[openib-general] ip over ib throughtput
Grant Grundler
iod00d at hp.com
Sat Jan 22 00:31:50 PST 2005
On Wed, Jan 12, 2005 at 05:34:06PM -0800, Grant Grundler wrote:
> RX2600 sending TCP_Stream packets to RX4640 via topspin 12port switch.
> Rx2600 has "Low Profile" (Cougarcub) and rx4640 has "Cougar" installed
> in "dual rope" slots.
>
> /opt/netperf/netperf -l 60 -H 10.0.1.81 -t TCP_STREAM -i 5,2 -I 99,5 -- -m 8192 -s 262144 -S 262144
...
> _spin_unlock_irqrestore is a clue we are spending time in
> interrupt handlers and that isn't getting measured.
David Mosberger gently reminded me he had added "-i" option
to monitor interrupt context as well.
Here is another run with "-i".
ionize:~# q-syscollect -i -k /opt/netperf/netperf -l 60 -H 10.0.1.81 -t TCP_STREAM -- -m 8192 -s 262144 -S 262144
TCP STREAM TEST to 10.0.1.81
Recv Send Send
Socket Socket Message Elapsed
Size Size Size Time Throughput
bytes bytes bytes secs. 10^6bits/sec
262142 262142 8192 60.00 1542.81
q-view .q/kernel-cpu0.info#0 | less
Flat profile of CPU_CYCLES in kernel-cpu0.hist#0:
Each histogram sample counts as 1.00034m seconds
% time self cumul calls self/call tot/call name
27.91 16.69 16.69 - - - default_idle
5.48 3.27 19.96 - - - ipt_do_table
4.28 2.56 22.52 - - - do_csum
3.81 2.28 24.80 - - - __copy_user
2.53 1.51 26.32 - - - local_bh_enable
2.52 1.51 27.82 - - - tcp_transmit_skb
2.51 1.50 29.33 - - - nf_iterate
2.33 1.39 30.72 - - - mthca_poll_cq
2.19 1.31 32.03 - - - tcp_sendmsg
1.73 1.03 33.06 - - - mthca_post_send
1.48 0.88 33.95 - - - ip_queue_xmit
1.33 0.80 34.74 - - - ip_conntrack_in
1.31 0.79 35.53 - - - mthca_eq_int
1.23 0.73 36.26 - - - _spin_lock
...
cpu1 was 99.87% of the time in default_idle.
hth,
grant
More information about the general
mailing list