AP Networking Network Monitoring and Performance Practice Test

1. The highest speed printed on a box
2. A copy of the DNS zone
3. A measured reference for comparable normal operation
4. One universal acceptable RTT

Correct answer: C. A measured reference for comparable normal operation

A baseline records expected behavior under stated conditions. It supports comparison; it is not a guarantee that every workload or time period behaves identically.

1. The maximum DNS TTL
2. The achieved transfer rate under the measurement conditions
3. The cable's color
4. The number of configured users

Correct answer: B. The achieved transfer rate under the measurement conditions

Throughput is measured delivery per unit time. It can be lower than nominal link capacity because of workload, overhead, contention, or endpoint limits.

1. The server's disk capacity
2. Time for a tested exchange to travel out and back, including relevant response delay
3. The size of the subnet
4. One-way delay automatically

Correct answer: B. Time for a tested exchange to travel out and back, including relevant response delay

RTT combines both directions and response behavior. It cannot be divided into a reliable one-way measure without additional assumptions.

1. The cable's maximum power
2. The total number of IP addresses
3. A DNS record type
4. Variation in packet delay

Correct answer: D. Variation in packet delay

Variable delay can disrupt real-time delivery even when average delay looks reasonable. The measurement definition and sample conditions should be stated.

1. CPU model number
2. VLAN ID
3. Packet loss rate
4. DNS label length

Correct answer: C. Packet loss rate

Loss rate requires a defined sent/received population and observation window. A percentage without its denominator and scope is hard to interpret.

1. Workload changes only the MAC address
2. A baseline replaces all testing
3. All workloads produce equal traffic
4. Changing demand can change the result without a configuration fault

Correct answer: D. Changing demand can change the result without a configuration fault

A fair comparison controls source, destination, protocol, load, and time context. Otherwise a difference may reflect changed demand.

1. The interface's hostname alone
2. Its total divided by the subnet prefix
3. Its change over a known interval
4. Only its largest lifetime value

Correct answer: C. Its change over a known interval

A cumulative total is not a current rate. Use differences between valid samples and account for resets or wraparound.

1. The percentage of valid DNS answers
2. Observed traffic rate relative to the relevant interface capacity
3. The server's storage fullness
4. The number of browser tabs

Correct answer: B. Observed traffic rate relative to the relevant interface capacity

State direction, interval, and capacity denominator. An aggregate or average can hide short bursts and directional pressure.

1. Send and receive capacity and demand are separate
2. Full duplex means only one direction works
3. Both directions always have equal traffic
4. A packet uses no reverse path

Correct answer: A. Send and receive capacity and demand are separate

A full-duplex link can transmit and receive simultaneously. One directional average does not characterize the other direction's load or queue behavior.

1. Flag a defined condition for investigation or action
2. Guarantee no false alarms
3. Prove the root cause automatically
4. Replace the service requirement

Correct answer: A. Flag a defined condition for investigation or action

Thresholds depend on the metric and operational context. A crossing is evidence of a condition, not a complete diagnosis.

1. Every student's web application
2. All DNS answers
3. That probe's echo path to that gateway at that time
4. Every server login

Correct answer: C. That probe's echo path to that gateway at that time

Probe location, target, and protocol limit its scope. A passing infrastructure check can coexist with an application-specific failure.

1. Counters always reveal page correctness
2. Only routers affect service availability
3. Application tests replace all interface measurements
4. Healthy devices do not necessarily imply successful service use

Correct answer: D. Healthy devices do not necessarily imply successful service use

A synthetic or real task checks dependencies closer to the user's goal. Infrastructure telemetry helps explain a task failure but does not replace task verification.

1. A new DNS lease
2. A misleading negative delta or implausible rate
3. The exact network packet loss
4. Guaranteed increased throughput

Correct answer: B. A misleading negative delta or implausible rate

A discontinuity invalidates an ordinary sample subtraction. Track restart/discontinuity state and establish a new measurement interval.

1. The subnet's maximum hosts
2. A position in an ordered set of measured delays
3. The speed of every packet
4. The number of server processes

Correct answer: B. A position in an ordered set of measured delays

A percentile describes a distribution under a defined sample method. It is not necessarily the same as the average or maximum.

