Hi all,
I did an experiment on interface capacity in vpp-agent and the results are as
follows:
1. af_packet: Less than 240 (when host memory is 8g)
created by:
ip link add name veth_lan$i type veth peer name lan$i
ip link set dev lan$i up
ip link set dev veth_lan$i up
vppctl create host-interface name lan$i
If I only creat ip dev without vpp host-interface, the number is far more than
1000, why does the af_packet interface take up so much memory? Is there any way
to create more af_packet interfaces?
2. memif: Less than 255
created by:
vppctl create interface memif id $i
The limiting factor for memif is node_index in src/vlib/error.h :
always_inline vlib_error_t
vlib_error_set (u32 node_index, u32 code)
{
ASSERT (node_index < (1 << 10));
ASSERT (code < (1 << 6));
return (node_index << 6) | code;
}
If I change node_index of error_set back to 1<<20, the number of memif
interfaces can reach 3000+. If so, What other effects?
3. tapV2: Less than 1024
created by:
vppctl create tap id $i
The limiting factor for tapV2 is TAP_MAX_INSTANCE 1024 in
src/vnet/devices/tap/tap.c.
Why set TAP_MAX_INSTANCE to 1024?If it is changed to 2048, what other effects?
--
姚洁
烽火通信科技股份有限公司(Fiberhome Telecommunication Technologies Co.,LTD)
宽带业务产出线（Broadband and Service Application Business Unit）
研发部（Research and Development Department）
Mobile：+86-18271867215
Email: yaoj5...@fiberhome.com
Address: 湖北省武汉市东湖新技术开发区高新四路6号
(No.6, Gaoxinsilu,East Lake High-tech Development zone，Wuhan, Hubei)
Postcode: 430205
http://www.fiberhome.com-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#10592): https://lists.fd.io/g/vpp-dev/message/10592
Mute This Topic: https://lists.fd.io/mt/25838930/21656
Group Owner: vpp-dev+ow...@lists.fd.io
Unsubscribe: https://lists.fd.io/g/vpp-dev/unsub [arch...@mail-archive.com]
-=-=-=-=-=-=-=-=-=-=-=-
