ok , thanks anyway.
mik...@yeah.net
From: Damjan Marion
Date: 2018-12-19 16:04
To: mik...@yeah.net
CC: vpp-dev
Subject: Re: [vpp-dev] dpdk-input : serious load imbalance
Yeah, I did not expect change, but at least we know that RSS is confiugred on
this card:
rss avail: ipv4 ipv4-tcp ipv4-udp ipv6 ipv6-tcp ipv6-udp port
vxlan geneve nvgre
rss active: ipv4 ipv4-tcp ipv4-udp ipv6 ipv6-tcp ipv6-udp
I'm affraid this is not VPP issue, you will need to talk with Cavium guys and
ask them why RSS doesn't work properly...
--
Damjan
On 19 Dec 2018, at 09:00, mik...@yeah.net wrote:
The result is the same. The 2rd queue only received 24 packets as it is in
18.07.
version : vpp v18.10-19~ga8e3001-dirty
vpp# show hardware-interfaces
Name Idx Link Hardware
VirtualFunctionEthernet5/0/2 1 up VirtualFunctionEthernet5/0/2
Ethernet address 12:90:b1:65:2d:19
Cavium ThunderX
carrier up full duplex speed 10000 mtu 9190
flags: admin-up pmd maybe-multiseg
rx: queues 2 (max 96), desc 1024 (min 0 max 65535 align 1)
tx: queues 2 (max 96), desc 1024 (min 0 max 65535 align 1)
pci: device 177d:a034 subsystem 177d:a234 address 0000:05:00.02 numa 0
module: unknown
max rx packet len: 9204
promiscuous: unicast off all-multicast off
vlan offload: strip off filter off qinq off
rx offload avail: vlan-strip ipv4-cksum udp-cksum tcp-cksum jumbo-frame
crc-strip scatter
rx offload active: jumbo-frame crc-strip scatter
tx offload avail: ipv4-cksum udp-cksum tcp-cksum outer-ipv4-cksum
tx offload active:
rss avail: ipv4 ipv4-tcp ipv4-udp ipv6 ipv6-tcp ipv6-udp port
vxlan geneve nvgre
rss active: ipv4 ipv4-tcp ipv4-udp ipv6 ipv6-tcp ipv6-udp
tx burst function: nicvf_xmit_pkts_multiseg
rx burst function: nicvf_recv_pkts_multiseg_no_offload
tx frames ok 11064
tx bytes ok 3042600
rx frames ok 3995658
rx bytes ok 687253292
rx missed 4342
extended stats:
rx good packets 3995658
tx good packets 11064
rx good bytes 687253292
tx good bytes 3042600
rx missed errors 4342
rx q0packets 3995634
rx q0bytes 687249164
rx q1packets 24
rx q1bytes 4128
tx q0packets 12
tx q0bytes 3300
tx q1packets 11052
tx q1bytes 3039300
mik...@yeah.net
From: Damjan Marion
Date: 2018-12-19 15:46
To: mik...@yeah.net
CC: vpp-dev
Subject: Re: [vpp-dev] dpdk-input : serious load imbalance
Can you try to capture "show hardw" with 18.10 ?
Looks like ThunderX is not acting as PCI device so part of the output is
suppressed in 18.07 and we changed that behaviour in 18.10.
I'm looking for someething like:
rss avail: ipv4 ipv4-tcp ipv4-udp ipv6 ipv6-tcp ipv6-udp ipv6-tcp-ex
ipv6-udp-ex ipv6-ex ipv6-tcp-ex ipv6-udp-ex
rss active: none
--
Damjan
On 19 Dec 2018, at 08:26, mik...@yeah.net wrote:
vpp v18.07.1-10~gc548f5d-dirty
mik...@yeah.net
From: Damjan Marion
Date: 2018-12-19 15:21
To: mik...@yeah.net
CC: vpp-dev
Subject: Re: [vpp-dev] dpdk-input : serious load imbalance
What version of VPP do you use.
I'm missing some outputs in "show hardware"...
