Measuring cpimport Performance Baseline with ioping

The purpose of this guide is to detail the process for establishing a performance baseline for the cpimport bulk data loading utility and diagnosing storage subsystem bottlenecks using ioping. Storage design choices—such as local disk deployment versus High Availability Network File Systems (NFS HA)—profoundly impact chunk ingestion execution times.

The reference benchmarks documented here were established on a system containing AWS 4vCPUs and 16GB of RAM.

Baseline Performance Reference Metrics

The following matrix reflects baseline loading rates recorded for a standard testing profile consisting of 10,000 records with 250 columns across varying disk architectures:

Infrastructure Layout	Ingestion Runtime	Observed Storage Latency (ioping)
No NFS (Local Storage)	5 to 6 seconds	data1: 450 µs | data2: 475 µs | data3: 430 µs
With NFS HA	29 to 32 seconds	data1: 3.7 ms | data2: 3.1 ms | data3: 3.1 ms
With NFS HA & Low Latency Storage at bulkrollback	14 to 15 seconds	data1: 3.3 ms (bulkRollback: 590 µs) data2: 3.2 ms (bulkRollback: 560 µs) data3: 3.8 ms (bulkRollback: 490 µs)

Manual Benchmarking Procedure

To run a manual performance check and track read latencies across your column store directories, execute the following steps in sequence:

1. Create the ColumnStore testing table: Initialize a test table (cptest) matching the 250-column layout using the COLUMNSTORE storage engine:

   CREATE TABLE cptest (
     group_name CHAR(2),
     number1 DECIMAL(12, 2),
     -- [Truncated for brevity; insert columns up to col250]
     col250 varchar(250)
   ) ENGINE=COLUMNSTORE;

2. Generate data and populate the table: Seed the table with 10,000 randomized evaluation rows using the system sequence loop engine:

   INSERT INTO cptest
   SELECT 
     CAST(ROUND(RAND() * 10, 0) AS CHAR),
     ROUND(RAND() * 10000, 2),
     -- [Truncated for brevity; complete selections to match your column layout]
     substring(MD5(RAND()),1,2)
   FROM seq_1_to_10000;

3. Export data to a local CSV file: Dump the resulting ColumnStore table fields out to a temporary flat comma-separated file:

   SELECT * FROM cptest INTO OUTFILE '/tmp/cptest.csv' FIELDS TERMINATED BY ',' OPTIONALLY ENCLOSED BY '"';

4. Benchmark the cpimport execution rate: Execute the ingestion tool directly to capture baseline loading speeds:

   cpimport -E '"' -s "," test cptest /tmp/cptest.csv

5. Measure disk sub-system latency using ioping: Install the latency verification tool and run a sequence of 30 tests directly on your active database storage directories to analyze performance consistency:

   sudo yum install ioping
   cd /var/lib/columnstore/dataX/
   ioping -c 30 .

Evaluating ioping Output Summary

When the test completes, evaluate the statistical summary metrics:

--- . (xfs /dev/xvda1) ioping statistics ---
29 requests completed in 15.1 ms, 116 KiB read, 1.92 k iops, 7.51 MiB/s
generated 30 requests in 29.0 s, 120 KiB, 1 iops, 4.14 KiB/s
min/avg/max/mdev = 407.4 us / 520.4 us / 694.0 us / 64.8 us

• Target Guideline: For consistent and dependable storage throughput, target a low AVG execution result under 700 µs, combined with a low standard deviation (mdev).

Automated Scripted Performance Testing

You can automate this benchmarking process by saving the comprehensive verification logic into an administrative script named cpimportBench.bash.

Script Implementation

#!/bin/bash
##################
# Purpose: Test cpimport using a 250 column test & ioping test
# Example: bash cpimportBench.bash root test
##################

USER=root
DATABASE=test
RECORDS=10000
TABLE=cptest
CSVPATH="/tmp/$TABLE.csv"
ITERATIONS=20

if [ ! -z "$1" ]; then USER=$1; fi
if [ ! -z "$2" ]; then DATABASE=$2; fi
if [ ! -z "$3" ]; then ITERATIONS=$3; fi

echo "Variable Summary"
echo "User: $USER"
echo "Database: $DATABASE"
echo "Table: $TABLE"
echo "CSV Path: $CSVPATH"
echo "Iterations: $ITERATIONS"

# Check daemon requirements
if [ -z $(pidof mariadbd) ]; then
    printf "[X] MariaDB is offline\n\n"
    exit 1
fi
if [ -z $(pidof PrimProc) ]; then
    printf "[X] Columnstore is offline\n\n"
    exit 1
fi

# Ensure dependency is present
if ! command -v ioping >/dev/null ; then
    yum install ioping
    printf "\n\n[X] Rerun the script now that ioping is installed\n\n"
    exit 1
fi

# Create database if not exists
if [ -z $(mariadb -s -e "SHOW DATABASES LIKE '%$DATABASE%'") ]; then
    echo "[+] Creating Database: $DATABASE ..."
    mariadb -e "CREATE DATABASE $DATABASE"
else
    echo "[*] Database Exists"
fi

# [Table Generation, Row Verification, and CSV Export Logic goes here based on Manual Steps 1-3]

printf "[~] Beginning Import Test\n"
i=0
results=()

while [ $i -le $ITERATIONS ]; do
    printf "\n\nStarting Iteration ${i}\n"
    start=`date +%s.%N`
    cpimport -E '"' -s "," $DATABASE $TABLE $CSVPATH
    end=`date +%s.%N`
    results+=($( echo "$end - $start" | bc -l )) 
    ((i++))
done

# Perform automated ioping test
if ! command -v ioping; then
    printf "\n\n[X] Please make sure ioping is installed and executable\n\n"
    exit 1
else
    printf "\n\n Checking Latency of Data1 DBroot \n\n"
    ioping -c 30 /var/lib/columnstore/data1
fi

printf "\n\nCpimport Summary Timings\n"
for t in "${results[@]}"; do
    echo $t
done

_{This page is: Copyright © 2026 MariaDB. All rights reserved.}