1. To assign public addresses
2. To encrypt every packet
3. To correlate changes and failures across observations
4. To stop TCP retransmission

Correct answer: C. To correlate changes and failures across observations

Time context allows comparison with workload, maintenance, and failures. Clock consistency and timezone interpretation matter.

1. The application completed successfully
2. The DNS server is necessarily offline
3. Every discard was caused by a bad cable
4. Packets were discarded at the measured queue under its defined semantics

Correct answer: D. Packets were discarded at the measured queue under its defined semantics

Queue drops are processing evidence at that observation point. Determine load, queue policy, and the affected traffic before assigning a cause.

1. Only the site's page title
2. Only the application password
3. Only the DNS TTL
4. The defined error type, trend, physical/link conditions, and affected path

Correct answer: D. The defined error type, trend, physical/link conditions, and affected path

Different error counters describe different events. Interpret the platform's definition and current rate rather than treating every lifetime total as an active fault.

1. Short outages never affect users
2. Monitoring intervals change the subnet
3. Brief spikes can be hidden inside the average
4. Averages contain every instant exactly

Correct answer: C. Brief spikes can be hidden inside the average

Sampling and aggregation can smooth peaks. Use finer evidence or events when the symptom lasts less than the collection interval.

1. The resource, condition, time, scope, and response context
2. Only something is wrong
3. Only a color with no metric
4. Only the router brand

Correct answer: A. The resource, condition, time, scope, and response context

An actionable alert lets a technician identify the affected service and decide what evidence to inspect next.

1. A duplicate MAC address
2. An automated test of a defined service action
3. An Ethernet trunk
4. A NAT port allocation

Correct answer: B. An automated test of a defined service action

A synthetic test can exercise a login or request path under a known identity and source. Its success still has the scope of that configured task.

1. Whether the link rate is zero
2. Loss of telemetry versus failure of the monitored service
3. Whether all clients must be offline
4. Whether every packet has been lost

Correct answer: B. Loss of telemetry versus failure of the monitored service

An agent, collector path, credentials, or schedule can fail while the service continues. Investigate monitoring health independently.

1. It proves every Internet service works
2. It removes all endpoint limits
3. It measures every router's CPU
4. It helps compare local capacity with the remote path

Correct answer: D. It helps compare local capacity with the remote path

A local test can isolate part of the path. It still depends on local endpoints and workload, so state what was held comparable.

1. The service, measurement window, success definition, and exclusions
2. Only the largest port number
3. Only a device's power rating
4. Only a DNS suffix

Correct answer: A. The service, measurement window, success definition, and exclusions

Availability is meaningful only when the measured service and rules are explicit. Device uptime and user-task availability may differ.

1. To prove every later fault was caused by that change
2. To replace all packet traces
3. To increase public address capacity
4. To investigate temporal relationships with behavior changes

Correct answer: D. To investigate temporal relationships with behavior changes

A timeline supplies leads for testing. Correlation with a change is not by itself proof of causation.

1. Only when the link is fiber
2. Yes; latency measures authorization
3. Yes; all fast responses are healthy
4. No; success status and task outcome must also be measured

Correct answer: D. No; success status and task outcome must also be measured

An immediate failure can have excellent latency. Measure correctness and availability along with timing.

1. 60 Mb/s
2. 48 Mb/s
3. 480 Mb/s
4. 6 Mb/s

Correct answer: B. 48 Mb/s

Convert bytes to bits and divide by time: 60,000,000 times 8 divided by 10 equals 48,000,000 bits per second. Under the stated capacity this is about 48% average utilization.

1. 75%
2. 150%
3. 25%
4. 1.33%

Correct answer: A. 75%

Divide measured rate by capacity: 150/200 equals 0.75. The result says nothing by itself about the opposite direction or short peaks.

1. 100 Mb/s
2. 8 Mb/s
3. 800 Mb/s
4. 12.5 Mb/s

Correct answer: A. 100 Mb/s

100 million bytes is 800 million bits; dividing by 8 seconds gives 100 million bits per second. State decimal units and payload scope to avoid byte/bit confusion.

1. The DHCP lease time
2. The effective 100 Mb/s link capacity
3. The server disk size
4. The marketing maximum 1 Gb/s regardless of negotiation

Correct answer: B. The effective 100 Mb/s link capacity

Current negotiated capacity controls the relevant comparison. Using the rated maximum can hide saturation on the actual link.

