How Alibaba Cloud Elasticsearch Serverless Cuts Log Costs by Over 70%

Overview: With the rapid growth of internet services, log data volume is huge; traditional log processing faces high cost and poor scalability. More enterprises turn to Alibaba Cloud Elasticsearch Serverless.

Feature Comparison

Traditional solution: self‑built Elasticsearch on ECS

Low resource utilization: waste during off‑peak, shortage during peak.

High maintenance cost: hardware plus manpower for monitoring, backup, security, upgrades.

Poor scalability: manual scaling needed for traffic spikes.

Alibaba Cloud Elasticsearch Serverless

Pay‑as‑you‑go: only pay for actual usage.

Automatic scaling: compute and storage adjust to demand.

Zero ops: Alibaba manages underlying infrastructure.

Scenario Simulation

Using real log business data, a one‑day write‑load simulation was performed:

1. Business curve: 1.2k–5.4k QPS, single bulk 3 MB writes, no queries.

Cost of self‑built ECS cluster (24 CPU × 6 nodes, 2048 GiB ESSD PL1): ¥4.68 per hour + ¥4.3008 per disk = ¥1293.24 per day.

2. ECS load monitoring (image).

3. Serverless monitoring (image).

Performance Compared to Self‑Built

Self‑built ES shows response time fluctuations even under low load.

During peak write pressure, self‑built ES RT spikes, while Serverless remains stable.

High load can cause request failures in self‑built cluster due to full thread‑pool queues.

Dynamic Shard Allocation

Serverless adjusts primary shard count based on write water‑level, keeping extra shards during drops and scaling up when needed.

After a drop, high shard count persists briefly to handle possible spikes.

If water‑level exceeds current shards, it raises shard count after a short delay.

When water‑level exceeds controllable range, shard increase accelerates.

Storage Compression

Serverless achieves >80 % storage cost reduction; daily data volume reaches hundreds of billions, with storage footprint about one‑sixth of self‑built ES.

Cost Savings

Serverless charges per CU hour; the test scenario without query load uses a minimum of 5 CU for queries. Overall cost is reduced by more than 70 % compared with self‑built, and up to 88.6 % when accounting for hot/cold storage tiers.

How to Enable Elasticsearch Serverless

Step 1: Activate the Service

Log in to the Elasticsearch Serverless console and click “Activate Now”.

Step 2: Create an Application

Enter the application creation page, configure basic info, select “General Retrieval” type, set network access (e.g., public with IP whitelist), set password for Kibana, and click “Create”.

After creation, the application appears in the management page; once its status becomes “Running”, the service is ready for use.