1.1.1 - IT Infrastructure and Cloud Computing
What you'll learn
- Trace a financial transaction through network, application and storage components.
- Distinguish virtualization from the characteristics that define cloud computing.
- Classify a cloud arrangement by what the provider operates and whom the infrastructure serves.
- Assign governance decisions and select evidence for controls retained by a cloud customer.
Estimated reading time: 19 minutes, including practice.
Why it matters
A billing system can acknowledge an order, lose the connection to its database and still report that its web servers are available. In an ISC scenario, tracing that order tells you which business objective is at risk, who can address the failure and which record could establish what happened.
Key terms
- Workload
- An application or processing job together with the resources it uses. A nightly receivables posting run is one workload.
- Instance
- A running copy of a computing resource, such as an application server. Multiple instances can run identical software.
- Tenant
- A customer or organizational environment logically separated from others using a service.
- Kernel
- The core of an operating system that manages resources such as memory, processors and devices.
- Failure domain
- A group of resources susceptible to the same failure, such as all equipment powered by one electrical circuit.
- Control plane
- The management functions used to create and configure resources. Authority to change a storage policy can reside here even when the administrator never uses the billing application.
- Residual risk
- The exposure that remains after controls have been applied.
Core content
1.1.1.1 Trace a transaction to its durable record
Start with the event the business recognizes. A sales clerk submits an order through a browser. The browser is a client, the application service applies business rules, and the database service records the accepted order. These are logical tiers. A small installation can run several tiers on one machine, while a large installation can spread one tier across many machines. Count the tiers to understand the work and trace their physical dependencies to understand what could fail together.
See Fig 1. The server cutaway locates three resources that behave differently when power is interrupted.
Key point: Working memory holds active processing state while a solid-state drive can retain written records after power loss.
{"type":"cross_section","describe":"The processor sits on the main board beside a row of narrow memory modules. A separate drive fits into a bay near the front of the chassis.","teaching_point":"A transaction in working memory can be lost before its durable record is written.","prompt":"Cross-section of a single rack server. Use a top oblique view with front drive bays nearest the viewer. Show a processor package on the motherboard, adjacent upright RAM modules and a front-mounted solid-state drive in a bay. MARKERS: 3 numbered circular markers, each a white digit on a solid dark circle, each joined to its target by a short thin leader line: 1 on processor; 2 on RAM module; 3 on solid-state drive. Use each digit exactly once, large and easy to read. NO words, captions or text labels anywhere in the image: digits 1 to 3 only. Use cool gray hardware and blue outlines to separate the three parts. Clean textbook illustration on white. Omit arrows, people and screens.","caption":"Fig 1. Inside a transaction server","size":"1536x1024"}- Central processing unit. Executes program instructions, including validation logic.
- Random-access memory. Holds active instructions and data while powered.
- Solid-state drive. Provides nonvolatile storage for data written to the device.
The operating system allocates machine resources to programs. Middleware connects software components, for example by queuing sales messages until a posting service can process them. The database management system (DBMS) interprets database requests, enforces configured permissions and coordinates commits. A commit makes a transaction's changes durable under the database's configured durability guarantees. A queue receipt can therefore be valid evidence of receipt while the downstream ledger posting is still pending.
Table 1 follows an order request across network components. The symptom narrows the investigation, but several components can produce similar symptoms.
| Component | Function | Illustrative failure |
|---|---|---|
| Switch | Forwards frames within a local network using link-layer addresses. | A port assigned to the wrong network isolates the clerk's terminal. |
| Router | Forwards packets between IP networks according to routes. | A missing route prevents the branch from reaching the application network. |
| Firewall | Permits or denies traffic according to configured rules. | A broad inbound rule permits an unintended path to the database. |
| Load balancer | Distributes requests across eligible service instances. | A shallow health check keeps sending requests to an instance that cannot post orders. |
| Domain Name System (DNS) | Resolves service names to records such as IP addresses. | A stale record sends clients to the retired service address. |
One appliance can perform several functions. A firewall can route packets, for example. The relevant control for a prohibited connection is the rule that permits or denies it. A virtual private network (VPN) can protect traffic across an untrusted network, but the user's database privileges still determine which records that user can change.
Follow one order number through these four records to locate a completeness exception.
