From: Andrew Morton <akpm@osdl.org>
To: Ravikiran G Thirumalai <kiran@scalex86.org>
Cc: dada1@cosmosbay.com, davem@davemloft.net,
linux-kernel@vger.kernel.org, shai@scalex86.org,
netdev@vger.kernel.org, pravins@calsoftinc.com
Subject: Re: [patch 3/4] net: Percpufy frequently used variables -- proto.sockets_allocated
Date: Fri, 27 Jan 2006 15:01:06 -0800 [thread overview]
Message-ID: <20060127150106.38b9e041.akpm@osdl.org> (raw)
In-Reply-To: <20060127224433.GB3565@localhost.localdomain>
Ravikiran G Thirumalai <kiran@scalex86.org> wrote:
>
> On Fri, Jan 27, 2006 at 12:16:02PM -0800, Andrew Morton wrote:
> > Ravikiran G Thirumalai <kiran@scalex86.org> wrote:
> > >
> > > which can be assumed as not frequent.
> > > At sk_stream_mem_schedule(), read_sockets_allocated() is invoked only
> > > certain conditions, under memory pressure -- on a large CPU count machine,
> > > you'd have large memory, and I don't think read_sockets_allocated would get
> > > called often. It did not atleast on our 8cpu/16G box. So this should be OK
> > > I think.
> >
> > That being said, the percpu_counters aren't a terribly successful concept
> > and probably do need a revisit due to the high inaccuracy at high CPU
> > counts. It might be better to do some generic version of vm_acct_memory()
> > instead.
>
> AFAICS vm_acct_memory is no better. The deviation on large cpu counts is the
> same as percpu_counters -- (NR_CPUS * NR_CPUS * 2) ...
I suppose so. Except vm_acct_memory() has
#define ACCT_THRESHOLD max(16, NR_CPUS * 2)
But if we were to perform similar tuning to percpu_counter, yes, they're
pretty similar.
Oh, and because vm_acct_memory() is counting a singleton object, it can use
DEFINE_PER_CPU rather than alloc_percpu(), so it saves on a bit of kmalloc
overhead.
> >
> > If the benchmarks say that we need to. If we cannot observe any problems
> > in testing of existing code and if we can't demonstrate any benefit from
> > the patched code then one option is to go off and do something else ;)
>
> We first tried plain per-CPU counters for memory_allocated, found that reads
> on memory_allocated was causing cacheline transfers, and then
> switched over to batching. So batching reads is useful. To avoid
> inaccuracy, we can maybe change percpu_counter_init to:
>
> void percpu_counter_init(struct percpu_counter *fbc, int maxdev)
>
> the percpu batching limit would then be maxdev/num_possible_cpus. One would
> use batching counters only when both reads and writes are frequent. With
> the above scheme, we would go fetch cachelines from other cpus for read
> often only on large cpu counts, which is not any worse than the global
> counter alternative, but it would still be beneficial on smaller machines,
> without sacrificing a pre-set deviation.
>
> Comments?
Sounds sane.
next prev parent reply other threads:[~2006-01-27 22:59 UTC|newest]
Thread overview: 39+ messages / expand[flat|nested] mbox.gz Atom feed top
2006-01-26 18:56 [patch 0/4] net: Percpufy frequently used variables on struct proto Ravikiran G Thirumalai
2006-01-26 18:59 ` [patch 1/4] net: Percpufy frequently used variables -- add percpu_counter_mod_bh Ravikiran G Thirumalai
2006-01-26 19:02 ` [patch 2/4] net: Percpufy frequently used variables -- struct proto.memory_allocated Ravikiran G Thirumalai
2006-01-27 9:01 ` Eric Dumazet
2006-01-26 19:03 ` [patch 3/4] net: Percpufy frequently used variables -- proto.sockets_allocated Ravikiran G Thirumalai
2006-01-27 8:53 ` Eric Dumazet
2006-01-27 19:52 ` Ravikiran G Thirumalai
2006-01-27 20:16 ` Andrew Morton
2006-01-27 22:30 ` Eric Dumazet
2006-01-27 22:50 ` Ravikiran G Thirumalai
2006-01-27 23:21 ` Eric Dumazet
2006-01-28 0:40 ` Ravikiran G Thirumalai
2006-01-27 22:44 ` Ravikiran G Thirumalai
2006-01-27 22:58 ` Eric Dumazet
2006-01-27 23:16 ` Andrew Morton
2006-01-28 0:28 ` Eric Dumazet
2006-01-28 0:35 ` Eric Dumazet
2006-01-28 4:52 ` Ravikiran G Thirumalai
2006-01-28 7:19 ` Eric Dumazet
2006-01-28 0:43 ` Andrew Morton
2006-01-28 1:10 ` Eric Dumazet
2006-01-28 1:18 ` Andrew Morton
2006-01-29 0:44 ` Benjamin LaHaise
2006-01-29 0:55 ` Andrew Morton
2006-01-29 1:19 ` Benjamin LaHaise
2006-01-29 1:29 ` Andrew Morton
2006-01-29 1:45 ` Kyle McMartin
2006-01-29 5:38 ` Andi Kleen
2006-01-29 6:54 ` Eric Dumazet
2006-01-29 19:52 ` Benjamin LaHaise
2006-01-27 23:01 ` Andrew Morton [this message]
2006-01-27 23:08 ` Andrew Morton
2006-01-28 0:01 ` Ravikiran G Thirumalai
2006-01-28 0:26 ` Andrew Morton
2006-02-03 3:05 ` Ravikiran G Thirumalai
2006-02-03 3:16 ` Andrew Morton
2006-02-03 19:37 ` Ravikiran G Thirumalai
2006-02-03 20:13 ` Andrew Morton
2006-01-26 19:05 ` [patch 4/4] net: Percpufy frequently used variables -- proto.inuse Ravikiran G Thirumalai
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20060127150106.38b9e041.akpm@osdl.org \
--to=akpm@osdl.org \
--cc=dada1@cosmosbay.com \
--cc=davem@davemloft.net \
--cc=kiran@scalex86.org \
--cc=linux-kernel@vger.kernel.org \
--cc=netdev@vger.kernel.org \
--cc=pravins@calsoftinc.com \
--cc=shai@scalex86.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).