KPIR-C: Keyword PIR with Arbitrary Server-side Computation

This repository provides the implementation and evaluation artifacts for our paper “KPIR-C: Keyword PIR with Arbitrary Server-side Computation,” accepted at NDSS 2027.

Contents

• Repository Layout
• Requirements
• Installation and Build
• Artifact Evaluation
• Configuration
• Running Single-Query KPIR
• Running Multi-Query KPIR-C
• Citation and License

Repository Layout

.
├── README.md
├── .gitignore
├── artifact/
│   └── README.md
├── scripts/
├── results/
└── tkpir/
    ├── Cargo.toml
    ├── Cargo.lock
    ├── benches/bench.rs
    ├── common_tool/
    └── src/
        ├── lib.rs
        └── bin/
            └── bench_kpir_c_multi.rs

Requirements

• Stable Rust toolchain with cargo
• Linux is recommended; macOS is also supported
  • Tested on Ubuntu 22.04.1 LTS (x86_64) with stable Rust/Cargo and TFHE-rs v0.10.0
  • Tested on macOS 26.5.2, Apple Silicon (arm64), with Rust/Cargo 1.97.1
• 8 GB RAM is expected to be sufficient for databases of size m ≤ 2^16; at least 32 GB RAM is recommended for m > 2^16

Installation and Build

Install Rust if needed:

curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh
source "$HOME/.cargo/env"

Then, from the repository root, build the artifact:

cargo build --manifest-path tkpir/Cargo.toml --release --all-targets

Artifact Evaluation

Detailed instructions for validating and reproducing the paper's experimental results are provided in artifact/README.md.

For a quick validation, run from the repository root:

bash scripts/quick_check.sh

The full reproduction scripts are:

bash scripts/reproduce_table_iv.sh
bash scripts/reproduce_table_v.sh
bash scripts/reproduce_table_vii.sh
bash scripts/reproduce_table_x.sh --quick

The helper scripts are convenience wrappers around the benchmark commands described below. See artifact/README.md for experiment configurations, expected outputs, resource requirements, and paper results.

Configuration

The benchmark entrypoints use the following environment variables:

• TKPIR_SCHEME: TCWKPIR, TCAKPIR, or TCAKPIR_OPT
• TKPIR_NUM_THREADS: number of threads
• TKPIR_MAX_KEYWORD_BITLEN: keyword bit length, where x corresponds to m = 2^x database entries
• TKPIR_MAX_DATA_BITLEN: entry size in bits
• TKPIR_NUM_WARMUP_RUNS: number of warmup runs (default: 0)
• TKPIR_NUM_MEASURED_RUNS: number of measured runs (default: 1)

By default, benchmarks use no warmup and one measured run for fast validation. For more stable runtime measurements consistent with the paper, set:

TKPIR_NUM_WARMUP_RUNS=1
TKPIR_NUM_MEASURED_RUNS=10

TCAKPIR_OPT is the implementation name for TCAKPIR+.

Running Single-Query KPIR

The single-query benchmark entrypoint is tests::test_tkpir in tkpir/src/lib.rs.

From tkpir/, use:

TKPIR_SCHEME=<SCHEME> \
TKPIR_NUM_THREADS=<THREADS> \
TKPIR_MAX_KEYWORD_BITLEN=<LOG_M> \
TKPIR_MAX_DATA_BITLEN=<DATA_BITS> \
TKPIR_NUM_WARMUP_RUNS=<WARMUP_RUNS> \
TKPIR_NUM_MEASURED_RUNS=<MEASURED_RUNS> \
cargo test --lib tests::test_tkpir --release -- --exact --nocapture

For example, to run TCAKPIR with m = 2^12, 8-byte entries, and one thread:

TKPIR_SCHEME=TCAKPIR \
TKPIR_NUM_THREADS=1 \
TKPIR_MAX_KEYWORD_BITLEN=12 \
TKPIR_MAX_DATA_BITLEN=64 \
TKPIR_NUM_WARMUP_RUNS=0 \
TKPIR_NUM_MEASURED_RUNS=1 \
cargo test --lib tests::test_tkpir --release -- --exact --nocapture

Supported schemes are:

TCWKPIR
TCAKPIR
TCAKPIR_OPT

Running Multi-Query KPIR-C

The multi-query benchmark entrypoint is bench_kpir_c_multi in tkpir/src/bin/bench_kpir_c_multi.rs.

From tkpir/, use:

TKPIR_NUM_THREADS=<THREADS> \
TKPIR_MAX_KEYWORD_BITLEN=<LOG_M> \
TKPIR_MAX_DATA_BITLEN=<DATA_BITS> \
TKPIR_NUM_WARMUP_RUNS=<WARMUP_RUNS> \
TKPIR_NUM_MEASURED_RUNS=<MEASURED_RUNS> \
cargo run --release --bin bench_kpir_c_multi -- \
  --scheme <SCHEME> \
  --function <FUNCTION>

Supported schemes are:

TCWKPIR
TCAKPIR
TCAKPIR_OPT

Supported functions are:

sum
threshold

For example:

TKPIR_NUM_THREADS=64 \
TKPIR_MAX_KEYWORD_BITLEN=12 \
TKPIR_MAX_DATA_BITLEN=64 \
TKPIR_NUM_WARMUP_RUNS=0 \
TKPIR_NUM_MEASURED_RUNS=1 \
cargo run --release --bin bench_kpir_c_multi -- \
  --scheme TCAKPIR \
  --function sum

The benchmark evaluates t ∈ {1, 2, 4, 8, 16} automatically.

To skip the KPIR Response stage, append:

--skip-kpir

Citation and License

Citation (ePrint version):

@article{DBLP:journals/iacr/ArastehfardLZSPQFH25,
  author       = {Ali Arastehfard and
                  Weiran Liu and
                  Qixian Zhou and
                  Zinan Shen and
                  Liqiang Peng and
                  Lin Qu and
                  Shuya Feng and
                  Yuan Hong},
  title        = {{KPIR-C:} Keyword {PIR} with Arbitrary Server-side Computation},
  journal      = {{IACR} Cryptol. ePrint Arch.},
  volume       = {2025},
  pages        = {1952},
  year         = {2025}
}

Licensing terms are provided in LICENSE.