A Locate the interrupted connection
A local area network (LAN) connects devices within a limited area, such as an office. A wide area network (WAN) can link a branch LAN to the central processing site. Suppose branch employees can still print locally, but all lose access to headquarters. The print job provides evidence about the local path it used. Test the shared path to headquarters before treating each workstation as a separate failure. The symptom alone cannot identify the failed device.
See Fig 2. Follow the cable used by a local print job and then the longer path needed to post a sale centrally.
Key point: A local print job can finish while the branch connection to headquarters is unavailable.
{"type":"instrument","describe":"At the branch, a terminal and a printer connect to the same switch. A separate path runs from that switch through the branch router, across a carrier link and through the headquarters router to a server.","teaching_point":"Only the central posting path depends on the carrier link in this arrangement.","prompt":"Instrument illustration of a two-site wired network. Use an elevated front view with branch equipment on the left and headquarters equipment on the right. At left connect a small terminal and printer separately to one switch, then connect that switch to a branch router. A long carrier link crosses the gap to the headquarters router, which connects to one server. MARKERS: 4 numbered circular markers, each a white digit on a solid dark circle, each joined to its target by a short thin leader line: 1 on the branch switch; 2 on the branch router; 3 on the carrier link; 4 on the headquarters router. Use each digit exactly once, large and easy to read. NO words, captions or text labels anywhere in the image: digits 1 to 4 only. Use cool blue for equipment, gray for connections and teal to highlight the carrier link. Clean textbook illustration on white. Omit arrows, screen contents, people and decorative objects.","caption":"Branch and headquarters connections","size":"1536x1024"}- LAN switch. Connects the terminal and printer on the branch network.
- Branch router. Forwards headquarters-bound packets toward the WAN.
- WAN link. Carries traffic between the two sites.
- Headquarters router. Forwards arriving packets toward the central server.
B Identify the storage interface
The business can call all three arrangements a place to keep files. Table 2 distinguishes them by what the consuming system actually requests.
| Arrangement | Unit addressed | Evidence to inspect |
|---|---|---|
| Storage area network (SAN) | Blocks on a volume presented to a host | Volume-to-host mappings and authorized host identifiers |
| Network-attached storage (NAS) | Named files and directories on a shared file system | Share permissions and effective file permissions |
| Object storage | An object identified by a key within a storage container | Object access policies and recorded service requests |
A database server can format a SAN volume with its own file system. Seeing database files on that server tells you little about how the underlying volume reached it. An unexpected second host mapping matters because it can expose the volume below the first server's file permissions.
With NAS, the storage service manages the shared file system. If a payroll folder becomes readable across departments, inspect the affected user's effective access through both the share and the file system. A network diagram can identify the appliance but cannot establish who could read the folder.
With object storage, an application might retrieve a payroll export by its key. Slashes in that key can resemble a directory path on a screen. Inspect how the application retrieves the export and which policy authorizes that request. The appearance of folders alone cannot establish that the application mounted a file share.
1.1.1.2 Distinguish isolation, pooling and elasticity
A hypervisor divides a host's physical computing resources among virtual machines (VMs). Each VM has a guest operating system and can run its own applications. Separate VMs can isolate software environments while sharing a processor, power supply and hypervisor. If two billing instances run on one host, either instance might survive an application crash in the other. Taking that host offline can stop both.
Containers package application code and its dependencies into isolated execution environments. Containers on a typical worker share the worker's kernel. Updating an application image can fix a library packaged in that image. A worker-kernel flaw calls for action at the worker layer. Containers themselves can run inside VMs, so the two techniques may appear in the same architecture.
Virtualization supplies a technical capability. Cloud computing adds a service delivery model. Table 3 distinguishes its five essential characteristics using evidence from a month-end reporting environment.
| Characteristic | Observable evidence | Control implication |
|---|---|---|
| On-demand self-service | An authorized analyst provisions a reporting workspace without a provider employee handling the request. | Restrict provisioning rights and record who creates resources. |
| Broad network access | Approved laptops and thin clients reach the service through standard network mechanisms. | Authenticate access across supported connection paths. |
| Resource pooling | The provider dynamically assigns capacity from a shared pool to different consumers. | Maintain logical separation as resources are reassigned. |
| Rapid elasticity | Workers expand for the closing run and contract when the backlog clears. | Apply security and logging configurations to new workers automatically. |
| Measured service | The platform records compute hours and storage consumed by each account. | Reconcile usage to authorized workloads and investigate unexpected consumption. |
Scalability is the ability to handle increased demand by adding resources. Scaling up adds capacity to an existing machine. Scaling out adds instances. Elasticity also accommodates declining demand, releasing capacity when it is no longer needed. A company that purchases a larger server for the next five years has scaled up. The purchase alone supplies little evidence of elasticity. Similarly, usage metering can support an internal chargeback even when no external vendor issues a bill.
