Performance Evaluation of Alibaba Cloud Hong Kong, China Node: Late Peak Packet Loss Rate and BGP Optimized Line Measurement

For many mainland webmasters and enterprises that do overseas business, cross-border e-commerce, or do not want to go through the cumbersome ICP filing process,

Alibaba Cloud Hong Kong, China Node

It has always been "sweet".

However, in major technical forums, the evaluation of Hong Kong nodes is always polarized: some people praise it as "as low as 20ms delay, comparable to intranet", while others complain that "it gets stuck as a dog and loses packets seriously at the peak of the evening".

Is Aliyun's BGP line not enough, or is your configuration and network wrong?

As an SEO website optimizer and veteran of operation and maintenance, I set up a special test node to continuously monitor Aliyun Hong Kong nodes for a week, focusing on in-depth measurement of network packet loss rate, delay jitter and actual performance of BGP lines during the evening peak (20:00-23:00).

1. test environment and measured configuration

In order to ensure the authenticity and objectivity of the test data, we avoid using a single Web page for speed measurement, but adopt a multi-node, multi-tool collaborative approach.

1. Test machine specifications

Instance type: Alibaba Cloud Hong Kong node ECS (computing/lightweight application server)

Network type: VPC

Public network bandwidth: fixed 10Mbps/peak 30Mbps

Operating System: Ubuntu 22.04 LTS

Test tools: MTR (My Traceroute), Iperf3, BestTrace, Speedtest-cli, Ping.pe

2. Test period division

Off-peak: 14:00 - 16:00 (low network congestion, baseline reference)

Late peak: 20:30 - 22:30 (peak congestion of international export backbone network, core assessment)

How to get to the Hong Kong BGP line in 2?

Before looking at the measured data, let's first understand the Aliyun Hong Kong node

Network base

.

Ordinary Hong Kong small computer rooms are often only connected to single-line or dynamic BGP, which is easily squeezed at the entrance and exit of the backbone network. While the Alibaba Cloud Hong Kong node mainly relies on its powerful

BGP multi-line optimized network

.

According to the Traceroute Routing (MTR) results, the network link of the Alibaba Cloud Hong Kong node has the following characteristics:

Telecommunications lines: both outbound and backhaul optimize the backbone nodes of telecommunications, and some high-quality lines or cloud products are directly connected with high-quality nodes, greatly shortening the number of inter-provincial routing hops.

Unicom line: take Unicom CUG/CUVIP(10010/4837 high-quality backbone network), directly connect to Guangzhou/Shenzhen export, delay performance is extremely stable.

Mobile line: mobile CMI is directly connected to the local POP point in Hong Kong, which is the line with the lowest delay and the most full bandwidth among the three major operators.

Analysis of 3. Core Measured Data

1. Comparison of Ping Delay in Many Plays in China

We selected four generations: South China (Guangzhou), East China (Shanghai), North China (Beijing), and Southwest (Chengdu).

Table node, which performs 1000 consecutive Ping package tests on Alibaba Cloud Hong Kong nodes.

Region/City

Operator

Average Daytime Delay (ms)

Average Late Peak Delay (ms)

Late Peak Delay Jitter Range (ms)

South China (Guangzhou)

Telecom/Unicom/Mobile

12.4

15.8

12 - 25

East China (Shanghai)

Telecom/Unicom/Mobile

31.2

36.5

30 - 48

North China (Beijing)

Telecom/Unicom/Mobile

42.8

48.1

42 - 65

Southwest (Chengdu)

Telecom/Unicom/Mobile

38.6

44.2

38 - 58

SEO observation: for websites deployed in hong kong, the access speed of users in the south is almost the same as that of domestic double-line computer rooms. The delay of users in the north is controlled within 50ms, which is fully in line with Google and Baidu's hard performance index of "high-quality first package response time (TTFB <200ms).

2. Measurement of packet loss rate in late peak (core hard test)

Many overseas servers run very fast during the day, and serious web pages cannot be opened and load overtime occur at 8 pm. This is often

Backbone network export packet loss

Caused by.

