Tair (Redis® OSS-Compatible):Performance white paper of persistent memory-optimized instances

Overview

Persistent memory-optimized instances use PMEM to provide large-capacity in-memory databases without relying on disks for data persistence. Every write operation is persisted directly to PMEM, maintaining the reliability of each operation. These instances cost 30% less than Redis Open-Source Edition instances while delivering almost the same throughput and latency.

For a product overview, see Persistent memory-optimized instances.

Test environment

Test tool

YCSB (Yahoo! Cloud Serving Benchmark) is an open source Java benchmarking framework that supports multiple database types. See YCSB on GitHub for installation instructions.

By default, YCSB tests only hashes. To benchmark strings, replace the RedisClient.java file in YCSB/redis/src/main/java/site/ycsb/db/ with the custom RedisClient.java after downloading YCSB.

Complex commands for sets and sorted sets are not included in these tests because their performance characteristics depend heavily on the specific use case. Design a custom workload for sets and sorted sets based on your application requirements.

Workloads

All tests use 8 GB of data with a Zipfian key distribution, which models real-world access patterns where a small fraction of keys receives the majority of requests. Three workloads are tested:

For workload definitions, see Core Workloads.

Test commands

# Load data
workload=a
./bin/ycsb load redis -s -P workloads/workload${workload} \
  -p "redis.host=${server_ip}" \
  -p "redis.port=${port}" \
  -p "redis.password=${password}" \
  -p "recordcount=${recordcount}" \
  -p "operationcount=${operationcount}" \
  -p "redis.timeout=30000" \
  -p "redis.command_group=${command_group}" \
  -p "fieldcount=${fieldcount}" \
  -p "fieldlength=${fieldlength}" \
  -threads ${threads} \
  -p "redis.password=***:*****"

# Run Workload C (100% read)
workload=c
./bin/ycsb run redis -s -P workloads/workload${workload} \
  -p "redis.host=${server_ip}" \
  -p "redis.port=${port}" \
  -p "redis.password=${password}" \
  -p "recordcount=${recordcount}" \
  -p "operationcount=${operationcount}" \
  -p "redis.timeout=30000" \
  -p "redis.command_group=${command_group}" \
  -p "fieldcount=${fieldcount}" \
  -p "fieldlength=${fieldlength}" \
  -threads ${threads} \
  -p "redis.password=***:*****"

# Run Workload A (50% read / 50% update)
workload=a
./bin/ycsb run redis -s -P workloads/workload${workload} \
  -p "redis.host=${server_ip}" \
  -p "redis.port=${port}" \
  -p "redis.password=${password}" \
  -p "recordcount=${recordcount}" \
  -p "operationcount=${operationcount}" \
  -p "redis.timeout=30000" \
  -p "redis.command_group=${command_group}" \
  -p "fieldcount=${fieldcount}" \
  -p "fieldlength=${fieldlength}" \
  -threads ${threads} \
  -p "redis.password=***:*****"

Test metrics

Test results for data loading

Data loading measures 100% write (INSERT) performance.

Strings

Hashes (10 fields per key)

Lists (10 elements per key)

Sets (10 members per key)

Sorted sets (10 elements per key)

Test results for Workload C

Workload C measures 100% read performance.

Strings

Hashes (10 fields per key)

Lists (10 elements per key)

Sets (10 members per key)

Sorted sets (10 elements per key)

Test results for Workload A

Workload A measures mixed-read/write performance: 50% reads and 50% updates.

Strings

Hashes (10 fields per key)

Lists (10 elements per key)

Sets (10 members per key)

Sorted sets (10 elements per key)

What's next

• Persistent memory-optimized instances — product overview and pricing

• Standard architecture — architecture details for the tested configuration

• Change or reset the password — manage instance credentials