Faster hash function

[itamarst]

Given pervasive use of dicts, Python's hash function gets used a lot. There are faster alternatives than SIP 2-4, e.g. ahash: https://github.com/tkaitchuck/aHash#comparison-with-other-hashers

ahash is still in need of cryptographic analysis to prove it has same protection as SIP 2-4, but (from using it in Rust) switching definitely has a meaningful performance improvement on hashmap lookups.

[gvanrossum]

The string hash doesn’t need to be cryptographically secure, doesn’t it? We use hash randomization with a cryptographically secure seed.

[itamarst]

My understanding (not an expert) is that there is cryptographically secure (e.g. SHA-512 or whatever), and that's definitely irrelevant to this use case.

But then there's denial-of-service proof, which is a much lower bar but still requires some of the same mathematical/cryptographic analysis to prove? And you do want DoS-proof. And that analysis has been done for SIP 2-4, which is why it's everyone's current default, but not for ahash (although the latter's author believes that it's DoS-proof: https://github.com/tkaitchuck/aHash/wiki/How-aHash-is-resists-DOS-attacks).

[itamarst]

So what I meant to say in initial issue was not so much "ahash is cryptographically insecure" (since that's fine) but rather "ahash needs more analysis to prove it's DoS-proof".

[gvanrossum]

How would an attacker DoS me with an "insecure" hash, given that hash randomization's seed is secure?

…

On Mon, Sep 27, 2021 at 8:25 AM Itamar Turner-Trauring < ***@***.***> wrote: So what I meant to say in initial issue was not so much "ahash is cryptographically insecure" (since that's fine) but rather "ahash needs more analysis to prove it's DoS-proof". — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#88 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAWCWMVRGR7WP3NKSH763Y3UECEFDANCNFSM5E2NXBDQ> . Triage notifications on the go with GitHub Mobile for iOS <https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675> or Android <https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub>.

-- --Guido van Rossum (python.org/~guido)

[itamarst]

https://www.python.org/dev/peps/pep-0456/ has a bunch of details; just learned from reading it that for a bad hash, the seed can be recovered by an attacker.

[gvanrossum]

In that case, we should probably table this improvement until the security folks have had time to look at this.

[markshannon]

String hashing is done once per string, when strings are created. It doesn't impact dict performance. The C-level mutability of strings has much more of an impact.

Reducing the time to hash strings will improve startup times, though. That is when much hashing occurs.

[gvanrossum]

String hashing is done once per string, when strings are created.

Correction: it's done lazily. So typically on the first time the string is looked up in any dict (including interning). A value of -1 indicates that the hash hasn't been computed yet.

I wouldn't want to change this, since most strings aren't used to look anything up in a dictionary (e.g. when reading lines from a file).

[pitrou]

I may be wrong, but the ahash benchmarks seem to have been run with Rust implementations of the hash algorithms.

My impression is that xxh3 (the third iteration of xxHash) is both widely used and extremely fast: https://cyan4973.github.io/xxHash/
The xxHash wiki also elaborates on small-string performance, which is certainly relevant for Python:
https://github.com/Cyan4973/xxHash/wiki/Performance-comparison#benchmarks-concentrating-on-small-data-

Note: xxHash APIs accept an optional seed, making it keyable. I have no idea whether it's DoS-resistant.

[itamarst]

That might be a good solution too, yes, the main point is that it is likely possible to switch to a faster hash than currently used one while still preserving anti-DoS guarantees.

[methane]

On Tue, Sep 28, 2021 at 1:02 AM Guido van Rossum ***@***.***> wrote: How would an attacker DoS me with an "insecure" hash, given that hash randomization's seed is secure?

