Le calcul d'un hachage MD5 est-il moins gourmand en ressources CPU que les fonctions de la famille SHA ?

Question

Le calcul d'un hachage MD5 est-il moins gourmand en ressources CPU que SHA-1 ou SHA-2 sur le matériel x86 "standard" des ordinateurs portables ? Je suis intéressé par des informations générales, pas spécifiques à une certaine puce.

UPDATE : Dans mon cas, je suis intéressé par le calcul du hachage d'un fichier. Si la taille du fichier est importante, supposons qu'il s'agit de 300K.

Answer 1

Sha1sum est un peu plus rapide sur Power9 que md5sum.

$ uname -mov
#1 SMP Mon May 13 12:16:08 EDT 2019 ppc64le GNU/Linux

$ cat /proc/cpuinfo
processor       : 0
cpu             : POWER9, altivec supported
clock           : 2166.000000MHz
revision        : 2.2 (pvr 004e 1202)

$ ls -l linux-master.tar
-rw-rw-r-- 1 x x 829685760 Jan 29 14:30 linux-master.tar

$ time sha1sum linux-master.tar
10fbf911e254c4fe8e5eb2e605c6c02d29a88563  linux-master.tar

real    0m1.685s
user    0m1.528s
sys     0m0.156s

$ time md5sum linux-master.tar
d476375abacda064ae437a683c537ec4  linux-master.tar

real    0m2.942s
user    0m2.806s
sys     0m0.136s

$ time sum linux-master.tar
36928 810240

real    0m2.186s
user    0m1.917s
sys     0m0.268s

Answer 2

[
Is MD5 faster or SHA1? ]

It's implementation dependent:

|*|
[
Theoretically the MD5 algorithm would do less work than SHA1, but the design of MD5 itself determined that the algorithm cannot effectively exploit computation parallelism (i.e. cannot effectively utilize a multi-processor system; or processors that utilize instruction-level parallelism). While SHA1 would provide better opportunity for so.

This is part of the reason why in some implementations SHA1 would outperform MD5. ]

|*|
[
There are also processors that provide dedicated hardware acceleration support for SHA1.

When properly utilized, such implementations tend to easily outperform software based MD5 implementations:

[ Quote dr-js @ CE 2021-01-28 10:31 UTC:
https://security.stackexchange.com/a/95697

2021 update with OpenSSL 1.1.1d: now we see md5 is often slower on newer CPU, and for larger chunks:

[
## PC i7-1165G7 @ 2.80GHz (2020)
OpenSSL 1.1.1d  10 Sep 2019 / built on: Mon Dec  7 20:44:45 2020 UTC
type      16 bytes    64 bytes    256 bytes   1024 bytes   8192 bytes  16384 bytes
md5     189018.70k  418310.85k   712090.28k   890189.14k   956293.12k   962560.00k
sha1    287134.62k  746529.17k  1474064.38k  1973607.08k  2197842.60k  2192179.20k
sha256  222301.71k  603962.47k  1213340.33k  1665262.59k  1849016.32k  1847388.84k

## Server AMD EPYC 7571 (2018)
OpenSSL 1.1.1d  10 Sep 2019 / built on: Mon Dec  7 20:44:45 2020 UTC
type      16 bytes    64 bytes    256 bytes   1024 bytes   8192 bytes  16384 bytes
md5      93668.33k  213979.18k   378971.56k   467472.38k   501205.67k   504064.68k
sha1    165020.82k  442991.72k   888443.48k  1188591.62k  1319236.95k  1330080.43k
sha256  142886.55k  375612.63k   791567.70k  1095950.34k  1234381.48k  1246827.86k

## Server E5-2682 v4 @ 2.50GHz (2016)
OpenSSL 1.1.1d  10 Sep 2019 / built on: Mon Dec  7 20:44:45 2020 UTC
type      16 bytes    64 bytes    256 bytes   1024 bytes   8192 bytes  16384 bytes
md5     101505.24k  207422.92k   393158.83k   453332.99k   527085.34k   490711.72k
sha1     98091.83k  249828.79k   389640.36k   675694.25k   686966.33k   721021.61k
sha256   55421.86k  130103.33k   251929.17k   302571.86k   296977.81k   338439.56k
] ]

Worth noticing that even SHA-256 could be faster than MD5 in such cases. ]

To put it in a simple (though not so accurate) statement:

|*| For high-end processors, SHA1 tends to be faster.
|*| For low-end processors, MD5 would be faster.

[ Quote Nyan @ CE 2020-12-10 10:18 UTC:
https://stackoverflow.com/a/64928816

Note that all of the above only applies to single buffer hashing (by far the most common use case). If you're fancy and computing multiple hashes in parallel, i.e. a multi-buffer SIMD approach, the behaviour changes somewhat. ]