Hi everyone,
I have a questions about pktgen-dpdk.
That is,
why the Total Tx is not equal to Rx, and had no errors at the same time?
Like this:
The relevant environment is:
OS type: CentOS 6.6 x64
pktgen-dpdk version: 2.8.1
DPDK version: 2.0.0
NIC: 82599ES
Thanks !
Best regards,
Leo
Hello Leo,
Welcome back to the Intel Embedded Community.
We have notice that you are using these addresses: 192.168.0.1/24 and 192.168.1.1/24, according to these IPs you are using different subnets.
Is there any device in between? any router?
Please send a network diagram.
Regards,
Gabriel Thomas
Hi Gabriel,
Thanks for reply.
Isn't it just a default settings ?
I just run the command:
./app/pktgen -c 0x1f -n 3 -- -P -m "2.0, 3.1" -l pktgen.log
My diagram is like below
Best regards,
Leo
Hello Leo_Yeh,
Thanks for the update!
We have consulted this with our DPDK engineers and as soon as we have more information we'll get back with you.
Best regards,
Jimmy.
Hello Leo_Yeh
The Supported Operating Systems for the Data Plane Development Kit (DPDK) are the following:
Fedora release 18 Red Hat* Enterprise Linux 6.3
Wind River* Linux* 5 Ubuntu* 12.04 LTS
SUSE Enterprise Linux* 11 SP2 FreeBSD 9.2
Please refer to Supported NICs and the following DPDK documentation.
Best Regards.
Josue.
Hi all,
I have a question again.
why does the performance seems not good?
I mean, port 0/1 are fine.
However, port 2/3 and 4/5, their performance seems bad, maybe 50~60 % only. (MBits/s Rx/Tx)
My 'page cpu' is:
How can I tune the performance ?
Any ideas?
BTW, my OS is ubuntu 14.04
NIC is 82599.
Thanks a lot.
Best regards,
Leo,
Hello Leo,
We are working in your case, we will contact you with additional information soon.
Regards,
Gabriel Thomas
Hi all,
Update the status.
I changed my number of NICs to 12 ports.
And I found the performance like below:
It looks pretty good !
Is it that achieve the line rate ?
is it possible 'total rate' to achieve 60000/60000 or 59999/59999 ?
Thanks !
Leo,
Great.
That shows the performance you are observing is - it is not CPU bound.
Since it was I/o bound, the more you added, it scaled with I/o.
And with 128 byte packet size you should get what you are looking for.