Re: Bootup and login performance
Hi,
(I started a new thread from "Kullervo")
On lauantai 26 tammikuu 2013, Thorsten Glaser wrote:
> Eero Tamminen dixit:
> >Sure, the main thing where it was annoyingly slower, was logging in
> >(as root). That hopefully doesn't run udev. ;-)
>
> No, it just runs sha512 10000 times in a row on the salted password
> to hash it instead of md5crypt or just unix crypt.
At least my /etc/pam.d/common-password doesn't have the rounds
keyword mentioned here:
http://forums.debian.net/viewtopic.php?f=30&t=60679&start=30
so I assume it does the hashing only once?
> (Or something like that, I’ve got the full paper on it by drepper
> mirrored here somewhere since I implemented it in Java™ for my
> coworkers. Not knowing any Java™, I might add.)
>
> I’ve got no idea how to change the default algorithm back to
> md5crypt, that’s a debian user question.
http://www.debian.org/doc/manuals/securing-debian-howto/ch4.en.html#s4.11
> Afterwards, just setting
> a new password with passwd should then use md5crypt.
>
> $ sudo sed -n /^root/p /etc/shadow
> gives you an indication: $6$ is the slow one, $1$ md5crypt, and
> no $ at all UNIX crypt().
After changing "sha512" in /etc/pam.d/common-password to "md5",
login is nearly instant. Thanks!
The difference is really huge, somebody should really look into
that at some point... Does any of the kernel profiling functionality
work on m68k port?
I checked Oprofile and at least according to its www-site it
doesn't support m68k and I haven't heard m68k to have performance
counter registers, so any tools using such are anyway out [1].
- Eero
[1] e.g. ARM performance counters are typically buggy, so with
Oprofile one often gets better results with sampling done
at suitable interval and just running the test longer.
Reply to: