Cloud Instance Recommender

• Size against p90, p99 or a per-row policy — each VM can be sized on a statistic chosen from its ENV, Criticality and Workload, and a safety veto holds a downsize when the VM's own Peak or p95 says it is saturating, marking the row in a Review Flag column. Criticality is a new optional column.
• Sizing confidence and a scale-out advisory — the input check says how well your data supports a sizing, and the App Portfolio flags VMs with a low median and a high peak as scale-out candidates, as guidance with its reasons.
• Validated Excel template — Download Excel Template gives a workbook whose ENV, OS, Workload and Criticality cells are dropdowns.
• Browser Storage and fleet snapshots — see everything the tool keeps in your browser, save a run as a named snapshot, compare it with a later run of the same estate, and export or import a snapshot as a file.
• A relative savings figure — the stats bar, the App Portfolio and the print report show how far Optimized ranks below Like-to-Like, as a percentage only, never a dollar amount.
• Input limits and a reserved-column notice — a file over 100,000 rows or 500 columns is refused with the reason, and a column named like one of this tool's own output columns is named in the input check.

• Filter presets — save the whole filter configuration under a name and re-apply it anytime from the bar above the Generate button; presets are per page and stay in your browser.
• Results Excel — download the results grid as a styled .xlsx (formatted header, autofilter, numeric columns) alongside the existing CSV.
• Nearest Miss column — no-match rows now name the closest qualifying instance and the filter(s) that excluded it, so you know exactly what to relax.
• Scenario comparison — pin generation runs (named, up to six) and diff them. Two runs give a detailed pairwise view: match-rate change, newly matched/unmatched VMs, the configuration settings that differed, and old → new values for every changed recommendation. Three or more give an N-way matrix — one column per run over the rows that differ across the set. Either view exports to CSV.
• Install & offline — the tool is an installable app that keeps working offline after your first visit; a page always loads the current version when you're online, and everything else updates automatically in the background.

• App Portfolio & executive Excel — add an App Name column and group your estate by application: a per-app dashboard (Open App Portfolio after generating) with a searchable app table you can filter to no-match or compliance-sensitive apps, a per-app App CSV export, and a styled multi-sheet .xlsx export of the whole portfolio.
• Excel upload — .xlsx workbooks are accepted alongside CSV. In a multi-sheet workbook the sheet whose columns best match an inventory is opened, and a picker lets you switch to any other sheet.
• Column auto-mapping — headers like vCPUs, RAM, or Hostname map automatically to the expected columns; ambiguous cases open a mapping panel, and confirmed mappings are remembered for repeat uploads.
• Region Check panel — after upload, chips show each region as recognized (green), auto-resolved (amber, e.g. us-east-1a → us-east-1), or unknown (red — those rows would use built-in sample data).
• Faster loading — instance data loads per region on demand; pages open with a few KB of data instead of ~25 MB.
• Real progress, no freezing — generation runs in a background worker with a row-accurate progress bar and automatic fallback.
• Preview search — filter the results preview live across all visible columns.
• Fit / headroom flag — a ▲% beside a like-for-like match shows how far the chosen instance over-provisions its worst axis (vCPU or memory) versus the size you requested — the unavoidable waste of discrete instance sizes and mismatched vCPU:RAM ratios. Shown only when it is worth flagging (≥ 25%), and more emphatically at ≥ 100%.
• No-Match export — download only the rows that got no recommendation, with reasons, to fix and re-upload.
• Executive print report — 🖨️ Print Report opens a print-ready one-page summary (headline stats plus the match-rate, recommended-family, and vCPU/RAM before → after charts) to print or save as PDF.
• Alternative recommendations — beside the best match, each row carries Most Cost Optimized, Workload Based, Newest Generation, and Best Network picks per provider (Best Network uses a real published field on AWS/Azure, a vCPU-count proxy on GCP) (separate columns; the Excel export gets one sheet per strategy).
• Dark mode — 🌓 toggle on every page; follows your OS setting until you choose manually.
• Keyboard & screen-reader support — collapsible sections, table sorting, and status updates are fully accessible.

Welcome to the Cloud Instance Recommender, a client-side, auditable rule-evaluation engine for cloud infrastructure decisions. Sizing across AWS, Azure, and Google Cloud Platform is its first and most developed policy set — evaluated entirely in your browser, with no data ever uploaded to an external server.

This distinction matters for a specific kind of buyer: regulated-industry infra teams, government contractors, air-gapped or data-residency-constrained shops, and MSPs handling client data often cannot accept a mainstream sizing platform's default SaaS connection back to the vendor, even where a self-hosted tier exists elsewhere in that platform's lineup. This tool has no such connection to disable or negotiate around — nothing it processes ever leaves the browser. Nothing here changes who can use the tool; it changes who this guide, and the product roadmap, are written for.

