Programming Languages & Database Benchmarks

Are programming languages really faster when applications perform actual database operations under load?

The Research Question

Programming languages are frequently described as faster, more efficient, or more scalable than their competitors. However, many published comparisons measure simple operations that do not represent actual applications.

Real applications communicate with databases, process results, manage connections, and serve multiple users simultaneously.

We wanted to investigate whether performance differences remain significant when different languages perform comparable database workloads.

How We Conducted the Experiments

We developed API implementations in different programming languages and connected them to database systems.

Each implementation was expected to perform the same logical operations and return correct results.

We measured performance at increasing concurrency levels, recording both successful requests and failures.

Throughput

The number of successful requests processed per second.

Latency

Average response time, median response time, and higher-percentile latency.

Reliability

Errors, timeouts, and the concurrency levels where implementations become unstable.

Resource Usage

Performance under controlled CPU and memory limits compared with broader resource access.

Programming Languages

The original comparison included thirteen languages:

C++, Elixir, Go, Haskell, Java, Nim, Node.js/JavaScript, PHP, Python, Ruby, Rust, V, and Zig.

Research Changes — October 2026

Haskell was discontinued from subsequent testing rounds because we decided not to invest additional development time in that implementation.

This decision does not establish that Haskell is universally slow or unsuitable.

C# was added for the final three database comparisons.

Database Systems

Our research involved several database technologies, including:

Not every database completed every experiment. Some systems were removed or deferred because configuration or execution problems prevented meaningful comparisons.

Research History

First Round

The initial comparison tested the original language implementations against Cassandra under limited resources. We collected throughput, latency, and error measurements.

Second Round

A second round was conducted to compare results and investigate whether initial findings could be reproduced.

CockroachDB

During preflight testing, a license-enforced concurrent-transaction limit affected the workload.

We documented the limitation rather than treating the resulting errors as valid evidence of programming language performance.

YugabyteDB

Configuration and execution problems prevented a fair completed comparison at that stage.

What We Learned

Programming language performance depends on more than the language itself.

Database behavior, connection management, concurrency, drivers, and implementation details can all affect measured performance.

A language performing well at low concurrency may experience errors or bottlenecks as concurrency increases.

Unexpected results require additional investigation before drawing broad conclusions.

We do not consider a single benchmark sufficient to declare one programming language universally superior.

Research Status

The benchmark work is retained as part of our research archive. Its measurements and findings can help inform technology decisions for future development projects.

The published results should be interpreted in the context of their specific workloads, configurations, and limitations.

Detailed Benchmark Results

The original benchmark results page contains the measurements collected during testing.

View Benchmark Results

Return to Research Projects