See Fig 3. Trace the physical connections that create a shared dependency beneath two hosts.
Key point: Separate hosts can lose connectivity together through one shared switch.
{"type":"instrument","describe":"Two rack servers each have a separate network cable. Following either cable upward leads to a different port on the same switch.","teaching_point":"Host redundancy leaves a network dependency when both paths terminate at one switch.","prompt":"Instrument illustration of a rack-mounted network switch and two server hosts. Use a rear elevation with one switch above two separate servers. Show a continuous blue cable from each server's network port to a distinct port on the same switch, with no other data path. MARKERS: 3 numbered circular markers, each a white digit on a solid dark circle, each joined to its target by a short thin leader line: 1 on the upper server; 2 on the lower server; 3 on the single switch chassis. Use each digit exactly once, large and easy to read. NO words, captions or text labels anywhere in the image: digits 1 to 3 only. Emphasize the shared cable destination with cool gray hardware and blue cables. Clean textbook illustration on white. Omit arrows, logos and screens.","caption":"Fig 2. Two hosts with a common network dependency","size":"1536x1024"}- First physical host. Supplies resources to workloads assigned to this machine.
- Second physical host. Can continue computing if only the first host fails.
- Shared switch. Provides both hosts with their illustrated network connection.
A close-run capacity test can establish whether added workers actually preserve processing.
A Checkpoint
Two application VMs remain powered on during an outage. In Fig 3, which numbered component could disconnect both hosts through the illustrated paths while leaving both hosts powered?
- Component 1
- Component 2
- Component 3
- Components 1 and 2 must both fail
B Find the host layer and the source of the next guest
A Type 1, or bare-metal, hypervisor runs directly on the hardware. A Type 2, or hosted, hypervisor runs on a conventional host operating system. Suppose a developer runs two test VMs on a laptop. If the virtualization software runs as an application on the laptop's operating system, a restart of that system interrupts both guests. Count the layers beneath the guests to classify the hypervisor. Guest count and workload size do not establish its type.
A VM template is the source configuration used to create guests. Suppose operations corrects a logging setting on all eight active guests on Friday. A ninth guest created on Monday starts with logging disabled. Compare the new guest's template version and startup configuration with the approved baseline. Friday's inspection can support the conclusion that eight guests were corrected that day. Testing a fresh deployment addresses whether the correction survives the next scale-out.
1.1.1.3 Locate the customer-managed layer
A provider can sell several services under one contract. Classify each service by what the customer receives and administers. The same company might rent virtual servers for a ledger interface, deploy code on a managed runtime for inventory and subscribe to a finished payroll application. Logging in through a browser is possible in all three arrangements.
Table 4 assigns typical operating duties. A separate managed-service agreement can delegate additional tasks, so inspect the terms for the particular workload.
| Model | Provider operates | Customer typically controls | Evidence from customer |
|---|---|---|---|
| Infrastructure as a service (IaaS) | Physical facilities, hardware and virtualization layer | Guest operating system, installed software, data and permitted network settings | Guest patch inventory and installation results |
| Platform as a service (PaaS) | Underlying infrastructure, operating system and supported runtime | Deployed application code, packaged dependencies, secrets and application configuration | Release approval tied to the deployed build |
| Software as a service (SaaS) | Finished application and supporting platform and infrastructure | Authorized users, permitted tenant settings, submitted data and use of outputs | Tenant role export and evidence of authorized changes |
In a conventional on-premises installation, the entity generally operates the full stack from facilities through applications, although it may contract for support. At a colocation site, the operator may supply space, power and cooling while the customer owns and administers the servers. Renting floor space changes the facilities operator. It does not by itself establish that the customer obtains pooled, on-demand computing.
See Fig 4. The branch points classify three separate offers from the same provider.
