[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CAJfpegtn1A=dL9VZJQ2GRWsOiP+YSs-4ezE9YgEYNmb-AF0OLA@mail.gmail.com>
Date: Tue, 31 Mar 2020 21:42:48 +0200
From: Miklos Szeredi <miklos@...redi.hu>
To: David Howells <dhowells@...hat.com>
Cc: Linus Torvalds <torvalds@...ux-foundation.org>,
Al Viro <viro@...iv.linux.org.uk>, dray@...hat.com,
Karel Zak <kzak@...hat.com>,
Miklos Szeredi <mszeredi@...hat.com>,
Steven Whitehouse <swhiteho@...hat.com>,
Jeff Layton <jlayton@...hat.com>, Ian Kent <raven@...maw.net>,
andres@...razel.de,
Christian Brauner <christian.brauner@...ntu.com>,
Lennart Poettering <mzxreary@...inter.de>,
keyrings@...r.kernel.org, linux-fsdevel@...r.kernel.org,
linux-kernel@...r.kernel.org
Subject: Re: Upcoming: Notifications, FS notifications and fsinfo()
On Tue, Mar 31, 2020 at 7:31 PM David Howells <dhowells@...hat.com> wrote:
>
> Miklos Szeredi <miklos@...redi.hu> wrote:
>
> > The basic problem in my view, is that the performance requirement of a
> > "get filesystem information" type of API just does not warrant a
> > binary coded interface. I've said this a number of times, but it fell
> > on deaf ears.
>
> It hasn't so fallen, but don't necessarily agree with you. Let's pin some
> numbers on this.
Cool, thanks for testing. Unfortunately the test-fsinfo-perf.c file
didn't make it into the patch. Can you please refresh and resend?
> Okay, the results:
>
> For 1000 mounts, f= 1514us f2= 1102us p= 6014us p2= 6935us; p=4.0*f p=5.5*f2 p=0.9*p2
> For 2000 mounts, f= 4712us f2= 3675us p= 20937us p2= 22878us; p=4.4*f p=5.7*f2 p=0.9*p2
> For 3000 mounts, f= 6795us f2= 5304us p= 31080us p2= 34056us; p=4.6*f p=5.9*f2 p=0.9*p2
> For 4000 mounts, f= 9291us f2= 7434us p= 40723us p2= 46479us; p=4.4*f p=5.5*f2 p=0.9*p2
> For 5000 mounts, f=11423us f2= 9219us p= 50878us p2= 58857us; p=4.5*f p=5.5*f2 p=0.9*p2
> For 10000 mounts, f=22899us f2=18240us p=101054us p2=117273us; p=4.4*f p=5.5*f2 p=0.9*p2
> For 20000 mounts, f=45811us f2=37211us p=203640us p2=237377us; p=4.4*f p=5.5*f2 p=0.9*p2
> For 30000 mounts, f=69703us f2=54800us p=306778us p2=357629us; p=4.4*f p=5.6*f2 p=0.9*p2
So even the p2 method will give at least 80k queries/s, which is quite
good, considering that the need to rescan the complete mount tree
should be exceedingly rare (and in case it mattered, could be
optimized by priming from /proc/self/mountinfo).
Thanks,
Miklos
Powered by blists - more mailing lists