[LU-16832] Alternative option of "tc" command for lnet

• Type: Improvement
• Resolution: Unresolved
• Priority: Minor
• Affects Version/s: Lustre 2.16.0
• Severity: 3
• Assignee: WC Triage
• Reporter: Shuichi Ihara

Description

"tc" command in linux helps to simulate network latency and bandwidth, it works for Lustre transparently, but it only works for ksocklnd.
"lctl net_delay_add" (added in LU-5435) which is good and an alternative option of "tc" command for latency simulation.
However, it needs some improvmenets.

• "<-l | --latency DELAY_LATENCY>" smaller value (less than 1 sec) should be configured. (msec or usec)
• currently, either -r or -i required, but it would also allow latency only.

eventually, it could do latency similar of tc command in lnet.

[root@c01 ~]# ping mds01-100g
PING mds01-100g (192.168.0.163) 56(84) bytes of data.
64 bytes from mds01-100g (192.168.0.163): icmp_seq=1 ttl=64 time=0.154 ms
64 bytes from mds01-100g (192.168.0.163): icmp_seq=2 ttl=64 time=0.159 ms
64 bytes from mds01-100g (192.168.0.163): icmp_seq=3 ttl=64 time=0.136 ms
^C
--- mds01-100g ping statistics ---
3 packets transmitted, 3 received, 0% packet loss, time 2052ms
rtt min/avg/max/mdev = 0.136/0.149/0.159/0.017 ms
[root@c01 ~]# tc qdisc add dev ens785np0 root netem delay 500ms
[root@c01 ~]# ping mds01-100g
PING mds01-100g (192.168.0.163) 56(84) bytes of data.
64 bytes from mds01-100g (192.168.0.163): icmp_seq=1 ttl=64 time=500 ms
64 bytes from mds01-100g (192.168.0.163): icmp_seq=2 ttl=64 time=500 ms
64 bytes from mds01-100g (192.168.0.163): icmp_seq=3 ttl=64 time=500 ms

Issue Links

• is related to: LU-14627 Lost ref on lnet_peer in discovery leads to LNetError: 24909:0:(peer.c:292:lnet_destroy_peer_locked()) ASSERTION( list_empty(&lp->lp_peer_nets) ) failed:
• is related to: LU-15957 LNet: add latency stats to lnet_selftest output
• is related to: LU-5435 Simulate message loss and high latency in LNet