PingCloud Help Center

Quick Start

Three minutes, three common tasks. All five tools live on the home page — the links below jump straight to each one.

Pick the best cloud region for your users

1. Open the Benchmark leaderboards, choose your user country and a 7- or 30-day window, and note the top regions by median latency.
2. Open the Cost & Latency Explorer, select the same country plus the vCPU / memory you need, and compare monthly cost against measured latency on the scatter plot.
3. Check the Pareto frontier: points on the frontier have no candidate that is both cheaper and faster. Pick A / B in the table to quantify the trade-off.
4. Before committing, run a live Cloud Ping Test from one or two of your user cities to confirm current conditions.

Compare prices for the same spec across clouds

1. Open Cloud Pricing, choose Cloud Servers or Cloud Disks, and set the spec filters (vCPU, memory, OS, class).
2. Use Exact Match to compare identical specs, or Similar Match to widen the net — differences are always shown per field.
3. Adjust monthly hours to your real usage: monthly cost = hourly price × usage hours. Select rows and compare to chart them.

Run a live latency test right now

1. Open Cloud Ping Test, pick a source (Global, region, country or city), a vendor and a region — or enter any custom host.
2. Choose ICMP Ping or HTTP, set the probe limit, and start. Results stream in live: MIN / AVG / MED / MAX and packet loss, or the full HTTP timing breakdown.
3. Want the view from your own machine? Use Cloud Regions & HTTP Ping — it measures browser-to-region HTTP latency directly.

Product Introduction

PingCloud is a public benchmark for multi-cloud network performance. Around the clock, it measures latency from more than 1,000 probe cities in 115 countries to 233 regions across 7 major cloud providers — AWS, Microsoft Azure, Google Cloud, Alibaba Cloud, Tencent Cloud, Huawei Cloud and Oracle Cloud — and publishes the results as daily rankings, live tests and an open API.

Latency alone is half the decision — the other half is cost. PingCloud also collects on-demand prices for cloud servers and cloud disks from official public price lists, and joins them with measured latency so you can compare regions on both axes at once.

Who it is for

• Cloud architects — choose and justify regions for new services with cross-vendor latency and cost evidence.
• Network & SRE engineers — monitor network quality between your users and your regions, and catch regressions early.
• FinOps & platform teams — compare list prices across providers and regions on a like-for-like spec basis.
• Developers & researchers — build on the public API and daily ranking data, free, with attribution.

What makes it different

• Real user networks: measurements run on the Globalping volunteer probe network — real ISP-level paths in 1,000+ cities, not synthetic datacenter-to-datacenter tests.
• Fair by construction: within each test cycle, the same probe measures every vendor's regions with identical parameters — same vantage point, same method.
• Rigorous statistics: 15 ICMP packets per measurement, medians computed from raw samples, sample-count-weighted aggregation, high-loss probes excluded.
• Official prices only: every stored price is traceable to an official public source — never estimated, never sampled.

The Five Tools

All five tools live on the home page. Each answers a different question — pick by the decision you are making.

How It Works

From probe to ranking, every number on PingCloud goes through the same pipeline.

1. Scheduling

The collector runs continuous refresh cycles. Each cycle re-tests the pairs that matter most for the rankings — top pairs more often, long-tail pairs less often — and samples up to 10 cities per country per round, probe-weighted at random. Across a multi-day window, successive rounds gradually cover more cities and more networks.

2. Measurement

Each measurement sends 15 ICMP packets from one probe to one cloud region endpoint. Within a cycle, the same probe measures every vendor's regions with identical parameters — this is what makes cross-vendor comparison fair. If a location temporarily has no online probes, the pair is retried and picked up again as soon as probes return.

3. Statistics

Median, average and jitter are computed from the full set of raw per-packet RTT samples — not taken from a pre-computed summary — so a single network hiccup cannot skew a result. Packet loss is the share of the 15 packets that never returned.

4. Aggregation

Per city and region, results are averaged over the window. City averages are then combined into a country-level number weighted by sample count, so a city with 10x more data has 10x the influence. Data from cities whose window packet loss exceeds 50% is excluded as unstable.

5. Rankings

Rankings are generated for 1-, 3-, 7-, 15- and 30-day windows: per country, the top 20 vendor regions by weighted median; per vendor region, the top 20 countries. Files are regenerated daily, and every entry carries its sample count, city count, ASN count and a day-by-day series so you can judge confidence yourself.

6. Pricing

A separate pipeline collects on-demand list prices from each provider's official public price API, normalizes them into one schema (vCPU, memory, architecture, class, USD per hour or per GB-month), and joins them to latency data by the unified region ID — the same region list the ping test uses.

Data & Methodology

Exact definitions of every metric you see, so numbers from different tools never leave you guessing.

Metric definitions

Time windows

1-day shows the freshest picture but is noisiest; 30-day is the most stable but reacts slowly to real changes. 7-day is a good default for decisions. Compare 7-day against 30-day — a region that climbs in the short window is worth watching.

Why P95 is N/A

P95 requires the raw per-packet samples of every measurement. PingCloud persists per-measurement summaries (median, average, loss), not the raw timings, so a true P95 cannot be computed — and it is never estimated. The Explorer offers average and p50 (median) only, for the same reason.

Update cadence

The collector runs continuously; ranking files are regenerated once a day, and the leaderboard shows the exact publish time. Prices are re-synced on a rolling schedule; a price older than its provider's staleness threshold is flagged as stale but still shown, with its collection date.