Key point: The delivered capability distinguishes a hosted application, a managed application platform and rented computing infrastructure.
flowchart TD
A[Examine delivered capability] --> B{Finished application supplied?}
B -->|Yes| C[Software as a service]
B -->|No| D{Provider manages application runtime?}
D -->|Yes, customer deploys code| E[Platform as a service]
D -->|Customer manages guest OS| F[Infrastructure as a service]Trace a software defect to the evidence held by the party that can correct it.
Worked case
A distributor purchased 12 virtual servers. The provider maintains the hypervisor and physical hosts. The distributor installs the guest operating system. Its policy requires a specified guest update on every active server by June 30. A June 30 inventory lists 12 active guests. The deployment log shows 12 successful installation events, but two are for the same guest. All identifiers are valid, and the other 10 events concern different guests. The provider also reports full host availability for June.
The question is whether the evidence supports complete guest patch coverage. The distributor owns that control under this agreement. Its denominator is 12 active guests. Matching successful events by unique guest accounts for 11 guests because the repeat installation contributes no additional server. One guest lacks installation evidence. Compare its current version and update history with the required version before deciding whether it actually missed the deadline. Twelve successful events is a count of actions, while 12 active guests is a count of assets. Host uptime addresses neither the installed version nor the unaccounted-for guest.
A Checkpoint
A customer uploads its own validation code. The provider maintains the runtime and operating system. A developer changes the code to accept invalid customer numbers. Which evidence most directly addresses the control over this change?
- Approval and test results linked to the deployed customer build
- Provider records of operating-system patch installation
- Customer review of the provider's physical access list
- Provider measurements of runtime availability
B Follow an event into a serverless function
In function as a service (FaaS), a provider can invoke customer code when a file arrives, a timer expires or a service receives a request. The provider provisions the execution capacity. The customer controls the deployed function and the permissions it grants the function under the arrangement. If a payroll import stops after a deployment, inspect the deployed version, event binding, execution role and invocation results. A count of healthy provider servers gives little evidence about that particular import.
Read the delivery guarantee. Under at-least-once delivery, the same event can trigger a retry after the first invocation has already posted the transaction. An idempotent posting produces the same business result when the same event is processed again. For example, enforce uniqueness on the source event identifier as part of the posting transaction. Test by submitting the same event twice and inspecting the ledger. Two invocations and one committed posting can be the expected result.
1.1.1.4 Determine who shares the infrastructure and what can fail together
A service model answers what capability is supplied. A deployment model answers who may use the cloud infrastructure and, for a hybrid arrangement, how distinct environments are linked. These classifications can coexist. An organization can obtain an application platform for its exclusive use or deploy its code on a platform offered to many unrelated customers.
Table 5 compares the access and integration conditions that distinguish deployment models.
| Model | Defining condition | Example |
|---|---|---|
| Private | Exclusive use by one organization, possibly across several business units | A third party operates dedicated cloud infrastructure for one bank. |
| Public | Infrastructure offered for open use by the general public | Unrelated subscribers can purchase isolated environments from the same offering. |
| Community | Exclusive use by a defined community of organizations with shared concerns | Member credit unions share infrastructure restricted to participating institutions. |
| Hybrid | Distinct cloud infrastructures remain separate but are bound to enable data or application portability. | A private cloud moves overflow workloads to a linked public cloud. |
Match each arrangement to its defining deployment feature.
- Private cloud:
- A hosted environment reserved for one organization and its divisions.
- Community cloud:
- Admission limited to participating organizations with common requirements.
- Public cloud:
- An offering available to unrelated subscribers from the general market.
- Hybrid cloud:
- Distinct environments linked to move workloads between them.
A private network segment inside a public offering can block outside access to a customer's servers. That network setting alone does not change the deployment model. Similarly, buying subscriptions from two cloud providers creates a multicloud sourcing arrangement. Hybrid classification requires the relevant connection and portability between distinct cloud infrastructures. A conventional local server connected to SaaS demonstrates integration, but that description alone does not establish a private cloud.
A region identifies a geographic deployment area. An availability zone is a provider-defined location or grouping within a region designed to limit common failures. Examine the actual design because providers define their boundaries differently. Separate zones can support resilience to a localized failure. Separate regions may address a regional disruption but can introduce data-location constraints, replication delays and dependence on a working failover configuration.
