[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <m1bpjjewkz.fsf@fess.ebiederm.org>
Date: Tue, 03 Nov 2009 02:23:08 -0800
From: ebiederm@...ssion.com (Eric W. Biederman)
To: Eric Dumazet <eric.dumazet@...il.com>
Cc: "David S. Miller" <davem@...emloft.net>, Greg KH <greg@...ah.com>,
Benjamin LaHaise <bcrl@...et.ca>,
Octavian Purdila <opurdila@...acom.com>,
netdev@...r.kernel.org, Cosmin Ratiu <cratiu@...acom.com>,
linux-kernel@...r.kernel.org
Subject: Re: [PATCH] sysctl: reduce ram usage by 40 %
Eric Dumazet <eric.dumazet@...il.com> writes:
> Eric Dumazet a écrit :
>
>> Its curious because in my tests the biggest problems come from
>> kernel/sysctl.c (__register_sysctl_paths) consuming 80% of cpu
>> in following attempt to create 20.000 devices
I bet that is Al's cute glue all the sysctl data structures together
patch. It improves readdir and lookup at a small cost at registration
time.
>> (disable hotplug before trying this, and ipv6 too !)
>> modprobe dummy numdummies=20000
>> I believe we should address __register_sysctl_paths() scalability
>> problems too.
Agreed.
>> I dont know what is the 'sentinel' we allocate after each struct ctl_table
>> But I suspect we could reduce size requirement of the 'sentinel' to include
>> only needed fields for the sentinel (and move them at start of ctl_table)
The sentinel is just a NULL terminator.
> Here is the patch to reduce ram usage of sysctl :
>
> [PATCH] sysctl: reduce ram usage by 40 %
>
> We currently reserve space for a so called sentinel, a full struct ctl_table
> for each ctl_table. We can cheat a bit since only needed fields of a sentinel
> are ctl_name and procname. Add a new structure (struct ctl_table_sentinel)
> that includes a full ctl_table and only required part of a sentinel.
Before we address sysctl I would like to get out my patchset that
makes sys_sysctl a wrapper around the ascii version of
/proc/sys/net. Once that goes in it becomes much easier to do things
and perform radical surgery on sysctl. Little things like .ctl_name and
.strategy go away.
Have you happened to look at the other cost of /proc proper? Hmm.
Except for /proc/net/dev_snmp6 it doesn't look like we keep per
interface directories in proc so without ivp6 you won't see the proc
generic code at all.
The practical consequence is if /proc/net/dev_snmp6 is not painful during
registration right now we can probably convert all of /proc/sys/net to proc
generic after my other changes are in.
Eric
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Powered by blists - more mailing lists