Reading the numbers correctly

• A country entry is an aggregate across cities and networks — it does not promise every ISP in that country sees the same latency. Use the live test to check your specific path.
• Rankings compare network latency only. They do not weigh service maturity, compliance, available instance types or reliability — fold those in yourself.
• List prices are not your prices: negotiated discounts, reserved capacity and savings plans are invisible to a public benchmark. Treat PingCloud prices as a comparable baseline, not a quote.

Global Probe Network

PingCloud's measurements run on the Globalping probe network — thousands of volunteer-operated probes embedded in real home, office and ISP networks across more than 1,000 cities. Because they sit on real user networks, the numbers reflect the paths your users actually traverse, not a datacenter cross-connect.

Current scale: 1,068 active probe cities across 115 countries, roughly 5,300 online probes. The live list is re-synced from Globalping every 3 days; cities whose probes go offline drop out of new measurements and return automatically when their probes come back.

Why ASN diversity matters

Two probes in the same city can sit on networks with wildly different international routing. That is why every ranking entry reports its ASN count, and why per-round city sampling is probe-weighted — the benchmark deliberately spreads measurements across networks so a single ISP's bad day cannot dominate a country's number.

Test quotas

The live Cloud Ping Test uses Globalping's shared rate limits: 100 tests per hour per IP anonymously, 250 per hour when authenticated. The quota display on the page shows your real remaining balance in real time. Collector-side measurements run on a separate, much larger schedule and never compete with your interactive tests.

Pricing Data

One hard rule: every stored price is traceable to an official public source. Each price row carries its source URL, collection time and effective date. If a provider publishes no price for a region, that region shows no price — estimated or sample numbers are never stored.

Sources by provider

Scope and conventions

• Prices are on-demand (pay-as-you-go) list prices, in USD, for Linux unless stated otherwise; Windows license surcharges are collected where published.
• Cloud servers are priced per hour; monthly cost = hourly × usage hours (default 730, adjustable). Cloud disks are priced per GB-month on one unified axis; monthly cost = per-GB-month × capacity (default 100 GB, adjustable).
• Non-USD display currencies use ECB reference rates (via frankfurter.dev); the original USD value is always kept.
• Excluded: network egress, support plans, reserved instances and savings plans, negotiated enterprise discounts. Region coverage mirrors the ping test's region list — the two tools can never drift apart.

Staleness

Each provider has a staleness threshold. Prices older than the threshold are marked stale and shown with their collection date, so you always know exactly how fresh a number is — nothing is silently refreshed or hidden.

Cost & Latency Methodology

The Explorer answers one question: for users in a given country and a given spec, which provider-region candidates are worth shortlisting? It plots monthly cost against measured latency and computes the Pareto frontier.

What is a candidate

One candidate = one (provider, region) pair, represented by its cheapest exactly-matching SKU. Exact match is required because Pareto comparison is only meaningful for the same resource spec — similar matches are not performance-comparable.

The latency axis

Same aggregation as the leaderboards: per-(city, region) window averages, test-count-weighted across cities, high-loss cities excluded. Windows of 24 hours, 7 days or 30 days; metric of average or p50. It is a country-level aggregate, not a single city or ISP result.

Pareto frontier, precisely

Candidate A dominates B if A's cost is less than or equal to B's AND A's latency is less than or equal to B's, with at least one strict inequality. The frontier is the set of candidates no other candidate dominates; a dominated candidate reports which candidate dominates it. Candidates with fewer than 3 samples in the window, or missing price or latency, are excluded from the frontier — never guessed.

Using it well

The frontier is an analysis, not a recommendation: a point on it can still be the wrong choice if the region lacks a service you need, fails a compliance requirement, or has a reliability history you cannot accept. Use the A/B trade-off panel to quantify what you would pay for the latency you save — then decide with the full picture.

API Reference

PingCloud exposes a free public REST API — no key, no registration. All endpoints return JSON; errors use one shape: {"error": {"code", "message"}}. Rate limit: 60 requests per minute per IP; a 429 response carries Retry-After.

Pricing API

Cost & Latency Explorer API

Examples

# Cheapest 4 vCPU / 8 GiB Linux instances in Japan
curl "https://pingcloud.io/api/v1/pricing/skus?service_category=compute&vcpu=4&memory_gib=8&operating_system=linux&countries=JP&sort=price_hour&order=asc&page_size=5"

# Cost x latency candidates for users in Germany, 7-day window
curl -X POST "https://pingcloud.io/api/v1/explorer/cost-latency/query" \
  -H "Content-Type: application/json" \
  -d '{"source_country": "DE", "vcpu": 4, "memory_gib": 8, "window_hours": 168}'

Ranking data files

The leaderboard's source files are public static JSON, refreshed daily: /static/data/rankings/region_ranking__p{N}d.json, country_ranking_..., geo_region_ranking_..., plus periods.json which points to the current date. N is one of 1, 3, 7, 15, 30. Each entry carries rank, median_ms, avg_ms, loss_pct, city_count, test_count, asn_count and a daily series — ideal for research and offline analysis.

Glossary

Citation & Data Use

Rankings, measurements and API responses are free to use for research, analysis, engineering and editorial purposes, with attribution. Suggested citation:

PingCloud.io — Multi-cloud region latency benchmark,
retrieved 2026-10-27, https://pingcloud.io

When publishing derived rankings or charts, link to the page or API URL you used and keep the retrieval date — the data refreshes daily. Bulk redistribution of the raw files as a dataset, or commercial use at scale, needs a quick word with us first — so we can keep the service free and fast for everyone.

FAQ

Contact & Feedback

Data anomalies, missing regions, API feature requests and partnerships — all welcome.

• Email: [email protected]
• GitHub: github.com/lexcodee/pingcloud.io