还用ovs+dpdk?

1 ovs+dpdk的性能仍然是非常差的，这里给出了一个数据比较

所以后续连包处理都要用VPP了马？

2 NDR rate : nondrop rate

3

VPP

https://docs.fd.io/vpp/17.04/d0/d83/vlib_2buffer_8h.html#a261854dad4fbc1792cd333a95b78c700

static void* vlib_buffer_get_current(vlib_buffer_t * b)  inlinestatic

Get pointer to current data to process.

Parameters b- (vlib_buffer_t *) pointer to the buffer Returns - (void *) (b->data + b->current_data)

Definition at line 188 of file buffer.h.

4

static vlib_buffer_t* vlib_get_buffer(vlib_main_t * vm,  u32 buffer_index  )   inlinestatic

Translate buffer index into buffer pointer.

Parameters vm- (vlib_main_t *) vlib main data structure pointerbuffer_index- (u32) buffer index Returns - ( vlib_buffer_t *) buffer pointer

Definition at line 57 of file buffer_funcs.h.

5

static void* vlib_frame_vector_args(vlib_frame_t * f)  inlinestatic

Get pointer to frame vector data.

Parameters fvlib_frame_t pointer Returns pointer to first vector element in frame

Definition at line 253 of file node_funcs.h.

6

#define VLIB_REGISTER_NODE( x,   ...  )  

Value: __VA_ARGS__ vlib_node_registration_t x; \ static void __vlib_add_node_registration_##x ( void) \ __attribute__((__constructor__)) ; \ static void __vlib_add_node_registration_##x ( void) \ { \ vlib_main_t * vm = vlib_get_main(); \ x.next_registration = vm->node_main.node_registrations; \ vm->node_main.node_registrations = &x; \ } \ __VA_ARGS__ vlib_node_registration_t x

Definition at line 143 of file node.h.

7

REF:

https://wiki.fd.io/view/VPP/What_is_VPP?