Attackers may guess the key by sending many hashes and measuring timings. So having seed is not enough for anti-HashDoS. FYI, I had made an issue to change the hash algorithm from SipHash-2-4 to SipHash-1-3, as Ruby and Rust had moved already. https://bugs.python.org/issue29410 Microbemchmark result: https://gist.github.com/methane/33c7b01c45ce23b67246f5ddaff9c9e7 Regards,

…

-- Inada Naoki ***@***.***>

[tiran]

Please be careful when comparing hashing performance benchmark. Some benchmarks show off with impressive numbers like X GB/sec throughput. Dictionary keys are usually short. AFAIK majority of dict key strings are in the range of 5 to 25 characters. Good hashing algorithms perform extra steps to finalize a hash in order to archive good diffusion. The finalization step is a fixed cost and impact short strings. SipHash-1-3 is already faster for shorter strings than SipHash-2-4. It performs one finalization round less.

[itamarst]

In my experience, when hashing 64-bit integers (equivalent to <8 char ASCII strings) ahash is faster than Rust's default hasher, which I believe is SipHash 1-3. So it's possible to have faster hashes than SipHash 1-3 even on latency measures of small data. Pretty sure xxh3 also does well on latency, though it doesn't graph a comparison to SipHash 1-3.

[markshannon]

I think we should wait until ahash has been comprehensively analyzed in the same way as siphash has been before we adopt it. Hopefully that will happen sooner rather than later.

[tiran]

A while ago I have created a test branch with xxhash3, python/cpython@main...tiran:test_xxhash3 . I selected xxhash3 because it was the least effort to get it working. Could somebody run benchmarks, please? I would like to see if a different hashing algorithms makes a difference at all. I don't have the benchmark tool installed and configured.

[gvanrossum]

@ericsnowcurrently Could you do this some time next week?

[ericsnowcurrently]

I ran the benchmarks on main (python/cpython@8a45ca5) and on @tiran's test_xxhash3 branch (tiran/cpython@2e91018). The geometric mean shows no change in performance.

results:

