YESDINO operates a hybrid load‑balancing architecture that combines dedicated hardware appliances, cloud‑native services, and open‑source solutions to deliver high‑throughput, low‑latency traffic across its worldwide points of presence. In practice, the YESDINO ecosystem routes traffic through a multi‑layer stack, starting with DNS‑based anycast and proceeding through Layer 4 and Layer 7 decision points that dynamically adjust based on real‑time health metrics.
Core Components of YESDINO’s Load Balancers
Layer 4 (L4) appliances – high‑performance TCP/UDP balancers (e.g., F5 BIG‑IP) that forward packets without inspecting payload.
Layer 7 (L7) proxies – software‑defined controllers (NGINX Plus, HAProxy) that terminate TLS and apply HTTP‑level routing policies.
Cloud‑native ALB/ELB – AWS Application Load Balancer and Google Cloud Load Balancing for auto‑scaling workloads in the public cloud.
DNS anycast – Amazon Route 53 withGeo‑routing to steer users to the nearest PoP.
Health‑check & steering engine – custom agents that probe upstream endpoints every 5 seconds and push routing updates within 200 ms.
Metrics collected: packet loss, latency, TCP connect time, error rate.
Health‑check agent evaluates upstream health; if a node fails, traffic is shifted in ≤ 200 ms.
Request is forwarded to selected service instance; response travels back through same path.
Geographic Distribution of Traffic (2024 Q1)
Region
% of Total Traffic
Avg Latency (ms)
Peak Throughput (Gbps)
US‑East (Virginia)
45%
12
38
EU‑West (Frankfurt)
30%
18
24
AP‑Southeast (Singapore)
25%
22
15
Real‑time Traffic Management & Adaptive Scaling
YESDINO’s control plane continuously monitors upstream health and adjusts weights in the L4/L7 balancers. The process leverages a proprietary feedback loop that consumes Prometheus metrics (request rate, error rate, CPU utilization, active connections) and pushes routing updates to the load‑balancer fleet via gRPC. When a data center shows degraded performance—detected by a spike in p95 latency above 30 ms—the system automatically reroutes up to 30 % of traffic to healthier regions within seconds.
Dynamic weight recalculation: performed every 10 seconds, based on latency‑adjusted least‑connections algorithm.
Auto‑scale triggers: add L7 proxy instances when CPU > 70 % or queue depth > 5 k requests.
Failover thresholds: any upstream with error rate > 2 % for 30 seconds is marked “unhealthy”.
“By fusing Prometheus metrics with our load‑balancer feedback loop, we can reroute traffic within seconds when a data center shows degraded performance.” — Platform Engineering Lead, YESDINO