1. Short demand peaks can exceed the queue/service capacity despite a low long average
2. The link never experienced pressure
3. The queue data must be impossible
4. The DNS resolver necessarily caused the burst

Correct answer: A. Short demand peaks can exceed the queue/service capacity despite a low long average

Different time scales reveal different behavior. Examine burst demand and queue policy rather than dismissing the symptom from the aggregate average.

1. Every request took 50 ms
2. The nominal link rate is 1800 Mb/s
3. Exactly 95% of requests failed
4. The slow tail of the measured requests

Correct answer: D. The slow tail of the measured requests

A low median can coexist with a substantial slow tail. Examine which requests, sources, or periods populate that tail.

1. Interpret the negative delta as negative traffic
2. Treat the older total as current demand
3. Mark the interval as discontinuous and begin a valid new interval
4. Add an arbitrary subnet-sized constant

Correct answer: C. Mark the interval as discontinuous and begin a valid new interval

A reset breaks normal delta calculation. Use uptime/discontinuity evidence and collect comparable samples after the reset.

1. Use ordinary negative subtraction
2. Assume the interface stopped sending
3. Account for the wrap and ensure the interval assumptions rule out ambiguous multiple wraps
4. Use the device hostname as the delta

Correct answer: C. Account for the wrap and ensure the interval assumptions rule out ambiguous multiple wraps

Modulo correction can handle a known single wrap. If several wraps are possible, the old samples may be insufficient; higher-capacity counters or shorter intervals help.

1. The tested gateway path is steady; other dependencies still need measurement
2. The whole application network is healthy
3. The server CPU cannot be involved
4. The complaint is impossible

Correct answer: A. The tested gateway path is steady; other dependencies still need measurement

A gateway echo probe excludes many remote and application stages. Compare the actual service path and endpoint processing.

1. A guaranteed local VLAN error
2. A universal client adapter failure
3. The additional remote path or remote endpoint workload
4. A guaranteed specific WAN router fault

Correct answer: C. The additional remote path or remote endpoint workload

A controlled comparison reduces suspicion of shared local dependencies. It does not distinguish every WAN segment from the remote server without further tests.

1. Disable all authorization
2. Treat administrator success as universal availability
3. Use a representative approved student task and identity
4. Increase the probe packet size alone

Correct answer: C. Use a representative approved student task and identity

The test identity bypasses the failing condition. Service monitoring should include the intended user context and safe representative permissions.

1. The link must be replaced immediately
2. Current deltas and their relation to the symptom
3. The lifetime number proves an active fault
4. Every application must be affected

Correct answer: B. Current deltas and their relation to the symptom

Historical accumulated errors do not quantify present behavior. Preserve the history but measure current changes over the affected period.

1. The DNS record type
2. The active-path capacity and monitored interface context
3. The subnet broadcast address
4. The application passwords

Correct answer: B. The active-path capacity and monitored interface context

The denominator and selected resource must match the effective path. Failover can greatly reduce available capacity while a stale dashboard understates pressure.

1. Connection, identity, response, and task success after name resolution
2. Whether a query existed
3. Whether the probe ran
4. Whether any DNS answer was returned

Correct answer: A. Connection, identity, response, and task success after name resolution

DNS success validates one dependency. A complete service test needs the intended transport and application stages.

1. Replace every laptop first
2. Change the approved backup schedule/rate and compare task delay and queue behavior
3. Disable DNS permanently
4. Assume correlation alone proves the backup caused everything

Correct answer: B. Change the approved backup schedule/rate and compare task delay and queue behavior

The timing and queue evidence justify a bounded experiment. Keep other conditions comparable and verify that the intended classroom task improves.

1. 3%
2. 0.3%
3. 30%
4. 97%

Correct answer: A. 3%

30 missing replies out of 1000 sent is 3%. This is loss for that test's protocol, path, and window, not automatically application packet loss.

1. 0.72 minutes
2. 7.2 hours
3. 72 hours
4. 30 hours

Correct answer: B. 7.2 hours

Thirty days is 720 hours. One percent is 7.2 hours; the result depends on the stated window and definition of unavailability.

1. Only one remote website's page content
2. Only the number of DHCP reservations
3. Shared branch egress, workload, and remote-path baselines
4. Only the local printer driver

Correct answer: C. Shared branch egress, workload, and remote-path baselines

The pattern suggests a shared remote dependency worth testing. It does not yet identify a single router or prove the provider is responsible.

