Hi Matias,
With your configure, I reproduced the hash issue.
Looks like the hash is too small for these huge flows.
The hash bucket number is roughly 1.25K. For 132M flows, the average
bucket has the length of 100+K.
So the lookup performance is deteriorated.
thanks,
Forrest
在
Hi Maxim,
I CCed lng-odp@lists.linaro.org, but not show on
http://patches.opendataplane.org/project/lng-odp/list/
I will resend it.
thanks,
Forrest
在 8/24/2016 6:03 PM, Maxim Uvarov 写道:
was v8 send to mailing list?
I do not see it here:
On 2016-08-09 15:46, Maxim Uvarov wrote:
> On 08/06/16 22:30, Anders Roxell wrote:
> >--- a/test/linux-generic/m4/configure.m4
> >>+++ b/test/linux-generic/m4/configure.m4
> >>@@ -1,5 +1,6 @@
> >> AC_CONFIG_FILES([test/linux-generic/Makefile
> >>
On 2016-08-08 09:59, Maxim Uvarov wrote:
> Hello Anders,
>
> thanks for review, there are 2 comments bellow. Will send v2.
>
> Maxim.
>
> On 08/06/16 22:30, Anders Roxell wrote:
> >On 2016-08-02 19:08, Maxim Uvarov wrote:
> >>add pcap play back test which takes 2 arguments: 1 - pcap file,
> >>2
On Tue, Aug 23, 2016 at 4:51 PM, Barry Spinney wrote:
> Both odp_pkt_queue.c and odp_traffic_mngr.c has similar bugs where
> an array was malloc'd but the valid indicies run from 1 to the number
> of entries malloc'd. This is because queue_num == 0 is used as a special
>
On Wed, Aug 24, 2016 at 11:10 AM, Oriol Arcas
wrote:
> Regarding the last question pointed out by Bill, max_queue_num =
> tm_system->next_queue_num, which is always "current queues in the TM + 1".
>
> So the loop, as it is now, visits all the elements in the array.
>
Regarding the last question pointed out by Bill, max_queue_num =
tm_system->next_queue_num, which is always "current queues in the TM + 1".
So the loop, as it is now, visits all the elements in the array.
--
Oriol Arcas
Software Engineer
Starflow Networks
On Wed, Aug 24, 2016 at 2:11 PM, Bill
On Tue, Aug 23, 2016 at 4:51 PM, Barry Spinney wrote:
> Both odp_pkt_queue.c and odp_traffic_mngr.c has similar bugs where
> an array was malloc'd but the valid indicies run from 1 to the number
> of entries malloc'd. This is because queue_num == 0 is used as a special
>
Hi Forrest,
Here’s my test setup:
$ sudo ./odp_l3fwd -i 0,1 -r 1.1.1.0/24,0 -r 2.2.2.0/24,1 -s hash -t 1
Both interfaces are connected to 40Gbps NIC ports (i40e).
Originally the generator was creating ~132M random flows (2 x (254 dst IPs, 254
src IPs, UDP, 1024 src ports, 1 dst port)).
In
> was v8 send to mailing list?
Looks like it wasn't. I received the patch directly from Forrest.
-Matias
>
> I do not see it here:
> http://patches.opendataplane.org/project/lng-odp/list/
>
>
> On 08/23/16 19:09, Mike Holmes wrote:
> > On 23 August 2016 at 09:28, Elo, Matias (Nokia -
was v8 send to mailing list?
I do not see it here:
http://patches.opendataplane.org/project/lng-odp/list/
On 08/23/16 19:09, Mike Holmes wrote:
On 23 August 2016 at 09:28, Elo, Matias (Nokia - FI/Espoo) <
matias@nokia-bell-labs.com> wrote:
Would you please have a review on this version?
Hi Matias,
About the hash lookup performance issue, would you please let me know
how many flowes in the testing?
I mean 5 tuples (dest_ip, src_ip, dest_port, src_port, proto) sent by HW
packet generator for each port.
I'm trying to reproduce the issue.
Thanks.
Forrest
在 8/24/2016
https://bugs.linaro.org/show_bug.cgi?id=2449
--- Comment #3 from Oriol Arcas ---
I think the patch from Barry fixes this issue.
http://patches.opendataplane.org/patch/6895/
--
You are receiving this mail because:
You are on the CC list for the bug.
You are the
13 matches
Mail list logo