[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <1412376969.17245.21.camel@edumazet-glaptop2.roam.corp.google.com>
Date: Fri, 03 Oct 2014 15:56:09 -0700
From: Eric Dumazet <eric.dumazet@...il.com>
To: Tom Herbert <therbert@...gle.com>
Cc: David Miller <davem@...emloft.net>,
Jesper Dangaard Brouer <brouer@...hat.com>,
Linux Netdev List <netdev@...r.kernel.org>,
Hannes Frederic Sowa <hannes@...essinduktion.org>,
Florian Westphal <fw@...len.de>,
Daniel Borkmann <dborkman@...hat.com>,
Jamal Hadi Salim <jhs@...atatu.com>,
Alexander Duyck <alexander.duyck@...il.com>,
John Fastabend <john.r.fastabend@...el.com>,
Dave Taht <dave.taht@...il.com>,
Toke Høiland-Jørgensen <toke@...e.dk>
Subject: Re: [net-next PATCH V6 0/2] qdisc: bulk dequeue support
On Fri, 2014-10-03 at 15:19 -0700, Tom Herbert wrote:
> On Fri, Oct 3, 2014 at 3:15 PM, Eric Dumazet <eric.dumazet@...il.com> wrote:
> > On Fri, 2014-10-03 at 14:57 -0700, Eric Dumazet wrote:
> >> On Fri, 2014-10-03 at 14:56 -0700, David Miller wrote:
> >>
> >> > I completely agree, and I sort of intended this to happen when
> >> > I split all the code into that new function.
> >> >
> >> > GSO segmentation of TX checksuming should not prevent other
> >> > > cpus from queueing other skbs in the qdisc.
> >> > >
> >> > > I will spend some time on this.
> >> >
> >> > Thanks!
> >>
> >> I just did my first reboot, will make sure everything is working well
> >> before sending the patch ;)
> >
> > This is awesome...
> >
> > 40Gb rate, with TSO=on or TSO=off, it does not matter anymore.
> >
> This is with or without GSO?
GSO=on
But with GSO off, I also get line rate, only spending more cpu.
25.05% [kernel] [k] copy_user_enhanced_fast_string
3.17% [kernel] [k] _raw_spin_lock
2.65% [kernel] [k] tcp_sendmsg
2.43% [kernel] [k] tcp_ack
2.08% [kernel] [k] __netif_receive_skb_core
1.88% [kernel] [k] menu_select
1.69% [kernel] [k] put_compound_page
1.50% [kernel] [k] flush_smp_call_function_queue
1.48% [kernel] [k] int_sqrt
1.43% [kernel] [k] call_function_single_interrupt
1.42% [kernel] [k] cpuidle_enter_state
1.21% perf [.] 0x00000000000353f2
1.10% [kernel] [k] tcp_init_tso_segs
1.05% [kernel] [k] memcpy
1.01% [kernel] [k] __skb_clone
0.91% [kernel] [k] irq_entries_start
0.84% [kernel] [k] get_nohz_timer_target
0.84% [kernel] [k] put_page
0.82% [kernel] [k] tcp_write_xmit
0.82% [kernel] [k] get_page_from_freelist
0.82% [kernel] [k] llist_reverse_order
0.80% [kernel] [k] native_sched_clock
0.79% [kernel] [k] cpu_startup_entry
0.78% [kernel] [k] kmem_cache_free
The killer combination is GSO=off and TX=off, of course.
9.90% [kernel] [k] csum_partial_copy_generic
7.64% [kernel] [k] _raw_spin_lock
6.10% [kernel] [k] tcp_ack
5.11% [kernel] [k] __skb_clone
3.90% [kernel] [k] __alloc_skb
3.78% [kernel] [k] skb_release_data
3.45% [kernel] [k] csum_partial
3.17% [kernel] [k] __kfree_skb
3.07% [kernel] [k] skb_clone
2.44% [kernel] [k] __kmalloc_node_track_caller
2.26% [kernel] [k] tcp_init_tso_segs
2.25% [kernel] [k] kfree
1.98% [kernel] [k] pfifo_fast_dequeue
1.61% [kernel] [k] tcp_set_skb_tso_segs
1.45% [kernel] [k] kmem_cache_free
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Powered by blists - more mailing lists