[dpdk-users] How to use more rx queues than RSS queue limit for ixgbe

[FQ Liu]

Hi

In production environment, all the packets have the same dst_ip, dst_port
and protocol, I cannot use vlan tag to identity in DCB mode.
If use Flow Director, one filter rule would direct to only one core,
Maybe I must modify the driver source code to meet the demand?

2016-08-16 21:35 GMT+08:00 Nishant Verma <vnish11 at gmail.com>:

> Hi,
>
> Yes, you can use flow director and other packet classification methods
> present, the thing is your hardware supports that. Check NIC datasheet if
> some thing you think is correct and still not working.
> For DCB, you can check example(vmdq_dcb) present in example folder in DPDK.
> For Flow Director, you can check test-pmd code.
>
> Just for your information. flow director can be used like this
> "
>     struct rte_eth_fdir_filter entry;
>
>     memset(&entry, 0, sizeof(struct rte_eth_fdir_filter));
>
>     entry.input.flow_type = 0x05;
>     entry.input.flow.udp4_flow.ip.dst_ip=0x11223344;
>     entry.input.flow.udp4_flow.dst_port=0x1111;
>     entry.action.rx_queue=0x02;
>     entry.action.report_status=0x02;
>
>     ret = rte_eth_dev_filter_ctrl(0, RTE_ETH_FILTER_FDIR,
>                          RTE_ETH_FILTER_ADD, &entry);
>
> "
> So, from now packet received from 0x11223344 will be re-directed to queue
> 2.
>
>
> On Tue, Aug 16, 2016 at 6:25 AM, FQ Liu <qleex001 at gmail.com> wrote:
>
>> Hi all
>>       I want to use 32 cores to process dns requests, which would need 32
>> rx queues, but the maximum RSS queue of the ethernet controller(Intel
>> Corporation 82599ES 10-Gigabit SFI/SFP+) is 16. If I can use dcb or flow
>> director to assign requests to 32 cores?
>>        The document (http://dpdk.org/doc/guides/nics/overview.html) shows
>> that ixgbe driver supports DCB and Flow director, but there isn't any
>> details about.
>>
>>
>>        OS: centos
>>        kernel: linux-3.18.34
>>        driver: igb_uio
>>
>
>
>
> --
> Rgds,
> Nishant
>
>
>
>