Performance Benchmarks

Hazelcast vs Redis

Hazelcast has performed a number of performance benchmarks against Redis providing a core set of conclusions:
  • Hazelcast consistently outperforms Redis at scale. Hazelcast is faster than Redis at scale and the gap widens as the workload increases. The benchmarks show that as the workload increases (thus requiring more concurrent threads), Hazelcast maintains its high-throughput and low-latency performance.
  • Published benchmarks that favor Redis do not use equal comparisons. In our latest benchmark reports, we call out several discrepancies in other benchmarks (not run by Hazelcast) that skew results in favor of Redis. Once the tests and configurations are normalized, Hazelcast performance outshines Redis. Also, Redis stops replicating data at high loads, which puts your system at risk for data loss (more details are available in this blog about tradeoffs in distributed systems).
  • The Hazelcast architecture is the key to performance. Hazelcast was built for performance; for example, its multi-threaded architecture (Redis is single-threaded), near-cachepipelining, and other capabilities give customers the highest throughput and lowest latency.

Recent Benchmark Results

Our latest performance benchmark tests compare Hazelcast Enterprise 3.12 versus Redis Open Source 5.0.3. The results are similar to past benchmarks, highlighting Hazelcast’s performance advantage at scale. The table below shows some key performance metrics:

Key MetricsHazelcast Enterprise 3.12Redis Open Source 5.0.3

Throughput using 128 threads1

1,190,000 ops/sec660,000 ops/sec

Throughput using 64 threads2

636,000 ops/sec614,000 ops/sec

Throughput as load scaled3

(see chart below)

Scaled linearly to 128 threadsMaxed out at 32 threads, performance degraded beyond that

1 In the Hazelcast Responds to Redis Labs’ Benchmark report.
2 In the Hazelcast IMDG Enterprise 3.12 vs Redis Open Source 5.0.3 report.
3 In the Hazelcast Responds to Redis Labs’ Benchmark report.

Hazelcast scales with greater load. Redis cannot scale beyond 32 concurrent threads.

See the results of the latest benchmarks and read more about the background of these tests: