# near.org RPC service by FASTNEAR\\n\\n## Organization\\n\\nFASTNEAR\\n\\n### Primary person\\\'s contact information\\nEvgeny Kuzyakov (ek@fastnear.com)\\n\\n### About\\n\\n[FASTNEAR](https://fastnear.com) is NEAR Protocol Infrastructure Provider.\\n\\n#### [Services](https://fastnear.com/services)\\n- RPC endpoints\\n- API for wallets\\n- Snapshot data\\n- Indexing and Lake solutions\\n- Validator services\\n\\n#### Key Features\\n1. **Dedicated to NEAR Protocol:** We exclusively serve the NEAR ecosystem.\\n2. **High-Performance Infrastructure:** \\n   - The majority of nodes and servers hosted on bare-metal.\\n   - Currently ~4,000 RPC requests per second. Resulting in 320M calls per day.\\n\\n#### Reliability and Security\\n- Experienced in handling:\\n  - Traffic surges of 30,000 RPS\\n  - DDoS attacks (recently)\\n  - Provider outages\\n- ~100% RPCs uptime for the last 3 months\\n- 250ms p95 over 60min view call latency\\n\\n#### Monitoring and Performance\\n- Internal monitoring and alert systems\\n- Global RPC latency monitoring dashboards\\n\\n## Body\\n\\n### Hardware\\n\\nBased on the current usage of the following:\\n- RPC Mainnet - 10k requests/s (while currently 6k RPS)\\n- RPC-Archival Mainnet - 2k requests/s\\n- RPC Testnet - 400 requests/s\\n- RPC-Archival Testnet - 150 requests/s\\n\\nWe recommend the following setup:\\n- RPC Mainnet: 20 nodes (estimating 500 RPS each)\\n- RPC-Archival Mainnet: 4 full NVME nodes, with 10Gbps (at 500 RPS)\\n- RPC Testnet - 4 nodes (for 2X redundancy)\\n- RPC-Archival Testnet - 2 full NVME nodes (for 2X redundancy)\\n\\nNote, that we don\\\'t rely on the read-rpc for this setup, due to the current setup complexity. We may consider switching to read-rpc over time.\\n\\n**Pricing**\\n\\nThree types of servers:\\n- RPC (mainnet/testnet): $250 with 2Tb NVME + 1Gbps\\n- RPC-Archive Mainnet: $2500 with 90Tb NVME + ~250TB of 10Gpbs traffic\\n- RPC-Archive Testnet: $1000 with ~20Tb NVME + 1Gbps\\n\\nTotal:\\n- 24 RPC nodes: $6000\\n- 4 Mainnet Archival nodes: $10000\\n- 2 Testnet Archival nodes: $2000\\n\\n**Total: $18000 per month**\\n\\n### Load-balancing\\n\\nOur current RPC setup is using a simple DNS-based load balancer. This works well for front-end clients which come from the different IPs. This setup has quite low cost overhead, but it can\\\'t be used to minimize GEO-spatial latency. There is a latency overhead, because each request might be routed not to the closest server, but instead to a server on a different continent. While it does increase latency by up to ~140ms, we don\\\'t consider it significant enough to justify costs associated with geo-spatial load-balancing.\\n\\n#### Expected load\\n\\n- ~12500 RPS\\n- 32.4B requests per month\\n- 93TB per month\\n\\nMost load balancer providers have costs associated with the number of requests. While GCP doesn\\\'t charge for the number of requests, it has limitations of load balancing across external endpoints and does charge for the traffic.\\n\\n*Note, we don\\\'t have enough experience with GCP load-balancing, so some assumptions might be wrong.*\\n\\nWe have considered the following options for this proposal:\\n1. Simple DNS-based load-balancing\\n3. **PREFERRED** Cloudflare Enterprise geo-spatial load-balancing.\\n4. Google Cloud Provider Load Balancing with proxy instances or GCP LB with simple traffic splitting.\\n\\n#### Option #1: Simple DNS-based load-balancing\\n\\nWe can use a CNAME record for each of the near.org endpoints to point to our endpoints(s).\\n\\nThis option might be fairly simple to setup at **no extra cost**.\\n\\nNote, Strict rate-limiting on HTTP referer will require extra work on nodes or an enterprise plan.\\n\\n#### Option #2: Cloudflare Enterprise Load-balancing (CF LB)\\n\\nSimple DNS-based load-balancing can be upgraded to geo-spatial load-balancing at our end.\\n\\nDuring the communication with Cloudflare sales, we\\\'ve received an estimate for 15B requests per month. Scaling it up to 32.4B requests, the price might be around **~$6000 per month**.\\n\\nCloudflare enterprise offers rate limiting based on HTTP headers, including HTTP Referer or HTTP Origin.\\n\\n#### Option #3: Google Cloud Provider Load Balancing (GCP LB)\\n\\nGCP external load balancer has bandwidth associated with it. \\nBased on our estimates, the LB traffic is around $10 per TB for ingress, and $10 per TB for egress.\\n\\nGCP LB can\\\'t be used for geo-spatial balancing without additional nodes on GCP or other cloud run/functions.\\n\\nSo there are 2 options for GCP:\\n- Full load-balancing. Requires RPC egress traffic with proxy nodes. Costs *estimated* **~$16000 per month**\\n- Splitting traffic (no geo-spatial routing). Costs around **~$2000 month**. Need to confirm feasibility.\\n\\n\\n## Goals / Milestones\\n\\n- Setup (2 weeks)\\n    - Setup RPC nodes and Archival RPC nodes\\n    - Setup 4 endpoints to point traffic\\n    - Rate-limiters based on IPs\\n- Status page for new endpoints (2 weeks)\\n- Setup external monitoring dashboard (4 weeks)\\n- Setup CF LB (4-8 weeks)\\n    - Negotiate deal with CF for LB\\n    - Setup rate limiter based on the HTTP referer.\\n- Full endpoints migration if necessary (3-6 months)\\n\\n## Metrics\\n\\n- Requests per second\\n    - All request types (expected to handle steady traffic of 12,500 RPS)\\n    - Function calls (majority of requests)\\n    - TX status\\n    - Broadcast TX\\n- Latency\\n    - Mean latency from 3 geo endpoints (~170ms for view calls)\\n    - P95 latency (~250ms for calls)\\n    - Lower mean/p95 latency after CF LB\\n- Rate limiting (number of throttled calls)\\n- Traffic (expected up to 3Tb per day)\\n- Uptime (expected 99.99%)\\n\\n### Compensation\\n\\nFor maintenance and development we ask for $12,000 a month.\\n\\n## Budget\\n\\n$36,000 per month (paid monthly) for the duration of 6 months.\\n- $18,000 nodes\\n- $6,000 CF LB\\n- $12,000 maintenance \\n\\nAfter 6 months, we expect a significant drop in the usage, so we should re-negotiate a deal based on the new requirements.\\n\\n

	Receipt	Action	Method	From		To	Value	Gas Limit

Detalye ng Transaksyon

Overview

Execution Plan

Enhanced Plan

Tree Plan

Receipt Summary

Comments