Singapore Telecom Computer Room Cn2 Comparison Test Bandwidth And Packet Loss Rate In-depth Evaluation

Based on multi-period and cross-network actual measurements, this article compares the differences in throughput and packet loss performance of several common links in and out of Singapore, and provides reproducible test methods and selection suggestions under different business scenarios to facilitate operation, maintenance and procurement decisions.

What are the testing environments and methods?

The test was conducted between three cloud hosts: one in the Singapore computer room (rented from the telecommunications computer room), and two in domestic and Hong Kong nodes. Tools employ iperf3 (bandwidth), mtr (routing and loss paths), continuous ping (loss and delay distribution), and Speedtest (supplementary verification). Each link is sampled for 1 hour during peak/off-peak periods. iperf3 uses a multi-stream concurrency test within 1 minute to simulate concurrent services; the packet loss statistics are based on the 95th percentile and the maximum observed value.

Where was the bandwidth and packet loss comparison test conducted?

Comparison objects include: CN2 dedicated lines provided by telecommunications equipment rooms, conventional operators' BGP ordinary exports, and some international direct connections. The test nodes are located at the Singapore computer room exit, Hong Kong transit node and domestic backbone node to cover the differences in common paths from Singapore to China.

Which computer room or line has better bandwidth performance?

The average results show that CN2’s one-way throughput is about 10%~30% higher than ordinary BGP on average. Specific measured data examples (concurrent 8-stream iperf3): CN2 average download 480~520 Mbps, ordinary BGP average 350~430 Mbps; during peak congestion periods, CN2 has small fluctuations and lower standard deviation, indicating that link queuing and jitter are better controlled. It should be noted that the actual bandwidth is affected by the peer release window, host network, and intermediate link speed limit.

Why do packet loss rates vary greatly on different links?

The differences in packet loss are mainly due to routing policies, bandwidth congestion control, network priorities, and queue policies of intermediate forwarding devices. CN2 usually adopts more direct backbone interconnection and QoS implementation, and the packet loss rate can be maintained at 0%~0.1% in most time periods (except for short-term bursts); when ordinary BGP is under heavy pressure, the packet loss can rise to more than 0.5%. Another common reason is that the link has a traffic shaping or burst loss protection policy, which results in intermittent high packet loss.

How to measure packet loss rate and delay to be more reliable?

It is recommended to adopt a multi-tool and multi-window strategy: 1) Use iperf3 to measure throughput under different concurrency numbers and observe retransmits; 2) Use mtr to sample paths for more than 10 minutes to locate which hops experience packet loss; 3) Use ping to sample high-frequency packets (such as 100ms) and count quantile delays; 4) Record different time periods (peak days on weekdays/night) to eliminate sporadic interference. Focus on the 95th/99th percentile rather than just the average.

How to choose whether to use the CN2 line based on the test results?

If the business has high real-time and stability requirements (VoIP, games, financial transactions, two-way video, etc.), give priority to CN2 to reduce the risk of jitter and packet loss. If only large file transmission or cost-sensitive services are required, the line level can be reduced as long as the bandwidth is satisfied and short-term packet loss is allowed. The final selection should be weighed against SLA, route visibility, and price.

What costs and deployment considerations need to be considered?

CN2 is usually more expensive than ordinary BGP, but the predictability and low packet loss it brings are of significant value to critical businesses. Note when deploying: 1) Confirm the interconnection points and redundant paths with the computer room; 2) Test whether the routing is stable in different time periods (whether there are frequent detours); 3) Monitor the real-time packet loss and RTT anomalies of the link, and set alarms; 4) Consider backup links and traffic diversion strategies to prevent single link failure.