req-1638816170-esnow - main
req-1638832245-esnow - test_xxhash3

2to3: Mean +- std dev: [/home/benchmarking/BENCH/REQUESTS/req-1638816170-esnow/results-data.json.gz] 271 ms +- 2 ms -> [/home/benchmarking/BENCH/REQUESTS/req-1638832245-esnow/results-data.json.gz] 271 ms +- 1 ms: 1.00x faster
chameleon: Mean +- std dev: [/home/benchmarking/BENCH/REQUESTS/req-1638816170-esnow/results-data.json.gz] 7.46 ms +- 0.06 ms -> [/home/benchmarking/BENCH/REQUESTS/req-1638832245-esnow/results-data.json.gz] 7.50 ms +- 0.09 ms: 1.01x slower
chaos: Mean +- std dev: [/home/benchmarking/BENCH/REQUESTS/req-1638816170-esnow/results-data.json.gz] 74.0 ms +- 0.7 ms -> [/home/benchmarking/BENCH/REQUESTS/req-1638832245-esnow/results-data.json.gz] 77.4 ms +- 0.8 ms: 1.05x slower
crypto_pyaes: Mean +- std dev: [/home/benchmarking/BENCH/REQUESTS/req-1638816170-esnow/results-data.json.gz] 87.0 ms +- 1.0 ms -> [/home/benchmarking/BENCH/REQUESTS/req-1638832245-esnow/results-data.json.gz] 86.4 ms +- 0.8 ms: 1.01x faster
deltablue: Mean +- std dev: [/home/benchmarking/BENCH/REQUESTS/req-1638816170-esnow/results-data.json.gz] 4.78 ms +- 0.07 ms -> [/home/benchmarking/BENCH/REQUESTS/req-1638832245-esnow/results-data.json.gz] 4.86 ms +- 0.06 ms: 1.02x slower
django_template: Mean +- std dev: [/home/benchmarking/BENCH/REQUESTS/req-1638816170-esnow/results-data.json.gz] 36.1 ms +- 0.5 ms -> [/home/benchmarking/BENCH/REQUESTS/req-1638832245-esnow/results-data.json.gz] 36.2 ms +- 0.3 ms: 1.00x slower
dulwich_log: Mean +- std dev: [/home/benchmarking/BENCH/REQUESTS/req-1638816170-esnow/results-data.json.gz] 67.0 ms +- 0.7 ms -> [/home/benchmarking/BENCH/REQUESTS/req-1638832245-esnow/results-data.json.gz] 67.8 ms +- 0.7 ms: 1.01x slower
fannkuch: Mean +- std dev: [/home/benchmarking/BENCH/REQUESTS/req-1638816170-esnow/results-data.json.gz] 386 ms +- 5 ms -> [/home/benchmarking/BENCH/REQUESTS/req-1638832245-esnow/results-data.json.gz] 414 ms +- 4 ms: 1.07x slower
float: Mean +- std dev: [/home/benchmarking/BENCH/REQUESTS/req-1638816170-esnow/results-data.json.gz] 80.1 ms +- 1.1 ms -> [/home/benchmarking/BENCH/REQUESTS/req-1638832245-esnow/results-data.json.gz] 82.3 ms +- 0.9 ms: 1.03x slower
go: Mean +- std dev: [/home/benchmarking/BENCH/REQUESTS/req-1638816170-esnow/results-data.json.gz] 162 ms +- 2 ms -> [/home/benchmarking/BENCH/REQUESTS/req-1638832245-esnow/results-data.json.gz] 166 ms +- 2 ms: 1.02x slower
json: Mean +- std dev: [/home/benchmarking/BENCH/REQUESTS/req-1638816170-esnow/results-data.json.gz] 5.10 ms +- 0.07 ms -> [/home/benchmarking/BENCH/REQUESTS/req-1638832245-esnow/results-data.json.gz] 4.84 ms +- 0.09 ms: 1.05x faster
json_loads: Mean +- std dev: [/home/benchmarking/BENCH/REQUESTS/req-1638816170-esnow/results-data.json.gz] 27.6 us +- 0.4 us -> [/home/benchmarking/BENCH/REQUESTS/req-1638832245-esnow/results-data.json.gz] 25.6 us +- 0.2 us: 1.08x faster
logging_format: Mean +- std dev: [/home/benchmarking/BENCH/REQUESTS/req-1638816170-esnow/results-data.json.gz] 6.53 us +- 0.10 us -> [/home/benchmarking/BENCH/REQUESTS/req-1638832245-esnow/results-data.json.gz] 6.57 us +- 0.10 us: 1.01x slower
logging_silent: Mean +- std dev: [/home/benchmarking/BENCH/REQUESTS/req-1638816170-esnow/results-data.json.gz] 115 ns +- 5 ns -> [/home/benchmarking/BENCH/REQUESTS/req-1638832245-esnow/results-data.json.gz] 109 ns +- 1 ns: 1.06x faster
logging_simple: Mean +- std dev: [/home/benchmarking/BENCH/REQUESTS/req-1638816170-esnow/results-data.json.gz] 6.00 us +- 0.06 us -> [/home/benchmarking/BENCH/REQUESTS/req-1638832245-esnow/results-data.json.gz] 6.08 us +- 0.10 us: 1.01x slower
meteor_contest: Mean +- std dev: [/home/benchmarking/BENCH/REQUESTS/req-1638816170-esnow/results-data.json.gz] 103 ms +- 2 ms -> [/home/benchmarking/BENCH/REQUESTS/req-1638832245-esnow/results-data.json.gz] 104 ms +- 1 ms: 1.01x slower