--
Damjan
On 19 Dec 2018, at 02:19, mik...@yeah.net wrote:
The "show hardw" is as follow, the statistics may be different from yesterday.
vpp# show hardware-interfaces
Name Idx Link Hardware
VirtualFunctionEthernet5/0/2 1 up VirtualFunctionEthernet5/0/2
Ethernet address 72:62:8a:40:43:12
Cavium ThunderX
carrier up full duplex speed 10000 mtu 9190
flags: admin-up pmd maybe-multiseg
rx queues 2, rx desc 1024, tx queues 2, tx desc 1024
cpu socket 0
tx frames ok 268302
tx bytes ok 74319654
rx frames ok 4000000
rx bytes ok 688000000
extended stats:
rx good packets 4000000
tx good packets 268302
rx good bytes 688000000
tx good bytes 74319654
rx q0packets 3999976
rx q0bytes 687995872
rx q1packets 24
rx q1bytes 4128
tx q0packets 12
tx q0bytes 3324
tx q1packets 268290
tx q1bytes 74316330
VirtualFunctionEthernet5/0/3 2 down VirtualFunctionEthernet5/0/3
Ethernet address 2a:f2:d5:47:67:f1
Cavium ThunderX
carrier down
flags: pmd maybe-multiseg
rx queues 2, rx desc 1024, tx queues 2, tx desc 1024
cpu socket 0
local0 0 down local0
local
mik...@yeah.net
From: Damjan Marion
Date: 2018-12-18 20:38
To: mik...@yeah.net
CC: vpp-dev@lists.fd.io
Subject: Re: [vpp-dev] dpdk-input : serious load imbalance
What kind of nic do you have? Can you capture "show hardw" ?
--
Damjan
On 18 Dec 2018, at 04:03, mik...@yeah.net wrote:
Hi,
I configured 2 worker thread and 2 dpdk rx-queuein startup.conf . Then I
forged 400w packets and send them to a single dpdk interface . It turns out
that the second thread received only 24 packets .I test it for several times
and the results are almost the same. Why did this happen ?
Here is some config and "show":
VPP : 18.07
startup.conf:
cpu {
main-core 1
corelist-workers 2,3
}
dpdk {
dev default {
num-rx-queues 2
num-tx-queues 2
}
}
packets:
these pkts share the same src mac , dst mac and ipv4 body , only ipv4 src ip
and dst ip are different from each other.
<Catch.jpg>
------------------------------------------------------------------
# sh runtime
Thread 1 vpp_wk_0 (lcore 2)
Time 69.8, average vectors/node 1.02, last 128 main loops 0.00 per node 0.00
vector rates in 5.7911e4, out 5.0225e2, drop 5.7783e4, punt 0.0000e0
Name State Calls Vectors
Suspends Clocks Vectors/Call
dpdk-input polling 104246554 3999630
0 7.84e2 .04
---------------
Thread 2 vpp_wk_1 (lcore 3)
Time 69.8, average vectors/node 1.00, last 128 main loops 0.00 per node 0.00
vector rates in 5.1557e-1, out 1.7186e-1, drop 5.1557e-1, punt 0.0000e0
Name State Calls Vectors
Suspends Clocks Vectors/Call
dpdk-input polling 132390000 24
0 1.59e8 0.00
-----------------------------------------------------------------
# show interface rx-placement
Thread 1 (vpp_wk_0):
node dpdk-input:
VirtualFunctionEthernet5/0/2 queue 0 (polling)
VirtualFunctionEthernet5/0/3 queue 0 (polling)
Thread 2 (vpp_wk_1):
node dpdk-input:
VirtualFunctionEthernet5/0/2 queue 1 (polling)
VirtualFunctionEthernet5/0/3 queue 1 (polling)
-----------------------------------------------------------------
vpp# show interface
Name Idx State MTU (L3/IP4/IP6/MPLS)
Counter Count
VirtualFunctionEthernet5/0/2 1 up 9000/0/0/0 rx packets
3999654
rx bytes 687940488
tx packets 35082
tx bytes 9647550
rx-miss 346
VirtualFunctionEthernet5/0/3 2 down 9000/0/0/0
local0 0 down 0/0/0/0
Thanks in advance.
Mikado
mik...@yeah.net
-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#11675): https://lists.fd.io/g/vpp-dev/message/11675
Mute This Topic: https://lists.fd.io/mt/28791566/675642
Group Owner: vpp-dev+ow...@lists.fd.io
Unsubscribe: https://lists.fd.io/g/vpp-dev/unsub [dmar...@me.com]
-=-=-=-=-=-=-=-=-=-=-=-
-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#11716): https://lists.fd.io/g/vpp-dev/message/11716
Mute This Topic: https://lists.fd.io/mt/28791566/21656
Group Owner: vpp-dev+ow...@lists.fd.io
Unsubscribe: https://lists.fd.io/g/vpp-dev/unsub [arch...@mail-archive.com]
-=-=-=-=-=-=-=-=-=-=-=-
