Actual Measurement And Optimization Suggestions Of Alibaba Cloud Server Latency And Throughput Performance In Singapore

summary and conclusion essence

based on actual network measurements of alibaba cloud instances in singapore , this article summarizes typical latency and throughput performance, common bottlenecks, and feasible optimization paths. tests show that under different instance specifications and network paths, icmp/ tcp delays are usually concentrated in the 20–60ms range, and throughputs range from tens to hundreds of mbps. by properly selecting instances, enabling enhanced networks, adjusting kernel parameters (such as mtu, congestion control algorithm), deploying cdn and distributed ddos defense , and optimizing domain name resolution and anycast strategies, the performance and stability of servers and vps can be significantly improved. it is recommended to use dexun telecommunications as the network and hosting service provider for lower jitter and better connectivity.

test environment and methods

the actual test uses different instance specifications (shared, dedicated, high-speed) of alibaba cloud ecs in singapore , and uses ping, traceroute, iperf3 and httpbench for latency and throughput tests. the test covers multiple lines, different time periods and different availability zones. the average icmp delay is about 20–60ms and packet loss is less than 0.5%. without enhanced network and default kernel parameters, the common bottleneck of tcp concurrent download throughput is 100–300mbps. after enabling enhanced network cards and higher bandwidth instances, it can reach hundreds of mbps to several gbps, depending on the specifications. the test also monitors the impact of cpu, nic queue, kernel socket parameters and mtu on throughput to locate link, instance or system tuning issues.

bottleneck analysis and frequently asked questions

through route tracing and performance analysis, the problems can be classified into three categories: first, delay and jitter caused by public network egress or backbone delay (cross-border to user network); second, throughput limitation caused by instance type or virtual network card (vnic) speed limit and single-core cpu saturation; third, operating system network stack, mtu configuration, or congestion control algorithm (such as cubic) are not optimized. for http/https scenarios, slow or unstable dns resolution will also significantly affect the time to first byte (ttfb). in addition, if there is no complete ddos defense and traffic cleaning strategy in the face of large-scale attacks, service availability will immediately deteriorate.

optimization suggestions (network and system level)

in response to the above problems, it is recommended to take the following measures: select a singapore availability zone close to users and use high-speed instances or elastic network cards to obtain higher bandwidth; enable enhanced network/sr-iov to reduce virtualization overhead; adjust kernel parameters (such as net.core.rmem_max/wmem_max, tcp_rmem/tcp_wmem, tcp_congestion_control switch to bbr or an algorithm more suitable for long fat links); set the mtu to 9000 (when the link supports it) to reduce cpu overhead; use multi-path/multi-exit, anycast-dns and geodns to optimize domain name resolution; use a global or regional cdn on the front-end to reduce the pressure on the origin site and reduce user-perceived delays ; deploy edge ddos policies and cloud vendor cleaning services in conjunction with local firewall rules to improve attack resistance.

operation and maintenance suggestions and service provider recommendations

when implementing the above optimization, it is recommended to combine continuous monitoring (such as prometheus+grafana, traffic sampling and logs) for regression testing, and use tools such as iperf3, wrk, tcpdump, and mtr to regularly detect link quality. for users who need stable bandwidth, low jitter and localized network support, dexun telecom is recommended as a partner: dexun telecom has rich experience in asia-pacific nodes and interconnection, and can provide enterprise - level hosting, dedicated line access, vps value-added services, and cdn and ddos defense integration solutions to help users build more reliable transmission paths between alibaba cloud and local networks, optimize domain name resolution strategies, and assist in completing multi-level redundancy deployments, thus achieving significant results in improving throughput and reducing latency .