[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <87setph63h.fsf@oldenburg.str.redhat.com>
Date: Tue, 24 Sep 2024 09:18:58 +0200
From: Florian Weimer <fweimer@...hat.com>
To: Schrodinger ZHU Yifan <i@...yi.fan>
Cc: "libc-coord@...ts.openwall.com" <libc-coord@...ts.openwall.com>
Subject: Re: getrandom via vDSO
* Schrodinger ZHU Yifan:
> 1 Is there a clear performance demand such that it becomes necessarily
> to avoid syscall inside getrandom​?
A pure kernelspace implementation of arc4random is unusable for many
applications due to system call overhead. If that's what libc offers,
some applications will have to use a bundled userspace implementation
instead.
Miroslav Lichvar reported a ~25% performance regression if glibc's
kernel-based arc4random is used:
| The new arc4random using getrandom() seems to have a significant
| impact on performance of chronyd operating as an NTP server. On an
| Intel E3-1220 CPU, I see that the maximum number of requests per
| second dropped by about 25%. That would be an issue for some public
| NTP servers.
<https://sourceware.org/bugzilla/show_bug.cgi?id=29437>
On newer CPUs, it's likely even worse.
Thanks,
Florian
Powered by blists - more mailing lists
Confused about mailing lists and their use?
Read about mailing lists on Wikipedia
and check out these
guidelines on proper formatting of your messages.
