. Identify | Ask what changed, read the logs, and decide the scope. Reproduce it if you can. Back up before you change a working system. Escalate when the impact or the skill exceeds you. |
| 2. Theory | Name a probable cause, including the obvious one. Look for one change that explains several symptoms. |
| 3. Test | Confirm or discard the theory. If it fails, build a new theory. Do not keep replacing parts to see what happens. |
| 4. Plan | Decide the change and tell the people it will affect. |
| 5. Implement | Make one change and test it. If it fails, reverse it when you can, then try the next change. |
| 6. Verify | Confirm the whole service, not only the light you were staring at. Add a preventive step when one exists. |
| 7. Root cause | After the service is back, ask why it failed. The root-cause write-up is not a substitute for restoring service. |
| 8. Document | Findings, actions, and outcomes throughout, not only a sentence at the end. |
Match the screen, the log, and the layer. Then use the tool that sees that layer.
POST runs before the operating system. A blue screen is Windows. A purple screen is a VMware ESXi crash. A kernel panic is Linux. A dead CMOS battery loses the clock and the firmware settings. Dust, heat, a loose heat sink, and a failed fan are cooling problems. Reseat a card before you condemn it. Use ESD protection. A predictive failure warning means replace the part before it dies, especially a disk in RAID 5.
A failed cache battery or a cache turned off makes writes slow. A rebuild makes the array busy and, on RAID 5, one more disk failure away from data loss. An HBA or a backplane failure hides disks. A corrupt boot sector or the wrong boot order produces "operating system not found," which is not a POST memory error. Mount and disk-management tools show whether the OS can see the volume. net use maps a Windows share.
Clock skew stops a domain logon even with the right password. NTP is the fix. A memory leak grows until you restart the process. A bad patch is reversed by uninstalling it or booting a previous kernel, safe mode, or single user mode. sudo runs one command with elevation. su switches user. runas is the Windows form. End of support means the vendor will not patch the flaw. Group Policy, SCCM, and configuration tools such as Ansible, Puppet, and Chef are how a fleet stays consistent.
No link light is a cable, a port, or power. APIPA means DHCP did not answer. "Unknown host" is DNS. "Destination unreachable" is routing. A hosts file can lie. ipconfig and ip addr show the address. ping tests reachability. tracert and traceroute show the path. nslookup and dig ask DNS. SELinux can deny a file that Unix permissions allow. Do not answer that with permissions of 777. Data-loss prevention can block a copy that the file ACL would allow.
No. They are an original summary of the public SK0-005 objectives, version 2.0. Confirm the current PDF with CompTIA before you book.
Extra marks a useful idea the SK0-005 objective bullets do not name, such as the older claim that Server+ never expires. Learn the unmarked items first.
RAID 6, which keeps two parity blocks. RAID 5 survives one failure. A second disk failure during a RAID 5 rebuild loses the array.
Use the flashcards to check whether these distinctions stick.
Exam format, the 750 passing score, four domain weights, and the October 2025 continuing-education change.
Reference notes for racking and RAID, administration, backups, and the troubleshooting order.
40 recall cards in four decks, plus memory notes for RAID, backups, and recovery sites.
200 original questions with custom exams, explanations, and an unofficial score on the 100-900 scale.
Useful companions while you study.
Generate strong random passwords with custom length, characters, symbols, and security options.
Convert length, weight, volume, temperature, area, speed, pressure, and other measurements.
Calculate elapsed time, add or subtract hours and minutes, and convert time to decimal hours.
Compare two text blocks and highlight added, removed, and changed content.
Certicab provides general informational, educational, and convenience resources. Calculations, conversions, estimates, and learning materials may contain errors or become outdated. Financial, tax, medical, legal, and travel information is not professional advice. Verify important results and current requirements with qualified professionals or authoritative sources. Protect sensitive files and personal information, review each tool’s privacy limitations, and use only content you are authorized to process. You are responsible for how you use and share results. Study resources are independent and do not guarantee exam success or imply certification-provider endorsement. CompTIA and Server+ are trademarks of CompTIA, Inc. Certicab is not affiliated with, endorsed by, or sponsored by CompTIA.
Certi