Hong Kong Vps $2 Network Quality Bandwidth And Packet Loss Rate Measurement Report

this article conducts a systematic network quality measurement on a low-priced cloud host with a price of about us$2 for hong kong vps : using iperf3, ping, mtr and other tools to repeatedly measure at different times and on multiple lines. the results show that this solution has usable throughput performance locally and in neighboring countries, but there are obvious fluctuations and packet loss for different operators and peak hours in mainland china. the overall network quality is suitable for lightweight applications (reverse proxy, lightweight website, vpn testing and development environment), businesses with high real-time requirements need to be carefully evaluated.

in multiple iperf3 long connection (60s) measurements, the average measured upload/download peak value for local nodes in the same city in hong kong fluctuated between 200-600 mbps (the supplier claimed that it was usually shared 1gbps), and the delay network quality was local rtt 1-5 ms, and the packet loss rate was close to 0%. for neighboring countries: the average throughput in the singapore/japan direction is 30–120 mbps, rtt 30–70 ms, and the packet loss rate is usually <0.5%. the tests for the three operators in mainland china are quite different: the telecom direction averages 20–60 mbps, rtt 25–60 ms, and the packet loss rate is 0.5–3%. the fluctuations in the china unicom/china mobile direction are more obvious, with some nodes experiencing 1–6% packet loss and jitter at the peak.

this actual test used three combinations of iperf3 (tcp/udp long connection), speedtest-cli (for public speedtest servers), ping and mtr (continuous route tracing). each set of tests was run 3–5 times in different time periods to obtain the average value. in order to reduce occasional interference, the test included weekday peak hours (12:00–14:00, 19:00–23:00) and off-peak hours (02:00–06:00), and compared multiple target nodes (operator nodes in the same city, singapore, japan, us west coast, guangzhou/shenzhen/shanghai/beijing), and the test period was one week.

there are obvious levels by region: intra-city/hong kong to hong kong pop performance is the best, with the lowest latency and almost 0 packet loss; southeast asia and japan are second best, with rtt of 30–100 ms, and throughput is acceptable; west america/europe has more transoceanic lines and transfers, rtt is generally 120–220 ms, and the packet loss rate increases slightly but is usually <1.5%. the difference is most obvious in mainland china. telecom links are relatively stable but peaks drop significantly. china unicom/china mobile are often affected by backhaul bgp, interconnection peering and gfw policies, resulting in packet loss or path jitter.

in actual observations, the main locations with higher packet loss rates include: first, the border routers of cross-border links (there will be queues at night or during peak hours, resulting in packet loss); second, the targets are some secondary operators in mainland china or city nodes with poor interconnection quality at the computer room exits (such as some china unicom/mobile nodes); third, the supplier's own network speed limit or port overbooking leads to short-term packet loss and bandwidth limitation. mtr route tracing can clearly locate the majority of packet losses in a certain hop cluster or peer routing policy.

the main factors causing fluctuations are: 1) bandwidth oversubscription and shared bandwidth policies common in low-price products lead to competition for resources during peak hours; 2) the provider's backbone and upstream bearer path quality and peering policies affect cross-border performance; 3) virtualization and host load: a large number of neighbor instances on the same host using large traffic at the same time will cause jitter; 4) access to mainland china is affected by intermediate devices (gfw, route filtering, deep packet inspection); 5) local traffic shaping or qos policies are handled differentially over udp/tcp.

purchasing advice: prioritize real reviews and the bandwidth guarantee provided by the provider (guaranteed bandwidth or burst policy), and choose a provider with multiple upstreams or good peering. you can request a trial or long-term (peak/off-peak) speed test during the refund period before purchasing. usage level: 1) if you need stable access to the mainland, consider deploying multiple exits at the same time or using a bgp/multi-line solution; 2) enable tcp acceleration, compression or maintaining a long connection to reduce retransmission overhead; 3) monitor traffic and delays (such as using prometheus/grafana + ping/iperf automation scripts), discover and change lines or computer rooms in a timely manner; 4) for real-time audio and video or game services, it is recommended to choose mid-to-high-end packages with sla and minimum packet loss guarantees.