1. A separate check for collection gaps and monitoring-path health
2. Raise all alarm thresholds
3. Assume silence proves normal service
4. Delete the outage record

Correct answer: A. A separate check for collection gaps and monitoring-path health

Monitoring absence is not normal-value evidence. Detect missing data explicitly so a collection failure cannot silently suppress service investigation.

1. Server processing/workload as a plausible limiting stage
2. Only a larger DNS cache TTL
3. Only VLAN renaming
4. Only higher cable bandwidth

Correct answer: A. Server processing/workload as a plausible limiting stage

The measurements support an endpoint lead. Test processing and dependency behavior rather than assume every latency symptom is a network-capacity problem.

1. The network is necessarily broken
2. The link must be half duplex
3. The sum may exceed 100% without either direction exceeding its own capacity
4. The counter values must be negative

Correct answer: C. The sum may exceed 100% without either direction exceeding its own capacity

Full-duplex directional capacity is independent. Report each direction or clearly define an aggregate denominator to avoid misleading saturation claims.

1. Assume every time period has the same demand
2. Only compare with the quietest hour
3. Delete the baseline entirely
4. Use normal lunch-period samples with a comparable task and load

Correct answer: D. Use normal lunch-period samples with a comparable task and load

Seasonal and daily demand matter. A relevant baseline matches the operating context rather than selecting an artificially quiet reference.

1. The configured shaping/policing intent and measured service requirement
2. Whether DNS has an A record
3. The cable's 1 Gb/s label alone
4. Whether the endpoint needs a public IP

Correct answer: A. The configured shaping/policing intent and measured service requirement

A policy limit can govern throughput below physical capacity. Confirm the intended restriction before calling the result a hardware failure.

1. Only measure the server's disk size
2. Separate success by affected source or user cohort
3. Declare all users healthy from the aggregate
4. Only increase the total probe count globally

Correct answer: B. Separate success by affected source or user cohort

A large healthy population can hide a small fully affected group. Break down results along the dimension relevant to the complaint.

1. Sampling or event collection that can observe the short periodic condition
2. Increase the DNS TTL to 15 minutes
3. Turn off timestamps
4. Assume it never happened because probes passed

Correct answer: A. Sampling or event collection that can observe the short periodic condition

A sparse sampling schedule can systematically miss short events. Choose a safe method suited to the symptom duration and frequency.

1. The average is 100% because cumulative bytes equal the rated bit capacity
2. The calculation proves no short queue pressure occurred during the interval
3. The average is 12.5% because conversion from bytes to bits is unnecessary
4. The interval average is 100 Mb/s, or 10% of that directional capacity

Correct answer: D. The interval average is 100 Mb/s, or 10% of that directional capacity

375 million times 8 divided by 30 gives 100 million bits per second. That corrects the average claim but does not exclude brief bursts between samples.

1. Attribute the load-dependent RTT increase solely to longer geographic distance
2. Buy more bandwidth without observing queue or endpoint behavior
3. Compare queue occupancy and task latency under controlled upload demand and an approved queue/rate policy
4. Change DNS TTLs without measuring whether name resolution contributed to the delay

Correct answer: C. Compare queue occupancy and task latency under controlled upload demand and an approved queue/rate policy

Load-dependent delay is consistent with queueing. Measure the queue and test a controlled management change; a larger link may help, but the observations do not alone dictate that purchase.

1. Convert the negative sample difference into a negative traffic rate
2. Keep the pre-restart counter as a valid current baseline
3. Reject ordinary delta inference for that interval and restart the rate baseline
4. Apply a single-wrap correction despite the uptime and discontinuity change

Correct answer: C. Reject ordinary delta inference for that interval and restart the rate baseline

The supplied evidence identifies a reset/discontinuity. A wrap-only correction could create a false huge rate.

1. Add another administrator probe inside the data center and treat it as student verification
2. Use branch gateway ping as proof that the student role can complete the application task
3. A safe branch student-role transaction that validates the intended task
4. Monitor only successful TCP setup and omit the observed role error

Correct answer: C. A safe branch student-role transaction that validates the intended task

Source and identity matter. The current probe misses the user context and stage that fail; design the synthetic test around the actual authorized task.

