Solution architectures

Two VPS in different Joy regions (for example Mumbai and Noida) sit behind the CDN with origin shield enabled. A managed MySQL primary lives in Mumbai with a read replica in Noida; uploads go to object storage so either web node can serve them. DNS uses a 30-second TTL and health-checked records, so failover is a DNS flip that completes in under a minute. Sessions are stored in Redis on the primary region with a replica alongside the database.

Why it works: the CDN absorbs most reads, the shield protects the small origins, and a regional failure loses at most 30 seconds of new writes. Scale-up path: add web nodes hourly through the API during campaigns; destroy them after.

Bare-metal game nodes sit on a private VLAN with no public addresses. Player traffic enters through Joy scrubbing and is delivered over Virtual Connect (GRE) to a pair of edge VPS running your own load balancer or the game's proxy layer; management traffic uses a separate /29 reachable only from your office IPs. NeuroMesh NetFlow scores flows in real time, so UDP floods are dropped before they reach the game ports.

Why it works: the game servers are unreachable except through the scrubbed path, and the edge VPS can be replaced hourly if an attack pattern changes. Scale-up path: add bare-metal nodes per region; hourly VPS for lobbies at peak.

Baseline capacity runs on monthly VPS in two regions behind the CDN (used here as a global anycast front door with caching disabled for dynamic paths). A queue drives background work; when queue depth crosses a threshold, a small controller calls POST /api/v1/instances with cycle: hourly to add workers, and DELETE them when the queue drains. Workers bootstrap from a snapshot-derived image and register themselves on boot.

Why it works: hourly billing means burst capacity costs cents, and the API's job model lets the controller wait for ready before routing work. Guard-rails: the API allows 10 deployments per hour per account — plan burst sizes accordingly or ask for a higher limit.

A regional ISP takes 1–10 G of IP transit at the Mumbai IX with Joy DDoS in the path, announces its prefixes with RPKI ROAs, and uses BGP communities for traffic engineering (prepend, no-export to specific upstreams, blackhole for attacked /32s). rDNS is delegated to the ISP's own nameservers. Optionally a second session in Noida provides geographic redundancy over dark fibre or an L2 circuit.

Why it works: deterministic paths through Tier-1 upstreams and 394 peers cut hops; scrubbing is upstream of the ISP's own edge so attacks never fill the transit port. Pricing: transit $500 / Gbps, DDoS-protected transit from $700 / Gbps, IX port from $150 / Gbps, cross-connect $100.

A Windows Server 2022 application server and a SQL Server instance on Large VPS, joined to a small Active Directory domain controller on a Micro VPS, all on a private VLAN with only the RDP gateway exposed. Users connect through the gateway with NLA and MFA; nightly snapshots before patch Tuesday, and off-node backups to Joy backup space.

Why it works: Windows licences are per-VM add-ons, private VLANs are free within a region, and snapshots make patching reversible.