See Fig 5. Physically trace both supply cords before accepting that the server has independent power paths.
Key point: Two working power supplies can both lose input when their shared distribution unit fails.
{"type":"instrument","describe":"The rear of one server shows two removable power supplies, each with its own cord. Both cords plug into different sockets on the same vertical power distribution unit.","teaching_point":"The common power distribution unit limits the protection afforded by dual supplies.","prompt":"Instrument illustration of dual server power supplies on one shared power distribution unit. Rear three-quarter view of a single server and one adjacent vertical rack power strip. Show two separate removable supply modules in the chassis, one blue cord from each module, and both uninterrupted cords ending in distinct sockets on the same gray distribution unit. MARKERS: 3 numbered circular markers, each a white digit on a solid dark circle, each joined to its target by a short thin leader line: 1 on the first supply module; 2 on the second supply module; 3 on the shared distribution unit. Use each digit exactly once, large and easy to read. NO words, captions or text labels anywhere in the image: digits 1 to 3 only. Emphasize both plug destinations. Clean textbook illustration on white. Omit arrows, logos and unrelated cabling.","caption":"Fig 4. Redundant supplies on one feed","size":"1536x1024"}- First power supply. Can power the server independently when sized for its load.
- Second power supply. Provides component redundancy if either supply can carry the load.
- Power distribution unit. Supplies electricity to both modules in this installation.
Apply the same physical reasoning to an order service described as operating in two zones.
A Compare the cost of entering, operating and leaving
Table 6 identifies costs to include when comparing a cloud migration with keeping the existing installation.
| Cost group | Examples | Question for the estimate |
|---|---|---|
| Transition | Data conversion, testing, training and parallel operation | Which costs arise before the old service can retire? |
| Recurring service | Compute, storage, support and outbound data transfer | What volume drives each charge? |
| Retained operations | Customer monitoring, integration support and remaining licenses | Which costs continue after migration? |
| Exit | Extraction, conversion and transition assistance | Does the decision horizon include a planned exit? |
A three-year proposal costs $70,000 annually plus $50,000 for migration. Keeping the local system costs $80,000 annually over the same period. With all other future costs equal and discounting excluded, cloud totals $260,000 and local totals $240,000. The $10,000 annual saving recovers only $30,000 of the migration cost within three years. A prior equipment purchase with no recoverable value is a sunk cost. Including it again would distort the decision.
Vendor lock-in. A company successfully exports every invoice, but its billing code calls a proprietary provider workflow for approval and release. An exit estimate must include rebuilding that workflow, converting configuration and testing the replacement. A stronger portability exercise runs a representative invoice through approval and release on the proposed platform, then reconciles its result. A readable export establishes data portability for the tested records. It leaves application portability untested.
Data residency. Map every copy covered by the requirement. Suppose a contract requires payroll identifiers to stay in a specified country. The primary database is there, but diagnostic logs containing employee identifiers are copied elsewhere. Inspect backup destinations, replica placement and log destinations against the contract, including the fields each copy contains. A region selected for the primary database establishes location only for that resource.
1.1.1.5 Connect decision rights to control evidence
IT governance sets direction for the use of technology and monitors whether it delivers business value within acceptable risk. A governing body evaluates stakeholder needs, directs priorities and monitors outcomes. IT management translates that direction into plans, budgets, implementations and corrective action. For example, a board may set the maximum acceptable interruption to billing, a steering committee may prioritize funding for resilience, and the operations manager may schedule the failover exercise. Committee titles vary. The charter and delegated authority establish who can make the decision.
Governance must reconcile benefits, risk and resources. A cheaper payroll service could meet its uptime target and still be unsuitable if it cannot return payroll history in usable form. Before committing, management evaluates the service against business requirements, classifies the data, identifies critical dependencies and considers provider concentration. Three outsourced applications may share one identity provider. Losing that provider can prevent access to all three despite separate application contracts.
A responsibility matrix maps a specific activity to its participants. In a RACI matrix, the responsible party performs work, the accountable party owns the result, consulted parties provide input and informed parties receive updates. Split a broad line such as 'access managed by vendor' into provider administrator access, tenant account authorization, provisioning and termination review. A provider can execute a removal promptly and still leave the user active for weeks if the customer never submits the termination.