We continuously trace the MTR link at 21:00 in the evening peak and send 5000 ICMP/UDP packets in total. The results are as follows:

Plaintext

Host Loss% Snt Last Avg Best Wrst StDev

1. 10.0.0.1 0.0 500 0.4 0.4 0.3 0.8 0.1

2.183. x.x.x (Guangdong Telecom Backbone Node) 0.0 500 3.2 3.5 2.8 8.1 0.6

3.2. x.x.x (telecommunications international exports) 1.2 500 14.1 15.2 12.5 35.0 2.1

4.7. x.x.x (Aliyun Hong Kong Portal) 0.4% 500 15.8 16.1 14.2 28.4 1.2

Measured summary:

Ordinary BGP lines: During the evening peak hours, the comprehensive average packet loss rate of the three major operators is controlled between 0.5 and 2.1. Compared with the "cliff" packet loss of up to 15%-30% in some small computer rooms in the market, Aliyun's performance

Very tough.

Mobile and Unicom users: the evening peak performance is almost perfect, the packet loss rate is close to 0%.

Telecom users: there are occasionally 1% ~ 2% slight packet losses at international exits, but they are protected by TCP retransmission mechanism, and the front-end web page loading is almost imperceptible to Caton.

3. Bandwidth Run-to-Full Rate and Throughput (Iperf3 Measured)

We use Iperf3 to test single-threaded and multi-threaded TCP throughput at 21:30 during the evening peak (public network limit 10Mbps):

Single-Threaded Downstream: Full 9.6 Mbps (96% of theoretical peak)

Multi-threaded concurrency: stable at 9.8 - 9.9 Mbps

Web Real Load (TTFB): Using the uncached WordPress home page test, the response time of the first byte of Guangzhou node is only 110ms, while that of Shanghai node is 145ms.

4. webmasters and business operations: how to maximize cost reduction?

The performance of Aliyun Hong Kong, China node is of course out of the group, but the price is slightly higher than that of ordinary nodes and non-optimized lines in mainland China. Especially when bandwidth upgrades (such as more than 5M) or high-end ECS instances are selected, the monthly budget tends to increase sharply.

From the perspective of SEO website operation and operation and maintenance cost control, there are two very practical money saving and optimization techniques:

1. Architecture level: lightweight application server + CDN / OSS combination

If your business is mainly a blog, an enterprise's official website or a small and medium-sized foreign trade station:

Hong Kong Lightweight Application Server is preferred: it usually comes with large bandwidth (such as 30Mbps peak) and fixed traffic packets, and the cost performance is much higher than that of ECS instances with the same bandwidth.

Static and dynamic separation: mount static resources such as pictures, JS, CSS, etc. to ariyun OSS, and turn on CDN acceleration to consume most of the traffic on cheap CDN, leaving the host only for pure API/dynamic rendering.

2. Capital level: use official agent channels (Aliyun recharge)

Many novice webmasters and enterprises purchase Aliyun servers directly from the original price of official website, or only receive a few small coupons, which virtually wastes a lot of budget.

In fact, a very common way to reduce costs and increase efficiency in the industry is to contact.

Alibaba Cloud official authorized service provider (agent)

. Through agents

Alibaba Cloud Recharge

, can enjoy the official original price basis

Additional discounts and exclusive cashback

.

💡what is alibaba cloud recharge? alibaba cloud authorized agents will provide recharge services for their associated accounts. Users do not need to change the original Alibaba Cloud account usage habits, but still place orders, manage servers, and issue official invoices in the official Alibaba Cloud background. However, at the time of settlement, through the Aliyun recharge method, you can get the discount or cash rebate subsidy provided by the agent, which can often save 15% - 30% of the hardware purchase cost. High bandwidth for Hong Kong nodes requiring long-term renewal

For ECS enterprises, this is a very cost-effective savings.

5. summary and purchase suggestions

After the depth measurement of the Alibaba Cloud Hong Kong, China node, the conclusions are as follows:

Performance and stability: 9.5 / 10 points. BGP line optimization is well done. Even in the most congested period of the evening peak, the packet loss rate can be stably suppressed within 2%, and the delay jitter is extremely small.

SEO friendly degree: high. Low latency and fast TTFB response can greatly improve the crawling efficiency and user experience score of search engine Spider (such as Googlebot and Baiduspider).

Suitable for business: cross-border e-commerce self-built stations (Shopify/WordPress/WooCommerce) are free of filing, and overseas API interface services and data transfer for web site enterprises facing the whole network in mainland China

Procurement proposal

: If it is a personal webmaster or test business, preferred

Hong Kong Lightweight Application Server

; If it is an enterprise-level core business, it is recommended to go directly to

ECS instance + Alibaba Cloud recharge channel

, which preserves high-performance and high-quality BGP lines, while significantly reducing annual IT infrastructure spending.