nbody: Mean +- std dev: [/home/benchmarking/BENCH/REQUESTS/req-1638816170-esnow/results-data.json.gz] 95.5 ms +- 0.8 ms -> [/home/benchmarking/BENCH/REQUESTS/req-1638832245-esnow/results-data.json.gz] 93.5 ms +- 2.1 ms: 1.02x faster
nqueens: Mean +- std dev: [/home/benchmarking/BENCH/REQUESTS/req-1638816170-esnow/results-data.json.gz] 86.4 ms +- 0.7 ms -> [/home/benchmarking/BENCH/REQUESTS/req-1638832245-esnow/results-data.json.gz] 85.9 ms +- 1.0 ms: 1.01x faster
pickle: Mean +- std dev: [/home/benchmarking/BENCH/REQUESTS/req-1638816170-esnow/results-data.json.gz] 10.1 us +- 0.3 us -> [/home/benchmarking/BENCH/REQUESTS/req-1638832245-esnow/results-data.json.gz] 9.90 us +- 0.08 us: 1.02x faster
pickle_dict: Mean +- std dev: [/home/benchmarking/BENCH/REQUESTS/req-1638816170-esnow/results-data.json.gz] 28.6 us +- 0.2 us -> [/home/benchmarking/BENCH/REQUESTS/req-1638832245-esnow/results-data.json.gz] 28.4 us +- 0.1 us: 1.00x faster
pickle_list: Mean +- std dev: [/home/benchmarking/BENCH/REQUESTS/req-1638816170-esnow/results-data.json.gz] 4.55 us +- 0.05 us -> [/home/benchmarking/BENCH/REQUESTS/req-1638832245-esnow/results-data.json.gz] 4.61 us +- 0.05 us: 1.01x slower
pickle_pure_python: Mean +- std dev: [/home/benchmarking/BENCH/REQUESTS/req-1638816170-esnow/results-data.json.gz] 358 us +- 3 us -> [/home/benchmarking/BENCH/REQUESTS/req-1638832245-esnow/results-data.json.gz] 356 us +- 3 us: 1.01x faster
pidigits: Mean +- std dev: [/home/benchmarking/BENCH/REQUESTS/req-1638816170-esnow/results-data.json.gz] 196 ms +- 0 ms -> [/home/benchmarking/BENCH/REQUESTS/req-1638832245-esnow/results-data.json.gz] 192 ms +- 1 ms: 1.02x faster
pycparser: Mean +- std dev: [/home/benchmarking/BENCH/REQUESTS/req-1638816170-esnow/results-data.json.gz] 1.27 sec +- 0.02 sec -> [/home/benchmarking/BENCH/REQUESTS/req-1638832245-esnow/results-data.json.gz] 1.22 sec +- 0.02 sec: 1.05x faster
pyflate: Mean +- std dev: [/home/benchmarking/BENCH/REQUESTS/req-1638816170-esnow/results-data.json.gz] 516 ms +- 3 ms -> [/home/benchmarking/BENCH/REQUESTS/req-1638832245-esnow/results-data.json.gz] 512 ms +- 4 ms: 1.01x faster
python_startup: Mean +- std dev: [/home/benchmarking/BENCH/REQUESTS/req-1638816170-esnow/results-data.json.gz] 13.4 ms +- 0.0 ms -> [/home/benchmarking/BENCH/REQUESTS/req-1638832245-esnow/results-data.json.gz] 13.2 ms +- 0.0 ms: 1.01x faster
python_startup_no_site: Mean +- std dev: [/home/benchmarking/BENCH/REQUESTS/req-1638816170-esnow/results-data.json.gz] 5.62 ms +- 0.00 ms -> [/home/benchmarking/BENCH/REQUESTS/req-1638832245-esnow/results-data.json.gz] 5.54 ms +- 0.00 ms: 1.01x faster
raytrace: Mean +- std dev: [/home/benchmarking/BENCH/REQUESTS/req-1638816170-esnow/results-data.json.gz] 333 ms +- 2 ms -> [/home/benchmarking/BENCH/REQUESTS/req-1638832245-esnow/results-data.json.gz] 331 ms +- 3 ms: 1.01x faster
regex_compile: Mean +- std dev: [/home/benchmarking/BENCH/REQUESTS/req-1638816170-esnow/results-data.json.gz] 141 ms +- 1 ms -> [/home/benchmarking/BENCH/REQUESTS/req-1638832245-esnow/results-data.json.gz] 140 ms +- 1 ms: 1.00x faster
regex_dna: Mean +- std dev: [/home/benchmarking/BENCH/REQUESTS/req-1638816170-esnow/results-data.json.gz] 207 ms +- 1 ms -> [/home/benchmarking/BENCH/REQUESTS/req-1638832245-esnow/results-data.json.gz] 213 ms +- 1 ms: 1.03x slower
regex_effbot: Mean +- std dev: [/home/benchmarking/BENCH/REQUESTS/req-1638816170-esnow/results-data.json.gz] 3.13 ms +- 0.04 ms -> [/home/benchmarking/BENCH/REQUESTS/req-1638832245-esnow/results-data.json.gz] 3.24 ms +- 0.08 ms: 1.03x slower
