Hash functions are widely used, so it is desirable to increase their speed and security. This package provides three hash functions that are resistant to hash flooding and outperform existing algorithms: a faster version of SipHash, a data-parallel variant of SipHash using tree hashing, and an even faster algorithm we call HighwayHash.
SipHash is a fast but cryptographically strong pseudo-random function by Aumasson and Bernstein [https://www.131002.net/siphash/siphash.pdf].
SipTreeHash slices inputs into 8-byte packets and computes their SipHash in parallel, which is faster when processing at least 96 bytes.
HighwayHash is a new way of mixing inputs which may inspire new cryptographically strong hashes. Large inputs are processed at a rate of 0.3 cycles per byte, and latency remains low even for small inputs. HighwayHash is faster than SipHash for all input sizes, with about 7 times higher throughput at 1 KiB.
Expected applications include DOS-proof hash tables and random generators.
SipHash is immune to hash flooding because multi-collisions are infeasible to compute. This makes it suitable for hash tables storing user-controlled data.
The output is also indistinguishable from a uniform random function, which means it can be used for choosing random subsets (e.g. for A/B experiments). Such generators are idempotent (repeatable/deterministic), which is useful in parallel algorithms and for testing/verification.
We have verified the bit distribution and avalanche properties of HighwayHash, though new cryptanalysis tools may need to be developed to analyze it.
Our SipHash implementation aims for maximum efficiency while remaining compatible with the reference C code. Outputs are identical for the given test cases (messages between 0 and 63 bytes).
Compared to Bernstein's SSE4.1 implementation (https://goo.gl/80GBSD), it is about 1.5x faster on the same machine. About 10% of that is tuning and the rest is achieved with AVX-2 independent-shift instructions.
The implementation uses custom vector classes with overloaded operators (e.g. const V2x64U a = b + c) for type-safety and improved readability vs. compiler intrinsics (e.g. const __m128i a = _mm_add_epi64(b, c)).
Even faster throughput can be achieved by logically partitioning inputs into interleaved streams and hashing them independently. The resulting hashes are then combined via original SipHash. Such "tree hash" constructions retain the safety guarantees of the underlying hash function.
Example: 64 byte input = 8 qwords. Interpret them as four interleaved streams: A0, A1, A2, A3, B0, B1, B2, B3. Compute four separate hashes of A0, A1, A2, A3 via four-way SIMD; update each of these with the second qwords B0, B1, B2, B3. Each independent hash result H_i includes A_i and B_i. Finally, combine the H_i into a single digest via SipHash.
By making full use of AVX-2 vectors, this leads to a 3x speedup vs. our compatible implementation. However, the output no longer matches SipHash.
We have devised a new way of mixing inputs with AVX-2 multiply and permute instructions. The multiplications are 32x32 -> 64 bits and therefore infeasible to reverse. Permuting equalizes the distribution of the resulting bytes.
The internal state occupies two 256-bit AVX-2 registers, v0 and v1. Due to limitations of the instruction set, the registers are partitioned into two halves that remain independent until the reduce phase. Operations are 64-bit, so each register half contains two lanes. To improve distribution, four bits of each v0 lane are hardwired to 1. Therefore, the effective size of each state half is 248 bits. The algorithm outputs 64 bit digests, which can easily be extended to 256 bits by no longer discarding the remaining bits.
In addition to high throughput, the algorithm is designed for low finalization cost. This enables a 2-3x speedup versus SipTreeHash, especially for smaller inputs.
Performance is measured as throughput for 1 KiB messages. The benchmark measures the minimum elapsed time among many repetitions (thus ensuring the inputs are cache-resident) but also updates an output variable to ensure the compiler does not elide anything. The C++ implementations are compiled with GCC 4.8.4 and run on a single core of a desktop Xeon E5-1650 v3 clocked at 3.5 GHz.
SipHash: 1.7 GB/s SipTreeHash: 4.8 GB/s HighwayHash: 11.3 GB/s
The software requires AVX-2-capable CPUs (Intel Haswell or upcoming AMD).
To build with Bazel (http://bazel.io/) : bazel build :all -c opt --copt=-mavx2
sip_hash.cc is the compatible implementation of SipHash, and also provides the final reduction for sip_tree_hash. sip_tree_hash.cc is the faster but incompatible SIMD j-lanes tree hash. highway_tree_hash.cc is our new, fast AVX-2 mixing algorithm. scalar_sip_tree_hash.cc and scalar_highway_tree_hash.cc are non-SIMD versions. vec2.h contains a wrapper class for 256-bit AVX-2 vectors with 64-bit lanes. vec.h provides a similar class for 128-bit vectors. code_annotation.h defines some compiler-dependent language extensions.
By Jan Wassenberg [email protected] and Jyrki Alakuijala [email protected], 2016-03-01
This is not an official Google product.