Saltar al contenido principal

Service Coverage Analysis: Minimum AWS Deployment

This document analyzes the service coverage of the Minimum AWS Deployment (aws-minimum) for the AlphaSwarm hosted platform. This deployment is optimized for a single-account, lowest-cost footprint (~$140/mo fixed) and serves as the baseline for individual operators and early-stage deployments.

1. Executive Summary

The Minimum AWS Deployment provides the essential compute, data, and identity services required to run the AlphaSwarm Admin dashboard, Core API, and Platform UI. It trades off high-availability, advanced streaming, and managed Kubernetes for a lean, ECS Fargate-based architecture using Amazon Bedrock as the primary LLM provider.

2. Service Coverage Matrix

CategoryServiceMinimum AWSFull AWS (Live)Role in Platform
ComputeECS FargateRuns Admin BFF, Frontends, and Core API.
EKS (Kubernetes)High-performance quant runtime and Celery workers.
IngressALBPublic HTTPS entry point and Cognito-gated routing.
CloudFrontGlobal edge delivery and Admin/AgentCore fronting.
IdentityAWS CognitoEnd-user authentication and identity brokerage.
DatabaseRDS PostgreSQLCanonical ledger, state, and vector storage.
CacheElastiCache RedisTask brokerage (Celery) and WebSocket pub/sub.
StreamingMSK KafkaHigh-throughput real-time market data streaming.
AI/LLMBedrock (Direct)LLM provider (Claude Haiku 4.5 baseline).
Bedrock AgentCoreManaged 8-hour sessions and agentic memory.
Bedrock KBManaged RAG and knowledge base indexing.
VectorOpenSearch SlsProduction-grade managed vector database.
StorageECRContainer image registry.
S3Terraform state, artifacts, and lakehouse storage.
ObsCloudWatchLogging, metrics, and basic infrastructure alarms.
Prom/GrafanaAdvanced metrics visualization and tracing.

3. Detailed Component Analysis

3.1 Compute & Orchestration

  • Minimum Architecture: Relies exclusively on ECS Fargate. All core services (Admin, Core API, UI, IDE) run as Fargate tasks.
  • Coverage Gap: Skips EKS and Karpenter. This means the environment lacks the auto-scaling worker tier used for heavy backtests, RL training, or high-frequency strategy execution.
  • Rationale: Simplifies the stack and eliminates the ~$72/mo base cost of an EKS cluster.

3.2 Data & Streaming

  • Minimum Architecture: Uses a single-AZ RDS PostgreSQL (db.t4g.medium) and a single-node ElastiCache Redis (cache.t4g.small).
  • Coverage Gap: Skips MSK Kafka. Real-time streaming is handled via Redis pub/sub or ignored for the minimum tier.
  • Rationale: MSK Kafka is a high-cost component (starts at ~$150-200/mo). Redis provides sufficient performance for single-operator workloads.

3.3 AI & Knowledge Management

  • Minimum Architecture: Direct integration with Amazon Bedrock (specifically Anthropic Claude Haiku 4.5).
  • Coverage Gap: Deliberately excludes Bedrock AgentCore, Knowledge Bases (RAG), and OpenSearch Serverless.
  • Rationale: These services add significant monthly overhead. The minimum tier assumes agents operate with short-term context or local vector storage (pgvector).

3.4 Networking & Security

  • Minimum Architecture: Standard VPC with interface endpoints for AWS services to reduce NAT Gateway bandwidth costs. Public entry via ALB with Cognito authentication.
  • Coverage Gap: No CloudFront distribution. The ALB DNS is used directly.
  • Rationale: Reduces DNS/CDN management overhead and eliminates CloudFront's base costs and WAF requirements for the initial launch.

4. Operational Readiness

The Minimum AWS Deployment is considered Ready for Deployment as of June 2026. It includes the necessary terraform modules in alphaswarm_platform and is recognized as a valid aws-minimum target in the platform topology.yaml.

5. Promotion Path

The architecture is designed to be additive. An operator can promote from the Minimum to the Full deployment by:

  1. Enabling CloudFront for the Admin and AgentCore edge.
  2. Adding the Bedrock Knowledge Base and OpenSearch Serverless modules.
  3. Provisioning the EKS cluster for the quant runtime tier.
  4. Retargeting the environment in topology.yaml from aws-minimum to aws.