regex_v8: Mean +- std dev: [/home/benchmarking/BENCH/REQUESTS/req-1638816170-esnow/results-data.json.gz] 23.8 ms +- 0.5 ms -> [/home/benchmarking/BENCH/REQUESTS/req-1638832245-esnow/results-data.json.gz] 23.5 ms +- 0.3 ms: 1.01x faster
richards: Mean +- std dev: [/home/benchmarking/BENCH/REQUESTS/req-1638816170-esnow/results-data.json.gz] 58.0 ms +- 1.0 ms -> [/home/benchmarking/BENCH/REQUESTS/req-1638832245-esnow/results-data.json.gz] 56.0 ms +- 0.5 ms: 1.04x faster
scimark_lu: Mean +- std dev: [/home/benchmarking/BENCH/REQUESTS/req-1638816170-esnow/results-data.json.gz] 112 ms +- 2 ms -> [/home/benchmarking/BENCH/REQUESTS/req-1638832245-esnow/results-data.json.gz] 111 ms +- 2 ms: 1.01x faster
scimark_sor: Mean +- std dev: [/home/benchmarking/BENCH/REQUESTS/req-1638816170-esnow/results-data.json.gz] 140 ms +- 1 ms -> [/home/benchmarking/BENCH/REQUESTS/req-1638832245-esnow/results-data.json.gz] 141 ms +- 3 ms: 1.01x slower
sympy_expand: Mean +- std dev: [/home/benchmarking/BENCH/REQUESTS/req-1638816170-esnow/results-data.json.gz] 500 ms +- 3 ms -> [/home/benchmarking/BENCH/REQUESTS/req-1638832245-esnow/results-data.json.gz] 499 ms +- 4 ms: 1.00x faster
sympy_integrate: Mean +- std dev: [/home/benchmarking/BENCH/REQUESTS/req-1638816170-esnow/results-data.json.gz] 21.7 ms +- 0.1 ms -> [/home/benchmarking/BENCH/REQUESTS/req-1638832245-esnow/results-data.json.gz] 21.8 ms +- 0.2 ms: 1.00x slower
sympy_str: Mean +- std dev: [/home/benchmarking/BENCH/REQUESTS/req-1638816170-esnow/results-data.json.gz] 301 ms +- 3 ms -> [/home/benchmarking/BENCH/REQUESTS/req-1638832245-esnow/results-data.json.gz] 303 ms +- 3 ms: 1.01x slower
telco: Mean +- std dev: [/home/benchmarking/BENCH/REQUESTS/req-1638816170-esnow/results-data.json.gz] 6.81 ms +- 0.15 ms -> [/home/benchmarking/BENCH/REQUESTS/req-1638832245-esnow/results-data.json.gz] 6.64 ms +- 0.17 ms: 1.03x faster
unpack_sequence: Mean +- std dev: [/home/benchmarking/BENCH/REQUESTS/req-1638816170-esnow/results-data.json.gz] 45.3 ns +- 1.1 ns -> [/home/benchmarking/BENCH/REQUESTS/req-1638832245-esnow/results-data.json.gz] 46.2 ns +- 0.5 ns: 1.02x slower
unpickle: Mean +- std dev: [/home/benchmarking/BENCH/REQUESTS/req-1638816170-esnow/results-data.json.gz] 13.9 us +- 0.6 us -> [/home/benchmarking/BENCH/REQUESTS/req-1638832245-esnow/results-data.json.gz] 13.5 us +- 0.1 us: 1.03x faster
unpickle_list: Mean +- std dev: [/home/benchmarking/BENCH/REQUESTS/req-1638816170-esnow/results-data.json.gz] 5.21 us +- 0.05 us -> [/home/benchmarking/BENCH/REQUESTS/req-1638832245-esnow/results-data.json.gz] 5.14 us +- 0.08 us: 1.01x faster
unpickle_pure_python: Mean +- std dev: [/home/benchmarking/BENCH/REQUESTS/req-1638816170-esnow/results-data.json.gz] 269 us +- 3 us -> [/home/benchmarking/BENCH/REQUESTS/req-1638832245-esnow/results-data.json.gz] 267 us +- 2 us: 1.01x faster
xml_etree_parse: Mean +- std dev: [/home/benchmarking/BENCH/REQUESTS/req-1638816170-esnow/results-data.json.gz] 152 ms +- 3 ms -> [/home/benchmarking/BENCH/REQUESTS/req-1638832245-esnow/results-data.json.gz] 150 ms +- 3 ms: 1.01x faster
xml_etree_iterparse: Mean +- std dev: [/home/benchmarking/BENCH/REQUESTS/req-1638816170-esnow/results-data.json.gz] 105 ms +- 2 ms -> [/home/benchmarking/BENCH/REQUESTS/req-1638832245-esnow/results-data.json.gz] 104 ms +- 2 ms: 1.01x faster
xml_etree_generate: Mean +- std dev: [/home/benchmarking/BENCH/REQUESTS/req-1638816170-esnow/results-data.json.gz] 79.4 ms +- 1.0 ms -> [/home/benchmarking/BENCH/REQUESTS/req-1638832245-esnow/results-data.json.gz] 79.1 ms +- 0.4 ms: 1.00x faster
xml_etree_process: Mean +- std dev: [/home/benchmarking/BENCH/REQUESTS/req-1638816170-esnow/results-data.json.gz] 58.0 ms +- 1.1 ms -> [/home/benchmarking/BENCH/REQUESTS/req-1638832245-esnow/results-data.json.gz] 57.2 ms +- 0.6 ms: 1.01x faster

