Blog
Designing High-Throughput Web3 RPC Infrastructure
Web3 applications rely on remote procedure calls (RPCs) to read chain state and submit transactions. Misjudge that layer and users experience “decentralization” as blank balances, timeouts, and stuck swaps.
Melbicom provides the server and network infrastructure for full and archival nodes, RPC gateways, indexers, and dApp backends. This article addresses a practical question: what does reliable, high-throughput, low-latency RPC infrastructure for modern Web3 apps require, and how should you size its dedicated server footprint?
Choose Melbicom— 1100+ ready-to-go servers — 21 Tier III & IV data centers — Free BGP sessions |
The Limits of Public RPC Endpoints
Public RPC endpoints are indispensable for experimentation and small projects. Production workloads can outgrow them through request volume, expensive methods, or requirements for fresher state.
In practice, public endpoints are ideal for:
- Local development and testnets
- Lightweight wallets
- Prototyping and small internal tools
RPC Demand vs. Published Endpoint Limits
Compare peak traffic, daily volume, and method availability separately. These examples use the limits published in each network’s documentation:
| Endpoint | Published Limits | Capacity-Planning Implication |
|---|---|---|
| Conflux free RPC | 50 calls/second; 100,000 calls/day | The daily quota can bind before the overall peak limit; several common methods have a 5 QPS cap |
| Conflux Standard RPC | 100 calls/second; 1,000,000 calls/day | A higher peak limit still leaves a separate daily budget |
| Solana public mainnet | 100 requests/10 seconds/IP; 40/10 seconds/IP for one method | Account for the full rate window and shared source IP; Solana directs production applications to private RPC access |
| NEAR public archival | Severely rate-limited; numerical cap not published | Choose documented capacity and historical-method support for sustained archival workloads |
For example, a Conflux application with 100 concurrent sessions, each making two uncached calls every five seconds, generates 40 RPS when requests are spread evenly. Sustained for 24 hours, that is 3,456,000 calls—above both daily quotas despite remaining below their overall per-second limits.
Dedicated nodes move the bottleneck from someone else’s shared infrastructure to capacity planning you control. Size capacity against the required method mix: a lightweight balance lookup and a historical trace place different demands on a node.
Reliable Web3 RPC Infrastructure
Treat RPC infra as a first‑class, multi‑region API platform backed by dedicated blockchain nodes, rather than a single dApp endpoint.
Sizing Dedicated RPC Nodes for Real Workloads
Start with the chain client’s requirements, then add capacity for the RPC methods, indexes, and concurrent requests the application will serve.
Ethereum full nodes:
- EIP-7870 recommends 4 cores/8 threads, 32 GB RAM, and 4 TB NVMe for combined execution and consensus clients.
- Add CPU, RAM, and disk headroom for production RPC, tracing, and concurrent queries.
- Choose archive retention separately from the resources needed to follow the chain.
Solana RPC nodes:
- Anza’s RPC guidance specifies 16 cores/32 threads or more and at least 256 GB RAM; allow 512 GB or more when enabling all account indexes.
- Use separate NVMe: ≥1 TB each for accounts and ledger; ≥500 GB for snapshots.
- Size ledger retention for the transaction history your API serves.
Indexers and analytics backends:
- Size CPU and memory for ingestion, database working sets, and concurrent queries.
- Use NVMe for active databases and object storage for snapshots and cold datasets.
- Account for traffic between RPC nodes, indexers, and query services.
Designing Geo‑Distributed RPC Clusters