The Rule Engine UI builds on the underlying rule engine with five capabilities: interactive Rule Engine UI dropdowns that set global defaults without CSV changes, a Minimum Generation filter to exclude EOL instance families (e.g., AWS m5/r5), conflict detection that highlights contradicting filter combinations in red, a split AWS Pricing Calculator Bulk Template that generates separate files for Like-to-Like and Optimized recommendations, and a Min Gen CSV column for per-row generation control.

🌐 Supported Cloud Providers

VM Name,App Name,CPU Count,Memory (GB),CPU Utilization,Memory Utilization,AWS Region,ENV,OS,Workload,Compliance,Min Gen,Exclude,Current Instance Type
web-server-01,Storefront,4,16,45,60,us-east-1,Production,Linux,Web Server,,6,,m5.xlarge
db-server-02,Billing,8,32,70,80,us-west-2,Production,Windows,Database,PCI,7,"Burstable,GPU",m5.2xlarge
app-server-03,Billing,2,8,35,45,eu-west-1,Dev,Linux,General,,,,t3.large
cache-server-04,Storefront,4,16,25,30,us-east-1,Staging,Linux,Cache,,6,Burstable,m5.xlarge
ml-server-05,Analytics,8,64,80,75,us-west-2,Production,Linux,ML/AI,HIPAA,7,,r5.2xlarge

The Rule Engine applies automatically when optional CSV columns are present. Rules narrow the candidate instance list before the cheapest-match selection runs — so every recommendation already satisfies your environment constraints.

🔢 Min Generation Rule (MG)

The Min Generation rule excludes instances older than a specified hardware generation. This is useful when older generations are approaching end-of-life (e.g., AWS m5/r5 EOL) or when you have a policy requiring modern silicon.

🎯 Workload Family Mappings

📋 Rules Reference Summary

Step 1 — Choose Your Cloud Provider

• Single provider: Navigate directly to the AWS, Azure, or GCP page for focused recommendations with provider-specific filtering options.
• Multi-cloud: Use the Multi-Cloud page to compare recommendations across all three providers in a single run from one CSV upload.

Step 2 — Download and Prepare the Template

1. Click Download Sample CSV on the provider page to get a template with the correct column headers. Or click Download Excel Template for a workbook whose ENV, OS, Workload, and Criticality cells are dropdowns, so a mistyped value cannot slip in. Its Read me and Allowed values sheets explain each column; the tool reads the Inventory sheet.
2. Replace the sample rows with your actual VM inventory. Include the optional ENV, OS, Workload, and Compliance columns where applicable.
3. Save the file as UTF-8 CSV.

Step 3 — Upload Your CSV

1. Drag and drop your CSV onto the upload zone, or click it to browse.
2. The tool validates the file, checks required columns, and shows a data preview and an input check (see the FAQ below).
3. Review any warnings before proceeding — common issues include missing region columns or non-numeric CPU/memory values.

Step 4 — Select Recommendation Type

Step 5 — Configure Optimization Settings

(Only relevant for Optimized or Both recommendation types)

Optimization Mode

• CPU-Based: Scale CPU according to utilisation thresholds
• Memory-Based: Scale memory according to utilisation thresholds
• Both can be enabled simultaneously

N/2, N, N+1 Thresholds (Industry-Standard Defaults)

Step 6 — Configure Rule Engine & Advanced Filtering (Optional)

Expand the Advanced Filtering section to access both the Rule Engine UI controls and provider-specific filters.

Rule Engine Defaults

Five dropdowns let you set rule defaults for the entire batch without modifying your CSV. Per-row CSV column values always take priority over these UI defaults.

Conflict Detection

When two selected options directly contradict each other, the conflicting dropdowns are highlighted in red with an explanatory warning. For example:

• OS = Windows + processor filter restricted to Graviton/ARM only
• ENV = Production or Staging + main families restricted to burstable types only (t/B/e2)
• Workload = ML/AI + processor filter restricted to Intel/AMD only (no GPU available)
• OS = macOS while an Azure or GCP provider is selected (macOS is AWS-only)

Resolve conflicts before generating recommendations — rules will still run but may produce unexpected results if contradictions are left in place.

