On 26 August 2014 15:49, Maxim Uvarov <[email protected]> wrote:

> On 08/26/2014 09:52 AM, Venkatesh Vivekanandan wrote:
>
>> Maxim,
>>
>> Need not apply this patch to upstream. It was sent, so that someone else
>> can validate the multi-queue support in ODP-DPDK.
>>
>> Thanks,
>> Venky
>>
>
> Venky I'm fine with that patch. Do we need to pass 12 to command line
> argument? 12 is number of cpus and number of dpdk threads, what will be if
> number of cpus will change? Might be set


Currently we can't pass from command line, one has to make the change in
the code. Since this is just a sample(simple :-) ) patch, sent it with
hard-coded values and not sure of the behavior with 128. I tested it on 4
ports(2 in pktgen and 2 in odp-dpdk) and 16 cpu machine, so 8 cpu's and 8
rings per odp-dpdk interface. From what I have seen, it varies based on
dstip and it requires 12 different ip address to exercise all 8 cpu's of
one interface in my test setup. It is good to use 128.

up this to 128 to blast 128 different ips?
>
> Maxim.
>
>
>>
>> On 25 August 2014 20:14, <[email protected] <mailto:
>> [email protected]>> wrote:
>>
>>     From: Venkatesh Vivekanandan <[email protected]
>>     <mailto:[email protected]>>
>>
>>
>>     setup:
>>
>>     - connect two ixgbe/igb cards to x86 box which has atleast 4 cpus.
>>     - Lets say we have eth0 & eth1 in one card and eth2 & eth3 in
>>     other card
>>     - eth0 and eth2 are connected back to back using cross cable(1g) or
>>       fibre(10g)
>>     - Similarly connect eth1 and eth3
>>
>>     test:
>>
>>     - say eth2 and eth3 are given to dpdk on which odp-dpdk l2fwd app
>>     is running.
>>       cmd is ./example/l2fwd/odp_l2fwd -i 0,1 -m 0
>>     - ensure eth0 and eth1 are brought up and put in promiscuous mode.
>>     - start odp_generator in rx mode in eth1. cmd is
>>       ./example/generator/odp_generator -I eth1 -m r
>>     - start odp_generator in tx mode in eth0. cmd is
>>       ./example/generator/odp_generator -I eth0 --srcmac
>> a0:36:9f:13:89:08
>>       --dstmac a0:36:9f:13:89:c8 -m u --dstip 201.0.0.0
>>     - replace srcmac(eth0) and dstmac(eth2) with actual mac addresses
>>     on the port.
>>     - choose suitable dstip, so that packets are distributed to different
>>       cpus on the machine.
>>
>>     Signed-off-by: Venkatesh Vivekanandan
>>     <[email protected]
>>     <mailto:[email protected]>>
>>
>>     ---
>>      example/generator/odp_generator.c | 4 +++-
>>      1 file changed, 3 insertions(+), 1 deletion(-)
>>
>>     diff --git a/example/generator/odp_generator.c
>>     b/example/generator/odp_generator.c
>>     index b10372e..d202a8a 100644
>>     --- a/example/generator/odp_generator.c
>>     +++ b/example/generator/odp_generator.c
>>     @@ -171,7 +171,7 @@ static int scan_mac(char *in, odp_ethaddr_t *des)
>>      static void pack_udp_pkt(odp_buffer_t obuf)
>>      {
>>             char *buf;
>>     -       int max;
>>     +       int max, i = 0;
>>             odp_packet_t pkt;
>>             odp_ethhdr_t *eth;
>>             odp_ipv4hdr_t *ip;
>>     @@ -195,6 +195,8 @@ static void pack_udp_pkt(odp_buffer_t obuf)
>>             /* ip */
>>             odp_packet_set_l3_offset(pkt, ODP_ETHHDR_LEN);
>>             ip = (odp_ipv4hdr_t *)(buf + ODP_ETHHDR_LEN);
>>     +       i = ((i+1) % 12);
>>     +       args->appl.dstip = args->appl.dstip + i;
>>             ip->dst_addr = odp_cpu_to_be_32(args->appl.dstip);
>>             ip->src_addr = odp_cpu_to_be_32(args->appl.srcip);
>>             ip->ver_ihl = ODP_IPV4 << 4 | ODP_IPV4HDR_IHL_MIN;
>>     --
>>     1.8.1.2
>>
>>
>>
>>
>> _______________________________________________
>> lng-odp mailing list
>> [email protected]
>> http://lists.linaro.org/mailman/listinfo/lng-odp
>>
>
>
> _______________________________________________
> lng-odp mailing list
> [email protected]
> http://lists.linaro.org/mailman/listinfo/lng-odp
>
_______________________________________________
lng-odp mailing list
[email protected]
http://lists.linaro.org/mailman/listinfo/lng-odp

Reply via email to