AWS Hong Kong Region Evening Peak Network Test: Will Triple Network Return Congestion?
In the current corporate and multinational business architecture,
AWS Hong Kong Region(ap-east-1)
It has always been the preferred node for many domestic teams. The geographical location is very close to South China, the physical delay is as low as 10~20ms, and there is no need for filing, which is extremely convenient to deploy.
However, all network engineers and SEO operation and maintenance personnel who have done mainland loop optimization have a common question in their hearts:
Is the return journey of the three networks (telecom, Unicom and mobile) stable during the "evening peak" period of the AWS Hong Kong node from 8 pm to 11 pm? Will there be large-scale congestion and packet loss?
This article will take you through the real measured data, route tracking disassembly and network architecture principles, and provide you with a set of practical optimization and cost control scheme.
Why is evening peak network testing so important?
When many teams deploy services during the day, the Ping delay is only 15ms and the page is open in seconds, thinking everything is fine. But once it's arrived
20:00 - 23:00 p.m.
User complaints follow:
Video loading is slow, API interface times out, game drops, and web pages cannot be opened directly.
The core reason for this phenomenon is
Late peak congestion in the international export backbone
.
The default public network line of AWS Hong Kong is
Standard public network BGP(DirectConnect / Internet Routing)
, and did not specifically for the three major domestic operators to buy expensive optimization lines (such as telecom CN2 GIA, Unicom 9929, etc.). Therefore, when backhaul data enters the domestic backbone network across submarine cables, it is bound to grab bandwidth with the personal traffic of the whole network.
AWS Hong Kong Region Late Peak Triple Network Return Measurement Analysis
In order to restore the most realistic usage scenario, we tested Ping delay, MTR route trace and packet loss rate for major provinces in mainland China at 21:30 in the evening peak.
Test Node: AWS Hong Kong (ap-east-1) EC2 Instance
Test time: working day evening peak 21:30 - 22:30
Test tools: MTR, PingPerf, 17CE multi-point dial test
1. China Telecom
Daytime performance: delay is about 20 - 45 ms, packet loss rate is almost 0%.
Late peak performance: delay soars to 120 - 220 ms, and packet loss rate in local areas can reach 10%-25%.
Routing analysis: Telecom backhaul defaults to the 163 backbone network (AS4134). During the evening rush hour, international exit routers located in Guangzhou, Shanghai, etc. will have a serious queue backlog. A large number of data packets are discarded when they cross the submarine cable into domestic 163 nodes, resulting in a sharp increase in TCP retransmission rate.
2. China Unicom (China
Unicom)
Daytime performance: delay about 30 - 50 ms, good experience.
Late peak performance: delay rises to 60 - 110 ms, packet loss rate is about 3%-8%.
Routing analysis: Unicom backhaul usually goes AS4837 (ordinary Unicom backbone network). Compared with telecom 163, Unicom's international export total bandwidth pressure is slightly smaller, so the performance is slightly better than telecom, but in the evening peak can still clearly feel the network "jitter" (Jitter).
3. China Mobile
Daytime performance: as low as 10 - 25 ms in southern China and 40 - 60 ms in northern China.
Late peak performance: South China maintained at 30 - 60 ms (packet loss rate <3%), while the northern region delayed to 90 - 140 ms (packet loss rate 5%-10%).
Routing analysis: Mobile backhaul mainly relies on CMI(AS9808) lines. Mobile has abundant direct connection bandwidth in Hong Kong, especially for mobile users in Guangdong and South China, and the experience of accessing AWS Hong Kong is relatively the best among the three networks. However, after the northern provinces cross-provincial transmission, limited by the domestic backbone network routing, there will still be a certain delay rise.
Summary of Evening Peak Performance of Three Networks
Operator
Average Daytime Delay
Late Peak Average Delay
Average packet loss rate in late peak
Backhaul Primary Network
Overall experience rating
China Telecom
25 - 40 ms
130 - 220 ms
10% - 25%
AS4134 (163)
⚠The congestion is obvious and needs to be optimized.
China Unicom
30 - 50 ms
70 - 120 ms
3% - 8%
AS4837
🟡Slight jitter, acceptable
China Mobile
15 - 35 ms
30 - 90 ms
1% - 5%
AS9808 (CMI)
🟢Relatively stable (especially in South China)
Conclusion: Will AWS Hong Kong's evening peak return journey be congested?