Step 7 — Generate and Download Results

1. Click 🔄 Generate Recommendations. A progress bar shows processing status for large files.
2. Once complete, the download section updates based on what was generated.
3. AWS page — both Like-to-Like and Optimized selected: three buttons appear:
   📊 Download Results (Excel) — the full workbook (the primary download); flat CSVs are behind the CSV ▾ menu
   🧾 Bulk Template (Like-to-Like) — AWS Pricing Calculator format, Like-to-Like instances only
   🧾 Bulk Template (Optimized) — AWS Pricing Calculator format, Optimized instances only
4. AWS page — single type selected: Download Results (Excel), the CSV ▾ menu, and one Bulk Template button.
5. Azure / GCP pages: Download Results (Excel), with the Results CSV behind the CSV ▾ menu.

The downloaded CSV preserves all original input columns and appends recommendation columns for each selected provider.

Original Input Columns (preserved)

VM Name, CPU Count, Memory (GB), CPU Utilization, Memory Utilization, [Provider] Region, ENV, OS, Workload, Compliance — all returned unchanged.

Recommendation Columns (added per provider)

The same pattern applies for Azure and GCP (replacing "AWS" with the provider name).

Status values for recommendation columns

The Advanced Filtering section contains two types of controls: the Rule Engine UI (global defaults for ENV/OS/Workload/Compliance/Min Gen) and provider-specific filters that narrow the instance candidate pool. Provider filters are applied first; the Rule Engine then runs on the already-filtered set.

🔶 AWS-Specific Filters

🔷 Azure-Specific Filters

🔴 GCP-Specific Filters

The Multi-Cloud page processes a single CSV against all three providers simultaneously, producing side-by-side recommendations in one output file. This is useful for migration planning, vendor selection, and lock-in risk assessment.

CSV Format for Multi-Cloud

Include region columns for every provider you want to compare. Providers with a blank or missing region column are skipped for that row.

VM Name,App Name,CPU Count,Memory (GB),CPU Utilization,Memory Utilization,AWS Region,Azure Region,GCP Region,ENV,OS,Workload,Compliance,AWS Min Gen,Azure Min Gen,GCP Min Gen,Exclude,Current Instance Type
web-server-01,Storefront,4,16,45,60,us-east-1,East US,us-central1-a,Production,Linux,Web Server,,,,,,m5.xlarge
db-server-02,Billing,8,32,70,80,us-west-2,West US 2,us-west1-b,Production,Windows,Database,PCI,,,,"Burstable,GPU",m5.2xlarge

Regional Equivalents Reference

Interpreting Multi-Cloud Output

The output CSV contains recommendation columns for every selected provider in the same row, making it easy to filter in Excel or Google Sheets:

• Sort by VM Name to see all providers for each workload
• Use the "Rules Applied" columns to understand why providers differ
• Provider pricing calculators should be used to compare actual costs — do not compare instance types directly as the vCPU/memory ratio differs across providers

File Upload Problems

Recommendation Problems

Browser / Performance Problems

📊 Data Collection

• Collect at least 30 days of utilisation data; 90 days is ideal to capture seasonal variation
• Use the P95 or P99 percentile for CPU if your workload is spiky — average metrics underestimate peak demand
• Memory utilisation from the OS is more reliable than hypervisor-reported figures — use OS-level agents where possible
• Tag VMs with their ENV, OS, and Workload in your CMDB and export that metadata directly into the CSV

🚀 Recommendation Implementation

• Start with Dev/Test environments — fast to change and low risk if something is wrong
• For Production, use Like-to-Like first to establish a safe baseline, then consider Optimized for a second wave
• Always have a rollback plan: snapshot or image the VM before resizing
• Monitor for 48–72 hours after a resize before treating it as complete

💰 Cost Strategy

• Use this tool to identify the right instance type — then separately apply Reserved Instance or Savings Plan pricing in the provider's pricing calculator
• Re-run the tool quarterly for dynamic workloads; semi-annually for stable production systems
• After a cloud migration, re-run with actual utilisation data from the new environment — the right-sizing picture often changes after migration

🔒 Security & Compliance

• The Compliance column tightens instance selection toward real hardware/software signals — it does not itself certify PCI-DSS, HIPAA, SOC 2, or FIPS compliance. Those are account- and program-level facts (your BAA, your AWS/Azure/GCP compliance program, your own controls), not properties of an instance type.
• Set Compliance = Current-Generation Hardware to keep recommendations on current-generation instances — a reasonable baseline for any regulated workload, on any of the three clouds.
• Add AWS Nitro Enclaves (comma-separated) to require Nitro-capable AWS instances — the real primitive many organizations use to isolate PCI/HIPAA-sensitive processing. Legacy PCI and HIPAA values still work and expand to both of the above automatically; legacy SOC2 and FIPS expand to Current-Generation Hardware alone.
• Add Confidential Computing for memory-encrypted instances — AWS reuses the Nitro Enclaves signal above; Azure requires its dc*/ec* confidential-VM families; GCP has no per-instance-type signal for this at all (a genuine gap in that feed, not an oversight), so the option has no effect there.
• Add Azure Trusted Launch for Secure Boot + vTPM instances (Azure only, distinct from Confidential Computing's memory encryption).
• For FIPS-boundary or data-residency requirements, manually verify that recommended regions are within your geographic compliance boundary — this is a region choice, not an instance-type one.
• No inventory data ever leaves your browser — the tool is fully client-side