Region placement and routing determine how quickly clients reach RPC nodes and whether the service survives a regional failure.
Best practice looks like this:
- At least two regions per critical chain, with enough surviving capacity to serve traffic during a regional failure.
- Local RPC for latency-sensitive paths—place trading engines, order books, and latency-critical DeFi services near the nodes they query.
- Global routing and failover, using DNS, an L7 load balancer, or BGP anycast, connected to checks for endpoint availability and chain synchronization.
DNS failover depends on caching; anycast needs health-driven route withdrawal. WebSocket clients must reconnect and resubscribe, then recover missed updates.
Melbicom’s data centers include these regional deployment options:
- North American users → RPC nodes in Atlanta and Los Angeles
- European users → RPC nodes in Amsterdam and Frankfurt or Warsaw
- Asian users → RPC nodes in Singapore and Tokyo
Security, API Design, and Observability Around RPC
Security research on Web3 supply chains calls out RPC endpoints explicitly: a compromised gateway can return false balances, transaction histories, or contract state. Applications that trust those responses can display misleading information or make incorrect decisions.
That reality drives a few architectural choices:
- Keep an independent RPC path. Self-operated dedicated nodes let you cross-check external responses at the same block and finality, and provide fallback capacity.
- Instrument RPC like any other core API. Collect p95/p99 latency, error codes, method-level throughput, and node sync lag into a single telemetry stack with host metrics.
- Protect the RPC gateway. Keep node RPC ports private; expose required methods through an authenticated HTTPS gateway with per-client limits and request logging, following Geth’s security guidance.
Comparing Dedicated Servers Suitable for Hosting Web3 Apps and Blockchain Nodes
Compare dedicated servers for RPC infrastructure by the chain client, RPC method mix, historical retention, and peak concurrency. Choose CPU, RAM, and storage that keep nodes synchronized while serving requests, then size network capacity and a second region for the same workload during failover.
Network and Bandwidth: Sizing RPC Ports
Network capacity must cover node synchronization, RPC responses, subscriptions, and traffic to indexers. Ethereum’s full-node baseline is 50 Mbps download and 15 Mbps upload; Anza specifies 1 Gbps symmetric for an unstaked Solana node. Production RPC traffic adds to those requirements.
For example, 1,000 responses per second averaging 100 kB each require 800 Mbps of outbound payload: 1,000 responses/s × 100,000 bytes/response × 8 bits/byte. Protocol overhead, subscriptions, and node traffic leave little headroom on a 1 Gbps port.
Melbicom offers ports up to 200 Gbps per server, with unmetered and guaranteed-throughput options without oversubscription ratios. Choose a larger port when combined traffic or failover demand exceeds the capacity of the smaller link.
Storage and State Growth: NVMe for Active Data
NVMe is the practical choice for active RPC databases. Archive capacity depends on the client and retained history: Geth’s path-based archive requires roughly 2 TB with full flat state history, or 6.5 TB with historical trie data. Its historical state data can reside on slower disks. Erigon recommends a 4 TB drive for an Ethereum archive node. Allocate additional space for growth and the requirements of any colocated services.
A practical pattern:
- Hot state and indexes on NVMe, with capacity for growth and client maintenance
- Cold archives and snapshots in object storage, such as S3-compatible buckets, plus a CDN for distributing snapshots to new nodes and remote teams
- Periodic, client-compatible snapshots with tested restores to accelerate recovery after node or storage failures
Melbicom’s S3-compatible object storage in Amsterdam provides a separate home for chain snapshots and datasets, with no ingress fees.
CPU, RAM, and Virtualization Choices
Choose deployment models according to the client and workload:
- Validators and RPC nodes: Single-tenant dedicated servers provide control over CPU scheduling and local storage. Ethereum also supports virtualized deployments; evaluate sustained CPU, disk latency, and network allocation against the client’s requirements.
- Indexers, ETL, and analytics jobs: These services can be containerized for repeatable deployment and scaled across dedicated hosts.
- Dev and staging environments: These can live on smaller dedicated servers or cloud virtual machines, but should mirror production topology closely enough to catch performance regressions.
Who Offers Reliable Web3 and RPC Servers?
Melbicom offers dedicated servers for Web3 and RPC infrastructure, with NVMe storage, high-memory configurations, and regional deployment choices. Compare servers against the chain client’s requirements, then check sustained CPU and disk performance, bandwidth options, and whether the remaining regions can carry traffic during a failure.

Server Provisioning and BGP Options
Once you have sized the clients and chosen regions, Melbicom offers:
- Ready-to-go dedicated servers typically provisioned in about 2 hours
- Custom high-RAM or storage-heavy servers delivered in 3–5 business days
- Free BGP sessions with BYOIP in every data center, supporting customer-operated anycast and regional failover
You manage the operating system, blockchain clients, monitoring, and RPC failover; we support the server hardware and network.
Owning the RPC Layer as Web3 Scales

Owning your RPC infrastructure on dedicated servers doesn’t mean ignoring managed services or decentralized networks. Use them within an architecture that assumes any one provider, network, or region can fail without taking your application down.
Practical Recommendations for RPC Infrastructure
Turn the architecture into deployment checks:
- Test the full request path. Measure UI → API → indexer → RPC → chain latency under the expected method mix and cache-miss rate; include historical queries and transaction submission.
- Exercise regional failover. Remove one region from service during a load test and check remaining capacity, node freshness, and WebSocket recovery.
- Test snapshot restores. Restore onto a separate host and verify synchronization before sending production traffic to it.
Order a Server
Ready to move beyond public endpoints and shared RPC bottlenecks? Order a dedicated server and explore Web3‑focused configurations for validators, RPC nodes, and indexing workloads—or talk with us about a custom build.