Evaluation Of Real And Fake Vietnam Servers, Multi-dimensional Comparison Of Real Latency And Bandwidth Performance

1.

test background and purpose

- the purpose of this evaluation is to verify the statement "whether the vietnam server is real and how the latency and bandwidth performance are".
- coverage scenarios: measure latency and bandwidth from domestic (beijing, shanghai, guangzhou) and asia-pacific neighboring nodes (hong kong, singapore) to vietnam (hanoi, ho chi minh).
- key indicators: average rtt (ms), jitter (ms), packet loss rate (%), tcp/udp throughput (mbps), port upper limit and concurrency capability.
- testing tools: ping/mtr, iperf3 (tcp/udp), curl download, web stress test and traceroute.
- the evaluation also includes the delay before and after cdn access, and the provider's ddos protection capabilities (cleaning bandwidth and strategies).

2.

test samples and server configuration (real case)

- case a: vps (kvm) in ho chi minh city, vietnam, configuration: 4 vcpu / 8gb memory / 100gb nvme / 1gbps dedicated line, computer room: vnpt computer room.
- case b: shared vps in hanoi, vietnam, configuration: 2 vcpu / 4gb / 50gb ssd / 100mbps shared port, computer room: fpt idc.
- case c: dedicated physical machine in vietnam's danai data center, configuration: 8-core xeon / 32gb / 2x1tb ssd / dual 1gbps bgp, including independent public network ip and hardware protection.
- test point sources: beijing unicom computer room, shanghai telecom computer room, guangzhou mobile node, hong kong isp, singapore data center.
- test time window: peak (12:00-14:00) and non-peak (02:00-04:00), each period of continuous testing for 30 minutes to take the average and 95th percentile.

3.

measured data on latency and packet loss (summary)

- the table lists the average rtt, jitter and packet loss rate from each test point to different vietnam computer rooms (sample data).
- note: the value is the combination of the icmp average value and the mtr path detection results within 30 minutes of each point.
- observation points: it is usually less than 50ms from the southern node in the country to ho chi minh, and about 40-70ms from the north to hanoi. the packet loss rate is between 0% and 1%.
- jitter is in the range of 0.5-5ms, and jitter and short-term packet loss will increase during cross-border peak hours.
- the following is a partial test result table (unit: ms / %):

source	purpose	average rtt	jitter	packet loss rate
beijing	hanoi (case b)	68	3.2	0.5%
guangzhou	ho chi minh (case a)	42	1.1	0.2%
hongkong	ho chi minh (case c)	28	0.6	0.0%

4.

actual measurement of bandwidth and throughput (iperf3 and http download)

- case a (1gbps port) was tested in the same area. the peak tcp throughput was about 950mbps (iperf3), and the non-peak value was stable at 700-900mbps.
- case b (100mbps sharing) tcp download peak value is about 85-95mbps. the actual http download jitter is large, and there are occasional sudden jitters.
- case c (dual 1gbps bgp) can stably maintain more than 900mbps in the multi-threaded tcp test, and supports 10 concurrent streams totaling more than 1.5gbps short-term bandwidth.
- udp test: without packet loss, case a can reach 900mbps udp with jitter <2ms; shared vps experiences 5-10% packet loss when concurrency is high.
- conclusion: the caliber (dedicated line/shared), computer room interconnection and peering relationship determine bandwidth stability. 1gbps dedicated line can be close to line speed, but 100mbps sharing is obviously limited.

5.

cdn and ddos protection comparison experiment

- cdn acceleration: after connecting to domestic/vietnam local cdn nodes, the beijing->ho chi minh static resource delay dropped from 68ms to 30-40ms, and the time to first byte ttfb dropped by about 40%-60%.
- bandwidth offloading: in the large file distribution test, the back-to-origin bandwidth dropped by more than 75% using cdn, and the bandwidth pressure on the origin site was significantly reduced.
- ddos protection: the provider in case c claims that the cleaning capability is 100gbps. in actual testing, when simulating a small-scale udp amplification attack (10gbps), the origin site can maintain normal business after cleaning; the shared vps does not have independent cleaning and is easily affected by black holes or traffic restrictions.
- strategy comparison: the solution with traffic cleaning + bgp anycast is better than the single black hole strategy when there is continuous large traffic.
- recommendation: for external services, it is recommended to deploy local cdn nodes and choose a supplier with at least 10-100gbps cleaning capabilities to avoid relying solely on small computer rooms with their own firewalls.

6.

comprehensive conclusions and deployment suggestions

- conclusion summary: vietnam servers are "really usable", but performance strongly depends on the quality of the computer room interconnection, port type (dedicated line vs shared) and the supplier's bgp/backbone connection.
- recommended scenario: for sites targeting vietnamese/southeast asian users, priority should be given to local computer rooms in ho chi minh/hanoi + local cdn.
- configuration recommendations: for those with high business stability requirements, choose a dedicated line of 1gbps or above, starting with 4-core 8gb, and cooperate with domestic exports and multi-line bgp.
- security recommendations: key businesses should purchase ip with cleaning capabilities or use cloud protection services, and specify sla and cleaning thresholds in the contract.
- real case reminder: after an e-commerce customer switched to case c and connected to the local cdn, the pv delay during peak activity dropped by 45%, the payment success rate increased by 3 percentage points, and there was no downtime due to attacks.