Table 7 is an extract from a fictional service agreement. Its separate measures matter when deciding whether a quarterly operating report provides relevant evidence.
| Commitment | Measurement basis | Owner |
|---|---|---|
| Gateway interruption | At most 30 unplanned minutes per calendar month, measured at the provider gateway. | Provider operations |
| Payroll export delivery | File available by 06:00 Eastern time on each scheduled payroll day. | Provider operations |
| Export reconciliation | Match the accepted payroll total to the imported ledger batch before release. | Customer payroll supervisor |
| Recovery test | Demonstrate restoration of an identifiable payroll file from a retained copy each quarter. | Provider performs and customer verifies output |
A service-level agreement (SLA) defines a commitment and how performance is measured. Inspect the measurement point, reporting window, treatment of planned maintenance, customer obligations and remedy for failure. A service credit may compensate for an interruption without restoring missing records. For an exit, establish the extraction format, access window, responsibility for keys, transition assistance and disposition of retained copies. A trial export opened and reconciled in the replacement system is stronger evidence of portability than a contract clause promising a download.
For financial reporting, a SOC 1 report addresses service organization controls likely to be relevant to user entities' internal control over financial reporting. A SOC 2 report addresses the applicable Trust Services Criteria in scope. Read the system description and opinion to determine whether the purchased service and relevant controls were covered. A Type 1 report addresses design as of a specified date. A Type 2 report also addresses operating effectiveness over a specified period. Inspect the actual tests, exceptions and period before using a conclusion about a control.
If a subservice organization's controls are carved out, its operating controls fall outside the service auditor's testing in that report. Determine whether the omitted work matters to the objective and obtain relevant evidence if needed. Complementary user entity controls are controls the service organization assumes customers implement in achieving specified objectives or criteria. Evaluate the applicable customer controls separately. A period after the report end date also requires consideration of changes and what additional evidence is needed.
Follow the evidence for a payroll authorization control that crosses the service boundary.
Worked case
A payroll provider supplies a Type 2 SOC 1 report covering January through September. A client's year ends December 31. The provider processed 26 payroll runs during the year, including six after September. The report describes customer approval of each uploaded instruction file as a complementary user entity control. The client has an unmodified provider opinion and a dashboard showing 100% gateway availability.
The objective is evidence that authorized payroll instructions were processed throughout the year. First establish that the service, controls and tests in the report address processing of this client's files. The report period leaves three months and six runs beyond its end. Inquire about subsequent changes and obtain evidence appropriate to the gap and assessed risk, such as later control testing or direct testing of relevant transactions. The six runs identify the uncovered population. They do not prescribe a sample size.
Next inspect the client's approval evidence across the period for which reliance is planned. A complete provider file-processing log can establish which files ran, yet it cannot establish that the customer approved their contents. The availability dashboard measures a different objective. If sufficient appropriate evidence remains unavailable, reassess reliance and design responsive procedures before drawing a reporting conclusion.
See also. Business Process and Application Controls covers validation and posting. Data Management and Databases and Data Analysis and Integrity develop the record-level evidence. Availability, Backup and Recovery addresses restoration objectives. IT Audit and Control Testing develops procedure selection. Security and Privacy Frameworks, Logical and Physical Access Controls, and Threats, Vulnerabilities and Incident Response extend the analysis of shared resources and privileged access.
A Map a decision to its COBIT domain
COBIT organizes governance objectives in one domain and management objectives in four. Table 8 uses the same billing investment throughout so that the action in the scenario determines the answer.
| Domain | Function | Billing example |
|---|---|---|
| Evaluate, Direct and Monitor (EDM) | Governance | The board evaluates the investment, directs priorities and monitors benefits. |
| Align, Plan and Organize (APO) | Management planning | Management develops a sourcing plan and budget aligned with the billing objective. |
| Build, Acquire and Implement (BAI) | Management implementation | The project team configures, tests and deploys the approved platform. |
| Deliver, Service and Support (DSS) | Management operations | Operations runs billing and resolves service interruptions. |
| Monitor, Evaluate and Assess (MEA) | Management assessment | Management assesses control performance and reports deficiencies. |
The word monitor occurs in both EDM and MEA. Ask what decision follows the monitoring. A board reconsidering whether an investment still serves the enterprise is exercising governance. A management team measuring control performance and following up on corrective actions is carrying out assessment.
