Comprehensive Evaluation Report Malaysia CN2 Covers Packet Loss Jitter And Packet Loss Recovery Capability

Overview of the comprehensive assessment

This report focuses on Malaysia's CN2 link, conducting a comprehensive performance comparison on the server side, covering packet loss, jitter, and packet loss recovery capabilities. We evaluated the performance of three categories: "Best (highest performance)," "Best (cost-effective)," and "Cheapest (lowest cost)" in real business scenarios, helping operations and architecture teams balance stability and cost.

Test objectives and scope

The tests focus on external server connectivity: including unidirectional/bidirectional latency, latency jitter, instantaneous packet loss rate, packet distribution (random and burst packet loss), and recovery efficiency under packet loss conditions (such as TCP retransmission, application layer FEC, and fast rerouting). The target is CN2 evaluation links stationed at Malaysian export points or transiting through Malaysia.

Test environment and topology

The hardware layer uses multiple cloud servers (installing the standard Linux kernel), declares multiple outlets via BGP, and compares CN2 lines with regular public network lines. Testing tools include iPerf3, mtr, ping, tcptraceroute, netem simulation packet loss/latency, as well as self-developed script records for retransmission and throughput fluctuations, ensuring fine-grained observation of server behavior.

Key indicator definitions

This evaluation uses the following key metrics: average RTT, RTT standard deviation (measuring jitter), instantaneous packet loss rate (short-term window), long-term packet loss percentage, distribution of burst packet loss duration, and packet loss recovery time (from the first packet loss to throughput stabilization). These metrics are directly related to service experience and the design of retry/timeout strategies.

Packet loss and jitter experimental design

In the comparative experiment, we injected 0.1%-5% random packet loss and burst packet loss (continuous loss of 4~20 packets) at different loads and time periods (peak/non-peak), recording throughput, retransmission counts, and connection disconnection rates for each path, thereby evaluating the impact of packet loss and jitter on real services (such as TCP downloads, real-time voice/video).

Methods for evaluating packet loss recovery capability

Packet loss recovery capability depends not only on the underlying packet loss rate but also on the recovery efficiency of the protocol and link layers. We measured TCP RTO trigger frequency, SACK/retransmission efficiency, QUIC recovery speed under packet loss, and both subjective and objective improvements after enabling FEC at the application layer. Special attention is given to whether the CN2 link on Malaysian nodes has additional packet loss repair mechanisms or shorter packet loss recovery times.

Summary of test results

Overall, the high-quality direct connection route via Malaysian CN2 outperforms the regular public network relay route in average latency and jitter, especially maintaining stable performance during peak periods. Random low packet loss (<0.5%) has limited impact on TCP throughput, but burst packet loss significantly impacts short connections and real-time applications. Packet loss recovery capability is strongly related to server TCP stack parameters and congestion control algorithms (such as BBR vs CUBIC).

Comparison of different options (best / best / cheapest).

"Best" solution: Choose servers with CN2 GIA/high-quality direct connection and global redundant exports, which are more expensive but have minimal latency and jitter, and fastest packet loss recovery; "Best" solution: cost-effective CN2 circuits + local optimization (kernel spec tuning, FEC), suitable for most services; "Cheapest" solution: Use a standard public network or low-cost data center, and compensate for packet loss and jitter fluctuations through retrys, extended timeouts, and application-level fault tolerance.

Optimization suggestions for servers

To improve performance on Malaysian CN2 links > , it is recommended to enable SACK and TCP Fast Open, adjust TCP retransmission and RTO strategies, prioritize BBR congestion control (beneficial for high-bandwidth latency products), and adopt FEC and adaptive bitrate in real-time applications. It also enables heartbeat and rapid reconnection mechanisms for long connections to reduce the impact of sudden packet loss.