🌟 General

Q: Is my data secure?
A: Yes. All processing happens in your browser — your VM inventory never leaves your machine. The tool loads a static instance catalogue at start-up; there is no backend server receiving your data.

Q: Why doesn't the output include a dollar amount?
A: Cloud pricing is not static — it changes with region, OS, discounts, Savings Plans, Reserved Instances, and enterprise agreements. Showing a dollar figure in the output would be misleading within weeks, so none ever appears — no price column, no CSV field, no export sheet. Since 3.17, the stats bar, App Portfolio Overview, and Executive Print Report show one aggregate, relative percentage per provider instead ("Optimized ranks ~X% lower than Like-to-Like"). This tool identifies the right instance type; use your provider's pricing calculator for accurate cost figures. The AWS Pricing Calculator Bulk Template export makes this workflow seamless for AWS.

Q: Can I use this tool offline?
A: Yes — after your first visit. A service worker caches the pages and any region data you've used, so those keep working without connectivity, and you can optionally install the site as an app from your browser. Regions you've never loaded still need one online visit. A page always loads the current version when you're online; region data and scripts update automatically in the background.

Q: Which file formats are supported?
A: CSV (UTF-8) and Excel .xlsx workbooks. A workbook with several sheets opens the one whose columns best match an inventory — an RVTools export lands on its vInfo sheet rather than its cover tab — and a picker lets you switch. Common header variants like vCPUs, RAM, or Hostname are mapped to the expected columns automatically.

Q: I mapped my columns wrong once and now it keeps doing it. How do I undo that?
A: When you confirm the column-mapping panel, that answer is remembered against your file's set of headers, and any later file with the same headers is mapped the same way without asking. Everything remembered is listed under Remembered column mappings in the upload section, with a Forget button on each. Forget one and the next file with those headers will ask again.

Q: I don't have a file at all. What are my options?
A: Three. Enter VMs manually takes them one at a time — and if you have several that are alike, set Copies and they are added numbered from the name you gave (web → web-01…web-05); any row can be corrected afterwards with the ✏️ button rather than deleted and retyped. Paste rows from a spreadsheet takes them straight from Excel. And under the sample template there are three ready-made datasets to load — a clean one, a 500-VM one, and a deliberately messy one that shows what the input check catches.

Q: Can I paste rows instead of uploading a file?
A: Yes — Or paste rows from a spreadsheet under the upload box. Copy the cells out of Excel or Google Sheets, header row included, and paste. It is read exactly like an uploaded file: the same column mapping, the same unit handling, the same input check.

Q: The tool says some of my rows look wrong. What is it checking?
A: After an upload, the input check names rows that will not size sensibly — a missing or zero CPU count or memory figure, a CPU count or memory size no provider sells (usually a unit that was never converted), a utilization outside 0–100%, or a blank VM name — and gives you the row numbers as your spreadsheet shows them. It is a report, not a gate: the file still loads and you can still generate. If the same VM name appears more than once, it asks whether those are one VM listed twice or different VMs that share a name, because only you know. If a column is named like one of this tool's own output columns (typically because an earlier export was uploaded again), it says so: a run ignores that column and writes its own, so the old values are never mixed into the new results. Rename the column in your file to keep its values.

Q: The input check says "How well the data supports sizing". What does High, Medium and Low mean?
A: It appears only when your file carries percentile or peak columns (p90, p95, p99 or Peak) and some VMs fall short of them. Per VM: High means every dimension the row reports (CPU, memory) has a percentile or peak reading; Medium means only one does; Low means an average only, or nothing, so bursts are invisible. It describes what the file carries, not which statistic a run will size on. The whole-file answer is separate and depends on how many VMs there are (under 25 Low, 25 to 249 Medium, 250 or more High): thousands of average-only rows make the file larger, not any row richer. A plain average-only upload does not show this note at all.

Q: Can I upload an AWS Application Discovery Service export?
A: Yes — the ADS import template is recognised and needs no mapping. Its logical-core count becomes the vCPU count (not the socket or physical-core count), RAM.TotalSizeInMB becomes memory, and memory utilisation — which ADS does not report as a percentage — is worked out from the memory used against the memory total. Without that last step your VMs could only be sized on CPU, and their memory would be left exactly as it is.