B Make strategic alignment measurable
Start with the enterprise outcome. If delayed invoices postpone collections, an IT proposal can target the proportion of invoices issued by the contractual deadline. Record the baseline, desired result, accountable business owner and date for measuring benefits. A report that the team deployed the platform on time supports project completion. Compare actual invoice timeliness with the baseline to assess whether the investment delivered its business benefit.
A key performance indicator (KPI) measures achievement of an objective. A key risk indicator (KRI) measures exposure and can trigger intervention. For the same billing service, the percentage of invoices issued on time measures delivery performance. The percentage of critical interfaces approaching the end of vendor support can warn of increasing continuity risk. Classify a measure by what management uses it to decide. A number on a board dashboard can serve either purpose.
C Identify the governing document
When an exhibit describes a rule, read the authority and level of detail in its wording. Table 9 follows a single requirement from management direction to implementation.
| Document | Purpose | Illustrative content |
|---|---|---|
| Policy | States a management requirement. | Production resources require authorization before deployment. |
| Standard | Sets mandatory, specific criteria. | Each production resource must carry an owner identifier and an approved change identifier. |
| Procedure | Prescribes the steps for carrying out an activity. | Validate approval, apply the identifiers, deploy and retain the result. |
| Guideline | Recommends a practice. | Use a descriptive workload name to help reviewers recognize its purpose. |
An examiner finds a resource with a descriptive name and no owner identifier. The resource satisfies the naming recommendation while failing the mandatory ownership standard. Inspect the approved rule and any authorized exception before classifying the finding. A procedure describing how to enter the identifier is evidence of the intended process. The resource configuration establishes whether that step produced the required result.
D Recalculate a percentage-based availability commitment
Use the contract's measurement window for the service being assessed. Let \(W\) equal eligible service minutes after contractual exclusions and \(D\) equal counted downtime minutes within that window. When each eligible minute is classified as available or unavailable, percentage availability is \(100(W-D)/W\). A monthly percentage cannot be evaluated from outage minutes alone without determining the eligible window.
A separate agreement covers a 30-day month around the clock and excludes a 120-minute maintenance window from eligible time. The eligible window is \(30\times24\times60-120=43{,}080\) minutes. With 36 minutes of counted downtime, availability is \(100(43{,}080-36)/43{,}080\approx99.9164\) percent. This meets a 99.9 percent minimum. Use the unrounded result for comparison unless the contract specifies rounding. Subtracting the excluded maintenance again as downtime would charge the same minutes twice.
E Recall check
Answer each question before revealing the matching explanation.
- What distinguishes a container from a VM at the operating-system layer?
- Containers on a worker typically share its kernel. A VM runs a guest operating system.
- Which cloud characteristic is demonstrated by releasing idle workers after month-end?
- Rapid elasticity accounts for contraction as well as expansion.
- What distinguishes community from private deployment?
- Community infrastructure serves a defined group of organizations with common concerns. Private infrastructure serves one organization.
- Which evidence addresses the customer's ability to leave a payroll provider?
- A trial export imported and reconciled in the replacement environment demonstrates usable data portability.
- What does a carve-out mean for subservice controls?
- The report excludes those operating controls from its testing. Their relevance determines the need for additional evidence.
Common mistakes
These descriptions sound reassuring until the requested objective is compared with the evidence.
| Tempting conclusion | Why it sounds plausible | Decisive detail |
|---|---|---|
| A password-protected tenant is a private cloud. | Only employees can enter the tenant. | Deployment classification turns on who may use the underlying offering. |
| A successful replica proves historical data can be recovered. | The second copy has current data. | The deletion propagated to the replica. Recovery needs a usable earlier version. |
| Two providers ensure independent service continuity. | Different vendors operate the applications. | Both require the same failed identity service. |
| A vendor uptime percentage establishes timely payroll delivery. | The gateway stayed reachable during the run. | The delivery commitment measures when the completed file became available. |
| A dashboard exception automatically changes the auditor's opinion. | The promised control appears to have failed. | The consequence depends on the affected objective, other evidence and significance of the failure. |
Approach
Use the business objective to follow one control across a hosted arrangement.
Self-check quiz
Question 1
A credit union deploys its own loan application to a managed runtime. Only participating credit unions with common supervisory requirements may use the infrastructure. Which combination describes the service?