The answer is: yes, and mainly focused on telecom users.
If you directly expose the public IP of AWS Hong Kong to domestic users:
Mobile users have minimal impact and perception is not obvious;
Unicom users have a slight increase in latency, which does not affect basic web browsing, but services with high real-time requirements will be affected;
Telecom users have a poor experience during the evening peak, with very obvious stuck, disconnection and slow loading.
4 Solutions to AWS Hong Kong Late Peak Congestion
Since the original public network line has a bottleneck in the evening peak, how can enterprises improve the access experience of domestic users?
1. Turn on AWS Global Accelerator(GA Global Acceleration)
AWS GA passes traffic through AWS's own global private backbone. Domestic traffic enters AWS nearby
After the edge node, take the AWS internal optimization route back to Hong Kong. This can effectively avoid the random congestion of the public network submarine cable and significantly reduce the packet loss rate.
2. Deploy a CloudFront CDN
Cache static resources (images, CSS, JS, and short videos) to the CloudFront node. Although the source station of AWS Hong Kong is affected by the evening peak, as long as the static resource hits the edge cache, the loading speed of domestic users is still in milliseconds.
3. Use transit node/SD-WAN line
For extremely sensitive financial, gaming, or real-time audio and video services, traffic can be transferred through a third-party high-quality computer room node with CN2 GIA / 9929 lines, or a direct Connect (direct line) can be established.
4. Optimize business architecture and resource deployment
Asynchronous tasks and background processing that are not sensitive to delay will remain in AWS Hong Kong, while the front-end access layer for high-frequency interaction will be deployed on domestic or dedicated nodes close to users.
Business out to sea and multi-cloud deployment: how to reduce AWS operating costs?
While addressing network performance issues, many enterprises (especially startups and small and medium-sized cross-border teams) tend to face another headache--
AWS official withholding and settlement costs are too high
.
AWS uses foreign currency credit card binding fees by default. For domestic enterprises, they often encounter the following pain points:
Payment restrictions: lack of large foreign currency credit cards, limited limits or easy to trigger bank wind control interception;
Financial reimbursement: AWS officially issues foreign Invoice, domestic companies in the financial accounting, withholding income tax and deduction of value-added tax is extremely cumbersome;
Expensive billing: AWS's traffic fee (Data Transfer Out) and GA acceleration fee are relatively high, and direct payment at the original price will lead to high operating costs.
Why are more and more enterprises choosing "Amazon Cloud Recharge"?
In order to solve the above payment and cost problems, the mature solution in the industry is to seek
AWS Official Senior Partner (AWS Partner Network)
Provided
Amazon Cloud Recharge
Service.
through the compliant
Amazon Cloud Recharge
channels, enterprises can obtain the following core advantages:
Flexible payment methods and domestic invoices: support the use of RMB to transfer, WeChat or Alipay direct settlement, and partners can issue formal domestic VAT invoices (including 6% service items), greatly simplifying the enterprise financial reimbursement and compliance process.
Get an extra discount on your bill: Official AWS partners have a larger discount quota. After an enterprise recharges through Amazon Cloud, the bound AWS account can usually enjoy a bill discount ranging from 10% to 20% (depending on the monthly consumption scale), which directly helps the enterprise save a lot of server and traffic expenses.
Specialized technical support and architecture diagnosis: a reliable replacement agent
It not only provides fund recharge, but also provides free network architecture tuning guidance (such as helping to enable free deduction coupons, optimizing CloudFront configuration, and even diagnosing network link failures at night peak).
Avoid credit card wind control: direct binding of personal credit cards or frequent replacement of cards can easily lead to AWS security wind control (Abuse / Suspicious Activity). Through partner account association or Amazon cloud recharge, the account security is higher, and the business is prevented from being suddenly blocked due to the failure of deduction.
Summary
With its superior geographical location and strong cloud ecology, AWS Hong Kong Region is still one of the best infrastructure for enterprises to go out to sea and cross-border business.
Network level: During the evening peak period, telecom backhaul congestion is relatively obvious. It is recommended to cooperate with AWS Global Accelerator or CloudFront for link acceleration;
Cost and financial level: It is recommended to combine Amazon cloud top-up service, which can not only solve the problems of foreign currency payment and domestic invoice issuance, but also enjoy the bill discount provided by official partners, so as to realize the double optimization of "network experience" and "operating cost.