1. Success/error rates and separate timing distributions for meaningful request groups
2. Assume a faster response always means a successful task
3. Exclude all error responses without reporting the changed failure rate
4. Report service improvement solely from the lower mixed average

Correct answer: A. Success/error rates and separate timing distributions for meaningful request groups

Mixing fast failures with successes can reverse the apparent trend. Report task correctness and group-specific latency before judging improvement.

1. Interpret counter-width validity as proof of unchanged resource identity
2. Resource identity, interface mapping, and replacement/discontinuity context
3. Assign old samples to the new fastest interface automatically
4. Continue the old trend solely because the numeric interface index looks unchanged

Correct answer: B. Resource identity, interface mapping, and replacement/discontinuity context

A stable-looking numeric key may no longer refer to the same physical/logical resource. Preserve continuity only when identity and semantics are confirmed.

1. Appropriate queue/drop or finer-grained telemetry tied to those bursts
2. Estimate exact burst peaks from a single coarse interval average
3. Dismiss the drops because the minute average is below capacity
4. Use another daily average to explain a 50-ms event

Correct answer: A. Appropriate queue/drop or finer-grained telemetry tied to those bursts

A measurement must resolve the event's time scale. Fine telemetry or queue events can explain pressure that coarse averages hide.

1. Infer that nominal upgraded capacity guarantees equal payload throughput
2. Receiver processing or application limits
3. Buy a second link upgrade without testing the now-saturated receiver
4. Treat low queueing as proof of unlimited server/application processing

Correct answer: B. Receiver processing or application limits

The bottleneck can move or already lie at an endpoint. Verify endpoint processing with a comparable workload before pursuing another network upgrade.

1. Treat the percentile label as proof of adequate sample population
2. Report an exact busy-period slow-tail rate from the ten quiet observations
3. Generalize the quiet-hour percentile to all busy-day users
4. No; small and unrepresentative sampling limits the claim

Correct answer: D. No; small and unrepresentative sampling limits the claim

The statistic's meaning depends on sample population and method. Collect representative observations and report sample size before generalizing to busy conditions.

1. An independent observation path for the service and missing telemetry
2. Rely exclusively on faster probes running on the powerless monitored device
3. Treat absence of local telemetry as a measured healthy value
4. Assume a co-located software monitor creates an independent failure domain

Correct answer: A. An independent observation path for the service and missing telemetry

Co-location creates a shared failure domain. Independent observation improves detection when the monitored component cannot report its own failure.

1. Verify only link-up state during an unloaded failover
2. Use the primary-link capacity as the backup utilization denominator
3. Required application performance and policy priorities under degraded capacity
4. Accept successful small echoes as proof that all upload and login goals are met

Correct answer: C. Required application performance and policy priorities under degraded capacity

Reachability and useful degraded service are different requirements. Test representative tasks and intended traffic treatment under the backup path's real limits.

1. Measurement offset rather than necessarily the network path
2. Report a proven physical-path speedup from the corrected clock offset
3. Reuse raw one-way differences without documenting clock behavior
4. Infer that a time correction changed the interface's forwarding capacity

Correct answer: A. Measurement offset rather than necessarily the network path

Clock correction can alter the measured difference without changing transit behavior. Use a defensible timing method and record its uncertainty.

1. Assume every interface error affects every user equally
2. The error is not yet evidence for these users' failure path
3. Select the highest error counter as the cause without checking the users' path
4. Replace that unrelated interface as the only repair and skip task verification

Correct answer: B. The error is not yet evidence for these users' failure path

A fault metric must belong to the affected dependency chain. Map the actual paths before joining coincident alarms to complaints.

1. No; the changed test conditions confound the comparison
2. Assume protocol and packet-size differences cannot affect policy or loss
3. Compare only the numeric percentages and omit target and time context
4. Attribute the lower percentage solely to the repair despite different tests

Correct answer: A. No; the changed test conditions confound the comparison

Controlled comparison requires relevant consistency. Standardize the task and report differing conditions rather than attribute the percentage change to one action.

1. Investigate headroom and demand against the service goal rather than declare an outage from the threshold alone
2. Declare no growth risk because current transactions still pass
3. Disable the interface before examining queues and the required service goal
4. Declare an outage solely because the utilization threshold crossed

Correct answer: A. Investigate headroom and demand against the service goal rather than declare an outage from the threshold alone

Threshold crossings merit context. They may indicate growth risk or expected load; assess queues, errors, task timing, and capacity requirements before labeling service failure.

