Message-Id: <200711140927.39796.nickpiggin@yahoo.com.au>
Date:	Wed, 14 Nov 2007 09:27:39 +1100
From:	Nick Piggin <nickpiggin@...oo.com.au>
To:	David Miller <davem@...emloft.net>
Cc:	clameter@....com, netdev@...r.kernel.org,
	herbert@...dor.apana.org.au, linux-kernel@...r.kernel.org
Subject: Re: 2.6.24-rc2: Network commit causes SLUB performance regression with tbench

On Wednesday 14 November 2007 17:37, David Miller wrote:
> From: Nick Piggin <nickpiggin@...oo.com.au>

> > I'm doing some oprofile runs now to see if I can get any more info.

OK, in vanilla kernels, the page allocator definitely shows higher
in the results (than with Herbert's patch reverted).

27516     2.7217  get_page_from_freelist                                        
21677     2.1442  __rmqueue_smallest                                            
20513     2.0290  __free_pages_ok                                               
18725     1.8522  get_pageblock_flags_group             

Just these account for nearly 10% of cycles. __alloc_skb shows up
higher too. free_hot_cold_page() shows a lot lower though, which
might indicate that actually there is more higher order allocation
activity (I'll check that next).

****
SLUB, avg throughput 1548

CPU: AMD64 family10, speed 1900 MHz (estimated)
Counted CPU_CLK_UNHALTED events (Cycles outside of halt state) with a unit 
mask of 0x00 (No unit mask) count 100000
samples  %        symbol name
94636     9.3609  copy_user_generic_string
38932     3.8509  ipt_do_table
34746     3.4369  tcp_v4_rcv
29539     2.9218  skb_release_data
27516     2.7217  get_page_from_freelist
26046     2.5763  tcp_sendmsg
24482     2.4216  local_bh_enable
22910     2.2661  ip_queue_xmit
22113     2.1873  ktime_get
21677     2.1442  __rmqueue_smallest
20513     2.0290  __free_pages_ok
18725     1.8522  get_pageblock_flags_group
18580     1.8378  tcp_recvmsg
18108     1.7911  __napi_schedule
17593     1.7402  schedule
16998     1.6813  tcp_ack
16102     1.5927  dev_hard_start_xmit
15751     1.5580  system_call
15707     1.5536  net_rx_action
15150     1.4986  __switch_to
14988     1.4825  tcp_transmit_skb
13921     1.3770  kmem_cache_free
13398     1.3253  __mod_timer
13243     1.3099  tcp_rcv_established
13109     1.2967  __tcp_select_window
11022     1.0902  __tcp_push_pending_frames
10732     1.0615  set_normalized_timespec
10561     1.0446  netif_rx
8840      0.8744  netif_receive_skb
7816      0.7731  nf_iterate
7300      0.7221  __update_rq_clock
6683      0.6610  _read_lock_bh
6504      0.6433  sys_recvfrom
6283      0.6215  nf_hook_slow
6188      0.6121  release_sock
6172      0.6105  loopback_xmit
5927      0.5863  __alloc_skb
5707      0.5645  tcp_cleanup_rbuf
5538      0.5478  tcp_event_data_recv
5517      0.5457  tcp_v4_do_rcv
5516      0.5456  process_backlog
****
SLUB, net patch reverted. Avg throughput 1933MB/s
Counted CPU_CLK_UNHALTED , count 100000
samples  %        symbol name
95895     9.5094  copy_user_generic_string
50259     4.9839  ipt_do_table
39408     3.9079  skb_release_data
37296     3.6984  tcp_v4_rcv
31309     3.1047  ip_queue_xmit
31308     3.1046  local_bh_enable
24052     2.3851  net_rx_action
23786     2.3587  __napi_schedule
21426     2.1247  tcp_recvmsg
21075     2.0899  schedule
20938     2.0763  dev_hard_start_xmit
20222     2.0053  tcp_sendmsg
19775     1.9610  tcp_ack
19717     1.9552  system_call
19495     1.9332  set_normalized_timespec
18584     1.8429  __switch_to
17022     1.6880  tcp_rcv_established
14655     1.4533  tcp_transmit_skb
14466     1.4345  __mod_timer
13820     1.3705  loopback_xmit
13776     1.3661  get_page_from_freelist
13288     1.3177  netif_receive_skb
9718      0.9637  _read_lock_bh
9625      0.9545  nf_iterate
9440      0.9361  netif_rx
9148      0.9072  free_hot_cold_page
8633      0.8561  __update_rq_clock
7668      0.7604  sys_recvfrom
7578      0.7515  __tcp_push_pending_frames
7311      0.7250  find_pid_ns
7178      0.7118  nf_hook_slow
6655      0.6599  sysret_check
6313      0.6260  release_sock
6290      0.6237  tcp_cleanup_rbuf
6263      0.6211  __tcp_select_window
6235      0.6183  process_backlog
5920      0.5871  ip_local_deliver_finish
5651      0.5604  ip_rcv
5239      0.5195  ip_finish_output
5058      0.5016  kmem_cache_free
5016      0.4974  thread_return


> Here are some other things you can play around with:
>
> 1) Monitor the values of skb->len and skb->data_len for packets
>    going over loopback.

OK, I've taken a log2 histogram of ->len and ->data_len sizes.
I'll attach the plots (xaxis is byte value of most significant
bit set, y axis is number of occurances, logscale on X so 0 isn't
there :( ). If you want to see the patch or raw data, let me know.

len looks very similar for both kernels, although the reverted
kernel has significantly higher frequency at most points. tbench
unfortunately is only possible to do it time-based, so this is
rougly expected.

data_len has a spike that shifts up to 512-1023, from 256-511,
when reverting Herbert's patch. Again, I believe the magnitudes
of the spikes should actually be pretty close if we were doing
an equal amount of work.

I can't see that these numbers show much useful, unfortunately.


> 2) Try removing NETIF_F_SG in drivers/net/loopback.c's dev->feastures
>    setting.

Will try that now.

Download attachment "data.png" of type "image/png" (3622 bytes)

Download attachment "len.png" of type "image/png" (4828 bytes)

Powered by blists - more mailing lists