Benchmark hidden because not significant (14): hexiom, json_dumps, mako, pathlib, scimark_fft, scimark_monte_carlo, scimark_sparse_mat_mult, spectral_norm, sqlalchemy_declarative, sqlalchemy_imperative, sqlite_synth, sympy_sum, thrift, tornado_http

Geometric mean: 1.00x faster

Grouped:

Slower (18):
- fannkuch: 386 ms +- 5 ms -> 414 ms +- 4 ms: 1.07x slower
- chaos: 74.0 ms +- 0.7 ms -> 77.4 ms +- 0.8 ms: 1.05x slower
- regex_effbot: 3.13 ms +- 0.04 ms -> 3.24 ms +- 0.08 ms: 1.03x slower
- float: 80.1 ms +- 1.1 ms -> 82.3 ms +- 0.9 ms: 1.03x slower
- regex_dna: 207 ms +- 1 ms -> 213 ms +- 1 ms: 1.03x slower
- go: 162 ms +- 2 ms -> 166 ms +- 2 ms: 1.02x slower
- unpack_sequence: 45.3 ns +- 1.1 ns -> 46.2 ns +- 0.5 ns: 1.02x slower
- deltablue: 4.78 ms +- 0.07 ms -> 4.86 ms +- 0.06 ms: 1.02x slower
- logging_simple: 6.00 us +- 0.06 us -> 6.08 us +- 0.10 us: 1.01x slower
- pickle_list: 4.55 us +- 0.05 us -> 4.61 us +- 0.05 us: 1.01x slower
- dulwich_log: 67.0 ms +- 0.7 ms -> 67.8 ms +- 0.7 ms: 1.01x slower
- meteor_contest: 103 ms +- 2 ms -> 104 ms +- 1 ms: 1.01x slower
- scimark_sor: 140 ms +- 1 ms -> 141 ms +- 3 ms: 1.01x slower
- logging_format: 6.53 us +- 0.10 us -> 6.57 us +- 0.10 us: 1.01x slower
- chameleon: 7.46 ms +- 0.06 ms -> 7.50 ms +- 0.09 ms: 1.01x slower
- sympy_str: 301 ms +- 3 ms -> 303 ms +- 3 ms: 1.01x slower
- sympy_integrate: 21.7 ms +- 0.1 ms -> 21.8 ms +- 0.2 ms: 1.00x slower
- django_template: 36.1 ms +- 0.5 ms -> 36.2 ms +- 0.3 ms: 1.00x slower