1. Treat unknown-data exclusion as proof that actual availability increased
2. It can inflate reported availability by removing failures from the denominator
3. Count every missing observation as healthy without distinguishing collection failure
4. Remove all failed tests from the reporting denominator without disclosure

Correct answer: B. It can inflate reported availability by removing failures from the denominator

Definitions of unavailable and unknown data affect the result. Separate missing telemetry from tested failure and make exclusions explicit rather than silently counting neither.

1. Investigate the database dependency and server-side waiting
2. Investigate only DNS despite its stable timing in the affected requests
3. Ignore server-side stage timing because every web delay must be a network delay
4. Upgrade the WAN solely because total page duration increased

Correct answer: A. Investigate the database dependency and server-side waiting

Stage timing locates the dominant delay more precisely than the overall page duration. Validate the dependency under a comparable workload.

1. Choose identical remedies because their long averages match
2. Their time distributions and queue pressure differ
3. Assume the smoother site must have the same short full queues
4. Use one aggregate rate as a complete description of user experience

Correct answer: B. Their time distributions and queue pressure differ

The same average can conceal different peaks and delay patterns. Choose interventions from burst and service evidence rather than one aggregate number.

1. Assume the probe updated its destination merely because DNS changed
2. Whether the monitor still exercises the intended production dependency
3. Remove destination identity from the monitor specification
4. Treat continued success against the old pinned host as current-production validation

Correct answer: B. Whether the monitor still exercises the intended production dependency

The probe can remain green against a retired or different endpoint. Match resolution and destination behavior to the intended task while keeping tests controlled.

1. Verify the policy only with idle reachability probes
2. Declare success from the improved critical service and omit required bulk deadlines
3. Assume prioritization creates enough capacity for every workload without tradeoffs
4. Both service goals under representative combined workload

Correct answer: D. Both service goals under representative combined workload

Traffic treatment trades resources among workloads. Verify all stated requirements and document any approved prioritization rather than judge one favored metric alone.

1. Disable recovery mechanisms before checking setup duration and endpoint limits
2. Treat one small transfer as a conclusive sustained-capacity test
3. Multiply the small-file result by a guessed correction factor
4. Use an approved test workload and duration suited to sustained transfer while accounting for endpoint limits

Correct answer: D. Use an approved test workload and duration suited to sustained transfer while accounting for endpoint limits

Small transfers measure setup and ramp behavior along with transfer. Match the experiment to the uncertainty while avoiding disruptive production load.

1. Use suitable high-capacity counters or shorten the interval enough to avoid ambiguity
2. Continue ambiguous samples because the poll schedule is consistent
3. Clamp all negative deltas to zero and report healthy traffic
4. Assume exactly one narrow-counter wrap in every ten-minute sample

Correct answer: A. Use suitable high-capacity counters or shorten the interval enough to avoid ambiguity

Multiple wraps cannot be reliably reconstructed from two narrow samples alone. Choose collection that preserves rate information and discontinuity tracking.

1. Assign every failure to the provider report without checking affected paths
2. Ignore policy events because a remote incident was also reported
3. Assign every failure to the latest local change without control comparisons
4. Compare affected and control paths, policy decisions, provider evidence, and timing

Correct answer: D. Compare affected and control paths, policy decisions, provider evidence, and timing

Competing causes can coexist. Controlled comparisons and boundary evidence distinguish their contribution better than one temporal association.

1. Count successful login alone as successful protected-resource use
2. Validate the authorized task outcome beyond login
3. Disable result authorization instead of testing the approved role and action
4. Treat TCP setup as a substitute for the required result page

Correct answer: B. Validate the authorized task outcome beyond login

A passed identity check does not prove resource authorization or task completion. Define success around the intended operation and its safe expected result.

1. Use one speed-test screenshot as the sole basis for all future purchases
2. Choose capacity solely from total lifetime bytes
3. Comparable demand distributions, service outcomes, effective capacities, constraints, and growth assumptions
4. Ignore peaks, policy limits, and service goals because average demand is known

Correct answer: C. Comparable demand distributions, service outcomes, effective capacities, constraints, and growth assumptions

Capacity planning needs workload context, peaks, bottlenecks, and explicit requirements. Preserve the evidence and assumptions so a purchase can be evaluated against the actual limitation.