Databend vs. Snowflake: Data Ingestion Benchmark

Overview

We conducted four specific benchmarks to evaluate Databend Cloud versus Snowflake:

1. TPC-H SF100 Dataset Loading: Focuses on loading performance and cost for a large-scale dataset (100GB, ~600 million rows).
2. ClickBench Hits Dataset Loading: Tests efficiency in loading a wide-table dataset (76GB, ~100 million rows, 105 columns), emphasizing challenges associated with high column counts.
3. 1-Second Freshness: Measures the platforms' ability to ingest data within a strict 1-second freshness requirement.
4. 5-Second Freshness: Compares the platforms' data ingestion capabilities under a 5-second freshness constraint.

Platforms

• Snowflake: A well-known cloud data platform emphasizing scalable compute, data sharing.
• Databend Cloud: A cloud-native data warehouse built on the open-source Databend project, focusing on scalability and cost-efficiency.

Benchmark Conditions

Conducted on a Small-Size warehouse (AWS us-east-2) using data from the same S3 bucket.

Performance and Cost Comparison

Performance and Cost

• TPC-H SF100 Data: Databend Cloud offers a 67% cost reduction over Snowflake.
• ClickBench Hits Data: Databend Cloud achieves a 91% cost reduction.
• 1-Second Freshness: Databend loads 400 times more data than Snowflake.
• 5-Second Freshness: Databend loads over 27 times more data.

Data Ingestion Benchmarks

TPC-H SF100 Dataset

Metric	Snowflake	Databend Cloud	Description
Total Time	695s	446s	Time to load the dataset.
Total Cost	$0.77	$0.25	Cost of data loading.

• Data Volume: 100GB
• Rows: Approx. 600 million

ClickBench Hits Dataset

Metric	Snowflake	Databend Cloud	Description
Total Time	51m 17s	9m 58s	Time to load the dataset.
Total Cost	$3.42	$0.30	Cost of data loading.

• Data Volume: 76GB
• Rows: Approx. 100 million
• Table Width: 105 columns

Freshness Benchmarks

1-Second Freshness Benchmark

Evaluates the volume of data ingested within a 1-second freshness requirement.

Metric	Snowflake	Databend Cloud	Description
Total Time	1s	1s	Loading time frame.
Total Rows	100 Rows	40,000 Rows	Volume of data successfully ingested within 1s.

5-Second Freshness Benchmark

Assesses the volume of data that can be ingested within a 5-second freshness requirement.

Metric	Snowflake	Databend Cloud	Description
Total Time	5s	5s	Loading time frame.
Total Rows	90,000 Rows	2,500,000 Rows	Volume of data successfully ingested within 5s.

Reproduce the Benchmark

You can reproduce the benchmark by following the steps below.

Benchmark Environment

Both Snowflake and Databend Cloud was tested under similar conditions:

Parameter	Snowflake	Databend Cloud
Warehouse Size	Small	Small
vCPU	16	16
Price	$4/hour	$2/hour
AWS Region	us-east-2	us-east-2
Storage	AWS S3	AWS S3

• The TPC-H SF100 dataset, sourced from Amazon Redshift.
• The ClickBench dataset, sourced from ClickBench.

Prerequisites

• Have a Snowflake account
• Create a Databend Cloud account

Data Ingestion Benchmark

The data ingestion benchmark can be reproduced using the following steps:

TPC-H Data Loading

1. Snowflake Data Load:

   • Log into your Snowflake account.
   • Create tables corresponding to the TPC-H schema. SQL Script.
   • Use the COPY INTO command to load the data from AWS S3. SQL Script.

2. Databend Cloud Data Load:

   • Sign in to your Databend Cloud account.
   • Create the necessary tables as per the TPC-H schema. SQL Script.
   • Utilize a similar method to Snowflake for loading data from AWS S3. SQL Script.

ClickBench Hits Data Loading

1. Snowflake Data Load:

   • Log into your Snowflake account.
   • Create tables corresponding to the hits schema. SQL Script.
   • Use the COPY INTO command to load the data from AWS S3. SQL Script.

2. Databend Cloud Data Load:

   • Sign in to your Databend Cloud account.
   • Create the necessary tables as per the hits schema. SQL Script.
   • Utilize a similar method to Snowflake for loading data from AWS S3. SQL Script.

Freshness Benchmark

Data generation and ingestion for the freshness benchmark can be reproduced using the following steps:

1. Create an external stage in Databend Cloud

CREATE STAGE hits_unload_stage
URL = 's3://unload/files/'
CONNECTION = (
    ACCESS_KEY_ID = '<your-access-key-id>',
    SECRET_ACCESS_KEY = '<your-secret-access-key>'
);

2. Unload data to the external stage.

CREATE or REPLACE FILE FORMAT tsv_unload_format_gzip
    TYPE = TSV,
    COMPRESSION = gzip;

 COPY INTO @hits_unload_stage
FROM (
    SELECT *
    FROM hits limit <the-rows-you-want>
)
FILE_FORMAT = (FORMAT_NAME = 'tsv_unload_format_gzip')
DETAILED_OUTPUT = true;

3. Load data from the external stage to the hits table.

COPY INTO hits
    FROM @hits_unload_stage
    PATTERN = '.*[.]tsv.gz'
    FILE_FORMAT = (TYPE = TSV,  COMPRESSION=auto);

4. Measure results from the dashboard.