Q: Can I upload an RVTools export directly?
A: Yes. The vInfo sheet is found and read, its VM column is used for the VM name — not Host, which is the ESXi server the VM runs on — and its Memory column, which holds MiB although the header does not say so, is converted to GB. The file status confirms it was recognised as an RVTools export.

Q: My memory column is in MB. Will that work?
A: Yes. A header that says so (Memory (MB), RAM MiB) is converted, as is a format that is recognised — an RVTools export reports MiB without saying so, and that is known. When neither the header nor the format says, the values are not taken as proof: if they look like MiB, the input check asks, and nothing is converted until you answer. A fleet of genuine 512 GB machines is unusual but real, and dividing it by 1024 would be its own kind of corruption. You can also set the unit yourself in the column-mapping panel.

Q: Can I save my filter configuration for next time?
A: Yes — use the ⭐ Filter presets bar above the Generate button to save the current configuration under a name and re-apply it later. Presets are per page and stored in your browser, and can be exported as a JSON file to move them to another browser or machine (see Section 9).

⚙ Technical

Q: Why does the tool recommend a different instance than our cloud provider's own advisor?
A: Provider advisors (AWS Trusted Advisor, Azure Advisor) typically look at longer time horizons (14–90 days) and use conservative thresholds. This tool applies your chosen utilisation window and thresholds. The recommendations can legitimately differ — treat both as data points rather than absolute answers.

Q: Why are the downsize thresholds set to 40 % instead of 50 %?
A: The 40 %/80 % thresholds align with the approach used by AWS Cost Optimisation, Azure Advisor, and GCP Recommender. At 40 %, there is still comfortable headroom for traffic spikes and OS/monitoring overhead. The old 50 % default was too aggressive for most production workloads. You can adjust both thresholds in the UI if your environment requires different values.

Q: I set ENV = Production but got a t3.large recommendation. Why?
A: Check the Rules Applied column — the rule engine reports exactly what happened. If "1a: Burstable excluded" is not listed, verify that the ENV column value is exactly Production (no extra spaces or different capitalisation).

Q: Can I use a mix of ENVs in the same CSV?
A: Yes — rules are applied per row. You can have Production, Staging, and Dev rows in the same file; each gets its own rule evaluation.

Q: What happens if no instance survives after all rules are applied?
A: The tool notes the failure in the Rules Applied column with a ⚠ warning and falls back to the best available instance from the pre-rules candidate set. This is a signal to either relax filters or review whether the requirements are achievable in that region.

💼 Business

Q: What typical savings can I expect?
A: Organisations commonly see 15–40 % reduction in compute spend from right-sizing alone, before applying Reserved Instances or Savings Plans. The actual figure depends heavily on how over-provisioned the original fleet is.

Q: How often should I re-run the tool?
A: Monthly for dynamic/seasonal workloads; quarterly for stable production workloads; immediately after any significant architectural change or traffic pattern shift.

Q: Can I share results with stakeholders who don't have access to the tool?
A: Yes — the downloaded CSV can be opened in Excel or Google Sheets. The "Rules Applied" column provides plain-English audit notes explaining every recommendation, making it straightforward to present to non-technical stakeholders.

Q: Is the AWS Pricing Calculator Bulk Template ready to import immediately?
A: It populates all required columns with On-Demand, Always On, Shared Instances defaults. Storage type, IOPS and throughput are left blank for the calculator to default, and the storage amount is filled in only when your file carried a Disk (GB) column; add the rest in the calculator after import if your estimate needs it.

The Cloud Instance Recommender gives you a fast, privacy-first starting point for cloud rightsizing at any scale — from a handful of VMs to thousands. By combining flexible like-to-like and optimised recommendation modes with a configurable rule engine, it produces recommendations that are not just cheap, but appropriate for the environment, OS, workload type, and compliance posture of each individual VM.

Key things to remember:

• Add ENV, OS, Workload, and Compliance columns to get environment-aware recommendations with a full audit trail
• Use the 40 %/80 % thresholds as a starting point — adjust them based on whether your utilisation data is averages or percentiles
• No absolute dollar figure is ever in the output — a relative savings percentage is shown since 3.17, and the AWS Pricing Calculator Bulk Template export is still the way to get an authoritative cost estimate
• Treat recommendations as an informed starting point, not a final answer — validate in a non-production environment first
• Re-run quarterly to keep pace with changing workloads and the cloud providers' expanding instance catalogues