- IaaS in a private cloud
- PaaS in a community cloud
- SaaS in a community cloud
- PaaS in a public cloud
Question 2
Using Table 7, the gateway had 26 unplanned minutes of interruption in April and four planned maintenance minutes. On one scheduled payroll day, the export became available at 06:08 Eastern time. The customer reconciled it before release. Which conclusion follows?
- The provider exceeded the gateway interruption allowance only.
- The customer missed the export reconciliation commitment only.
- Both gateway interruption and export delivery commitments were missed.
- The provider missed the export delivery commitment only among the measures described.
Question 3
A firm wants evidence that no unauthorized journal file was accepted by its hosted ledger. The provider shows that every submitted file was processed without alteration. Its service description assigns file authorization to the customer. Which procedure best addresses the remaining concern?
- Match accepted file identifiers to dated customer approvals for those files.
- Compare total submitted bytes with total bytes received by the provider.
- Inspect the provider's reconciliation of accepted files to processed files.
- Inspect the provider's review of physical access to storage devices.
Question 4
A security team updates a library in every deployed container image. An assessor's remaining finding concerns the shared kernel on the workers. Which evidence would address it most directly?
- Image build records showing the updated application library
- Logs showing that every container restarted after the image release
- Worker inventory matched to the running kernel versions after remediation
- A list of application users with permission to schedule jobs
Question 5
The board has delegated acceptance of moderate payroll continuity risk to the payroll executive. An exercise identifies moderate residual risk within that limit. Which action best demonstrates that governance operated as designed?
- The infrastructure manager closes the risk because the exercise ran successfully.
- The payroll executive documents the decision, rationale and conditions for reconsideration.
- The CPA authorizes continued use after confirming that management disclosed the exposure.
- The cloud provider accepts the client's business exposure by issuing a service credit.
Answer key
Question 1 - Answer: B. Customer-deployed code on a managed runtime indicates PaaS, and membership defined by shared supervisory requirements indicates a community cloud. A mistakes restricted membership for exclusive use by one entity and assumes guest administration. C treats a hosted runtime as a finished loan application. D overlooks the membership restriction.
Question 2 - Answer: D. The file arrived eight minutes after the 06:00 deadline. A incorrectly treats the gateway allowance as exceeded when only 26 minutes count. B mistakes late provider delivery for a failure of the customer's before-release reconciliation. C combines the real delivery miss with a gateway breach unsupported by the measurements.
Question 3 - Answer: A. Starting with accepted files and tracing to their approvals can identify a processed file the customer never authorized. B addresses transfer volume. C addresses completeness of processing for accepted files. D addresses physical access, which gives no file-specific authorization evidence.
Question 4 - Answer: C. The affected component runs on each worker, so match the worker population to evidence of the running kernel version. A demonstrates a library change in the image. B demonstrates a restart that could reuse the same worker kernel. D addresses use of scheduling privileges.
Question 5 - Answer: B. The named executive holds delegated authority for this level of exposure and can document an informed decision. A confuses completion of a technical exercise with authority to accept business risk. C assigns a management decision to the independent evaluator. D treats contractual compensation as the client's acceptance of operational exposure.
Rapid review
- Receipt in a queue → trace onward to the committed business record.
- Two instances → inspect the host, network and power they share.
- Customer maintains the guest → start with the IaaS responsibility boundary.
- Customer ships code to a managed runtime → inspect deployment controls under PaaS.
- Provider runs the finished application → examine SaaS tenant permissions and settings.
- Private or community → count eligible organizations and identify the admission condition.
- Contract performance → compare actual results with the stated measurement point and window.
- Provider report → match the service, scope, period and customer obligations.
- Residual risk acceptance → locate the owner with authority for that exposure.
- Host receives a disk volume → inspect block-storage mappings.
- New guests recreate an old defect → test deployment from the approved template.
- Repeated delivery of one event → verify one business effect using its source identifier.
- Successful data export → test dependent workflows before concluding the application can move.
- Board evaluates investment benefits → classify the action under EDM.
- Percentage uptime target → determine eligible minutes before calculating the result.
Keep going with the full CPA Information Systems and Controls (ISC) Prep Course
You just studied the way members do, in the same lesson interface. Full access unlocks every lesson, the question bank and flashcards for a full year.
One payment per program