Alibaba Cloud Singapore And Hong Kong Cn2 Capacity Planning And Emergency Plan During The E-commerce Peak Period

1.

overall goals and assumptions

estimated peak number of concurrent users: 200,000;
target p95 response time < 500ms;
target cdn cache hit rate ≥ 95%;
planned maximum carrying bandwidth of origin: 40 gbps;
ddos protection preparation: peak cleaning capacity ≥ 80 gbps;
the traffic path preferentially uses the cn2 dedicated line to reduce cross-border jitter and packet loss.

2.

regional capacity planning (numeric display)

planning goals are broken down by region and link, as shown in the table below:

area	number of ecs units	single unit specifications	single unit bandwidth	total bandwidth	estimated qps
singapore (cn2)	80	8vcpu/32gb	200mbps	16 gbps	60,000
hong kong (cn2)	100	8vcpu/32gb	240mbps	24 gbps	100,000

note: the bandwidth in the table is a conservative configuration and can be expanded through slb and elastic public ip;
reserve additional disaster recovery bandwidth of 10 gbps for cross-region handover;
database master-slave read and write separation, rds specification example: 4-core 16gb master, latency ≤10ms;
cache redis instance: 3 master and slave nodes, 64gb of memory each, supporting millions of gets per second.

3.

architectural components and technical points

edge: alibaba cloud cdn+iaas edge nodes cover near-origin cache;
network: cn2 direct connection priority, bgp multi-exit redundancy, cross-availability zone anycast dns;
load: slb (layer four + layer seven) performs session maintenance and health checks;
storage: oss sharding + back-to-source compression to reduce origin bandwidth;
protection: anti-ddos pro is linked with cloud shield waf, and the policy is based on ip reputation and behavioral fingerprints.

4.

emergency plans and drill steps

monitoring links: real-time alarms on bandwidth, number of connections, qps, and error rate;
automatic expansion: trigger the ecs elastic scaling group expansion strategy based on cloudmonitor;
cleaning strategy: exceeding the threshold (for example, bandwidth >60 gbps) triggers ddos cleaning and traffic mirroring;
downgrade strategy: static page and read-only mode current limiting to protect the payment link;
switching process: dns weight + slb black and white list to quickly switch to the backup area, rto target ≤ 2 minutes;
drill frequency: 3 full-process desktop drills + 1 full-link stress test before the peak.

5.

real cases and configuration examples

case: an e-commerce company used the cn2 line during the 11.11 period, and the peak bandwidth reached 32 gbps. origin’s error rate increased due to insufficient cdn hit rate;
disposal: the elastic expansion was started on the same day from 40 to 200 ecs units, and slb was expanded horizontally to 8 instances;
results: after loading cache strategy optimization and anti-ddos cleaning, the origin error rate dropped by 90% and the payment success rate returned to 99.8%;
example ecs configuration (hong kong): ecs.g6.8xlarge 8vcpu/32gb, public network bandwidth 240mbps, system disk 100gb ssd;
database example: rds.mysql.s1.large master node 8vcpu/32gb, backup delay < 1s, rpo ≤ 5min;
recommendation: pre-purchase annual bandwidth or reserve dedicated channels to ensure link stability during peak periods.