eBPF based tools for MTL#

This directory contains eBPF based tools for MTL.

1. Dependencies#

Dependencies: libbpf, bpftool, libxdp, clang, llvm, gcc, libelf, zlib.

If you has problems to install the dependencies for libbpf, bpftool, or libxdp, you might need to install these dependencies manually, follow libbpf page libbpf/libbpf, bpftool page libbpf/bpftool, libxdp page xdp-project/xdp-tools to build and install from source code directly.

2. Run#

2.1. lcore_monitor#

lcore_monitor: a eBPF based tool to monitor the scheduler and interrupt event on one MTL lcore. The MTL achieves high data packet throughput and low latency by defaulting to busy polling, also referred to as busy-waiting or spinning. To ensure peak performance, it is crucial to verify that no other tasks are running on the same logical core (lcore) that MTL utilizes. For debugging purposes, the tool provides a status overview via an eBPF (extended Berkeley Packet Filter) trace point hook, which monitors the activities on a single core.

Build:

make lcore_monitor

Run:

sudo ./lcore_monitor --lcore 30 --t_pid 194145

The output is like below, inspect the time to check if the lcore is suspending for a long time.

main, load bpf object lcore_monitor_kern.o succ
lm_event_handler: sched out 7.789us as comm: migration/30
lm_event_handler: sched out 7.405us as comm: migration/30

The lcore and t_pid can be get from MTL running log.

MT: MT: 2024-01-17 15:45:14, * *    M T    D E V   S T A T E   * *
MT: MT: 2024-01-17 15:45:14, DEV(0): Avr rate, tx: 2610.440314 Mb/s, rx: 0.000278 Mb/s, pkts, tx: 2465879, rx: 6
MT: MT: 2024-01-17 15:45:14, DEV(1): Avr rate, tx: 0.000000 Mb/s, rx: 2602.470600 Mb/s, pkts, tx: 0, rx: 2465811
MT: MT: 2024-01-17 15:45:14, SCH(0:sch_0): tasklets 3, lcore 29(t_pid: 190637), avg loop 105 ns
MT: MT: 2024-01-17 15:45:14, SCH(1:sch_1): tasklets 1, lcore 30(t_pid: 190638), avg loop 45 ns

2.2. udp_monitor#

udp_monitor: a eBPF based tool to monitor UDP streaming on the network, please note the promiscuous mode will be enabled on the selected interface. “promiscuous mode” means that a network interface card will pass all packets received up to the software for processing, versus the traditional mode of operation wherein only packets destined for the NIC’s MAC address are passed to software. Generally, promiscuous mode is used to “sniff” all traffic on the wire.