Faster (29):
- json_loads: 27.6 us +- 0.4 us -> 25.6 us +- 0.2 us: 1.08x faster
- logging_silent: 115 ns +- 5 ns -> 109 ns +- 1 ns: 1.06x faster
- json: 5.10 ms +- 0.07 ms -> 4.84 ms +- 0.09 ms: 1.05x faster
- pycparser: 1.27 sec +- 0.02 sec -> 1.22 sec +- 0.02 sec: 1.05x faster
- richards: 58.0 ms +- 1.0 ms -> 56.0 ms +- 0.5 ms: 1.04x faster
- unpickle: 13.9 us +- 0.6 us -> 13.5 us +- 0.1 us: 1.03x faster
- telco: 6.81 ms +- 0.15 ms -> 6.64 ms +- 0.17 ms: 1.03x faster
- pidigits: 196 ms +- 0 ms -> 192 ms +- 1 ms: 1.02x faster
- nbody: 95.5 ms +- 0.8 ms -> 93.5 ms +- 2.1 ms: 1.02x faster
- pickle: 10.1 us +- 0.3 us -> 9.90 us +- 0.08 us: 1.02x faster
- python_startup_no_site: 5.62 ms +- 0.00 ms -> 5.54 ms +- 0.00 ms: 1.01x faster
- regex_v8: 23.8 ms +- 0.5 ms -> 23.5 ms +- 0.3 ms: 1.01x faster
- python_startup: 13.4 ms +- 0.0 ms -> 13.2 ms +- 0.0 ms: 1.01x faster
- xml_etree_process: 58.0 ms +- 1.1 ms -> 57.2 ms +- 0.6 ms: 1.01x faster
- unpickle_list: 5.21 us +- 0.05 us -> 5.14 us +- 0.08 us: 1.01x faster
- xml_etree_iterparse: 105 ms +- 2 ms -> 104 ms +- 2 ms: 1.01x faster
- xml_etree_parse: 152 ms +- 3 ms -> 150 ms +- 3 ms: 1.01x faster
- scimark_lu: 112 ms +- 2 ms -> 111 ms +- 2 ms: 1.01x faster
- unpickle_pure_python: 269 us +- 3 us -> 267 us +- 2 us: 1.01x faster
- crypto_pyaes: 87.0 ms +- 1.0 ms -> 86.4 ms +- 0.8 ms: 1.01x faster
- pickle_pure_python: 358 us +- 3 us -> 356 us +- 3 us: 1.01x faster
- raytrace: 333 ms +- 2 ms -> 331 ms +- 3 ms: 1.01x faster
- pyflate: 516 ms +- 3 ms -> 512 ms +- 4 ms: 1.01x faster
- nqueens: 86.4 ms +- 0.7 ms -> 85.9 ms +- 1.0 ms: 1.01x faster
- xml_etree_generate: 79.4 ms +- 1.0 ms -> 79.1 ms +- 0.4 ms: 1.00x faster
- pickle_dict: 28.6 us +- 0.2 us -> 28.4 us +- 0.1 us: 1.00x faster
- 2to3: 271 ms +- 2 ms -> 271 ms +- 1 ms: 1.00x faster
- regex_compile: 141 ms +- 1 ms -> 140 ms +- 1 ms: 1.00x faster
- sympy_expand: 500 ms +- 3 ms -> 499 ms +- 4 ms: 1.00x faster

[pitrou]

It would be worth checking a microbenchmark of hashing or dict-lookup of largish strings.