VPP配置网卡多队列no bufs问题

在VPP配置文件/etc/vpp/startup.conf中启用网卡的多队列功能，指定接收和发送队列的数量。

dpdk {

   dev default {

      num-rx-queues 4

      num-tx-queues 4

   }

在测试中发现如下情况，网卡的rx no bufs错误计数非常高。

vpp # show hardware-interfaces

GigabitEthernet2/0/0                               1     up   GigabitEthernet2/0/0

  Link speed: 10 Gbps

  RX Queues:

    queue thread         mode

    0     vpp_wk_0 (1)   polling

    1     vpp_wk_1 (2)   polling

    2     vpp_wk_2 (3)   polling

    3     vpp_wk_3 (4)   polling

  Intel 82599

    carrier up full duplex mtu 1500

    flags: admin-up pmd maybe-multiseg tx-offload intel-phdr-cksum rx-ip4-cksum int-supported

    Devargs:

    rx: queues 4 (max 128), desc 1024 (min 32 max 4096 align 8)

    tx: queues 4 (max 64), desc 1024 (min 32 max 4096 align 8)

    tx frames ok                                        1228

    tx bytes ok                                        73998

    rx frames ok                                        2639

    rx bytes ok                                       158486

    rx missed                                     1106737634

    rx no bufs                                  379872039168

网上查询可以通过增加socket-mem或者num-mbufs的数量来解决此问题，但是VPP目前都不再支持这两个参数。查询startup.conf文件，发现可通过设置buffers-per-numa解决。其默认值为16384，修改为128000之后，问题解决。

    ## Increase number of buffers allocated, needed only in scenarios with

    ## large number of interfaces and worker threads. Value is per numa node.

    ## Default is 16384 (8192 if running unpriviledged)

buffers {

        buffers-per-numa 128000

        default data-size 2048

VPP启动之后可见，buffers的总数量为128016，可用的还有97560，使用了29845，默认值16384明显不够使用。

vpp# show buffers

Pool Name            Index NUMA  Size  Data Size  Total  Avail  Cached   Used

default-numa-0         0     0   2752     2048   128016  97560    611    29845

VPP文件src/vlib/buffer.c定义了默认的buffers数量。

#define VLIB_BUFFER_DEFAULT_BUFFERS_PER_NUMA 16384

#define VLIB_BUFFER_DEFAULT_BUFFERS_PER_NUMA_UNPRIV 8192