[Bug 14391] New: sha1sum slow on x64 and possibly others
bugzilla at busybox.net
bugzilla at busybox.net
Thu Nov 25 21:28:30 UTC 2021
https://bugs.busybox.net/show_bug.cgi?id=14391
Bug ID: 14391
Summary: sha1sum slow on x64 and possibly others
Product: Busybox
Version: unspecified
Hardware: All
OS: All
Status: NEW
Severity: enhancement
Priority: P5
Component: Other
Assignee: unassigned at busybox.net
Reporter: blazejroszkowski at gmail.com
CC: busybox-cvs at busybox.net
Target Milestone: ---
Created attachment 9171
--> https://bugs.busybox.net/attachment.cgi?id=9171&action=edit
dot config file from my build on Fedora VM
sha1sum in BusyBox is over twice as slow as decently optimized C
implementation.
All tests were done on x64-86 (VMs and real OSes, Windws and Linux, two
laptops). I don't know if this is a performance problem on other architectures
(I'd guess it is).
Test file (1 GiB): dd if=/dev/urandom of=gig.gig bs=1024 count=$((1024 * 1024))
GNU coreutils sha1sum (Git for Windows, no libcrypto use) and my own
implementation take (on my personal laptop) 2.8 seconds.
BusyBox takes 6.3 seconds. It's around 2x slower on my work laptop as well.
This is present in at least versions: 1.34.1 (my own build on Fedora 34,
.config attached), 1.34.1 in latest Alpine (3.15.0), 1.34.1 from Fedora 34
repos, 1.34.0 on Windows.
I also remember it being present on Ubuntu 18.04 LTS and 20.04 LTS as well
(busybox from the repos).
My optimized plain C sha1 implementation (that I'm happy to contribute) is
here: https://github.com/FRex/blasha1
The only downside I see from using optimized C version is potential increase in
binary size, since optimized code is heavily unrolled (but I didn't investigate
this increase).
I've searched for "performance", "sha1", "sha1sum", and "speed" on this
Bugzilla and found nothing about this.
I understand it's an old semi-obsolete algorithm but if BusyBox provides this
util, I assume it's better that it's faster rather than slower.
--
You are receiving this mail because:
You are on the CC list for the bug.
More information about the busybox-cvs
mailing list