How We Cut Search Latency by 80% with an AWS Cloud‑Native Architecture

Infrastructure Layer Optimization

1.1 AWS Placement Group

Network topology : Use a Cluster Placement Group to provide a dedicated 10 Gbps backbone between EC2 instances.

Configuration code :

aws ec2 create-placement-group \
  --group-name easybom-net-group \
  --strategy cluster \
  --region us-west-1

Performance verification : iperf3 test shows a transfer of 11.2 GB over 10 seconds, achieving 9.62 Gbits/sec.

1.2 OS‑Level Tuning (Windows Server)

Key parameters were adjusted to improve TCP handling, memory locking, and power settings.

TCP connections : Default 1000 → Optimized 10000 (Set‑NetTcpSetting -SettingName InternetCustom -MaxSynRetransmissions 10)

Memory lock pages : Disabled → Enabled (LockPagesInMemory=1 for SQL Server)

Power plan : Balanced → High performance (powercfg /setactive 8c5e7fda-e8bf-4a96-9a85-a6e23a8c635c)

Elasticsearch Cluster Construction Guide

2.1 Node Expansion Steps

Freeze existing cluster to prevent shard migration:

{
  "persistent": {
    "cluster.routing.rebalance.enable": "none"
  }
}

Securely add new nodes with core elasticsearch.yml settings:

discovery.seed_hosts: ["172.31.88.146:9300", "172.31.88.61:9300"]
cluster.initial_master_nodes: ["node-1", "node-2"]
xpack.security.transport.ssl.keystore.path: certs/elastic-certificates.p12

Certificate issuance (Let’s Encrypt example shown later).

2.2 Performance Monitoring Metrics

Key visual dashboards (shown in the original images) illustrate cluster health and query latency.

indices.breaker.total.limit: 70%

thread_pool.search.size: 2 * cores + 1

SQL Server Read/Write Splitting

3.1 Publication/Subscription Matrix

Sync method : Transactional replication (primary DB priority, read‑only DB immediate update).

Conflict handling : Primary‑first, automatic rollback on the read‑only side.

Bandwidth usage : Compressed SMB 3.1.1 for primary, decompressed on replica.

3.2 Migration Checklist

Pre‑condition verification (check tables without primary keys):

SELECT name FROM sys.tables WHERE object_id NOT IN (
  SELECT parent_obj FROM sys.key_constraints
);

Publication configuration script :

EXEC sp_addpublication @publication='EasyBom_Repl',
  @sync_method='database snapshot',
  @repl_freq='continuous';

Fail‑over test (simulate primary outage):

# PowerShell
Stop-Service MSSQLSERVER -Force
# Verify read‑only routing (max 15 s latency)

Intelligent Load Balancing

4.1 Cloudflare Rule Configuration

{
  "rules": [
    {
      "name": "Asia_Traffic",
      "condition": "ip.src.region eq 'APAC'",
      "action": "route_to_origin(origin='tokyo-edge')"
    },
    {
      "fallback": "global_health_check",
      "monitor": "http://172.31.88.211:9200/_cluster/health"
    }
  ]
}

4.2 Certificate Management Automation

# Renew Let’s Encrypt wildcard certificate
certbot certonly \
  --dns-cloudflare \
  --dns-cloudflare-credentials ~/cloudflare.ini \
  -d '*.easybom.com'

Deep Monitoring System

5.1 Alert Thresholds

ES JVM heap – warning 75 %, critical 85 % (check every 30 s)

SQL lock wait – warning 5 %, critical 15 % (check every 1 min)

Internal network latency – warning 2 ms, critical 5 ms (check every 10 s)

5.2 Log Analysis Command

# High‑frequency query analysis in Elasticsearch
GET _search/template
{
  "id": "slow_query_analysis",
  "params": {"threshold": "100ms"}
}