LinuxQuestions.org - IFB with HFSC queuing

- Linux - Networking (https://www.linuxquestions.org/questions/linux-networking-3/)

- - IFB with HFSC queuing - TX dropped packets (https://www.linuxquestions.org/questions/linux-networking-3/ifb-with-hfsc-queuing-tx-dropped-packets-4175576484/)

IFB with HFSC queuing - TX dropped packets

Hi
Last month on my ISP router I moved with queueing from IMQ to IFB (for incoming traffic on eth1 LAN interface). Everything seems to working fine however I noticed that when queueing is enabled on eth1 (LAN) interface which ifb0 is connected to there is a gradually increasing number of TX dropped packets. It looks like below. Is it normal behavior for queueing mechanisms for doing this or this is a problem of too small socket buffers? How to fix that?

Code:

root@access1:~# netstat -i

Kernel Interface table

Iface  MTU Met  RX-OK RX-ERR RX-DRP RX-OVR    TX-OK TX-ERR TX-DRP TX-OVR Flg

eth0      1500 0  1596046184      1  1992 0      1048035349      0      0      0 BMRU

eth1      1500 0  1112127683      6      0 0      1610512236      0      0      0 BMRU

eth0.6    1500 0  994109903      0      0 0      1036972784      0      0      0 BMRU

eth0.6:1  1500 0      - no statistics available -                        BMRU

eth0.6:2  1500 0      - no statistics available -                        BMRU

eth0.6:3  1500 0      - no statistics available -                        BMRU

eth0.6:4  1500 0      - no statistics available -                        BMRU

eth0.6:5  1500 0      - no statistics available -                        BMRU

eth1.8    1500 0    26630      0      0 0        25235      0      0      0 BMRU

eth1.10    1500 0  100159397      0      0 0      79595979      0  16306      0 BMRU

eth1.10:1  1500 0      - no statistics available -                        BMRU

eth1.10:2  1500 0      - no statistics available -                        BMRU

eth1.20    1500 0  97316370      0      0 0      93154573      0  21278      0 BMRU

eth1.20:1  1500 0      - no statistics available -                        BMRU

eth1.20:2  1500 0      - no statistics available -                        BMRU

eth1.100  1500 0  354601567      0      2 0      320990005      0  36936      0 BMRU

eth1.100:1  1500 0      - no statistics available -                        BMRU

eth1.180  1500 0  110145136      0      0 0      125649429      0  46503      0 BMRU

eth1.180:1  1500 0      - no statistics available -                        BMRU

eth1.190  1500 0  3000576      0      0 0      2994657      0      0      0 BMRU

eth1.200  1500 0  392698570      0      0 0      415288471      0  59740      0 BMRU

eth1.200:1  1500 0      - no statistics available -                        BMRU

ifb0      1500 0  1054757621      0      0 0      1054757621      0      0      0 BORU

lo        65536 0        2      0      0 0            2      0      0      0 LRU

IFB is for ingress traffic shaping. Do you have egress traffic shaping?

Yes. eth1 is my LAN interface and I queue download on eth1 and upload on ifb0 which is related to eth1.

You can try increase length of queue on eth1 and relative to VLAN.

What exactly do you mean by that? I've tried to increase interface queue by using ethtool without any result. Which option should I change?

How do you achieve egress traffic shaping?

Both eth1 and ifb0 have hfsc queueuing.

The burst parameter need be changed.

There is no such thing as burst in hfsc. Moreover it (the problem) doesn't appear while using IMQ instead of IFB so it's not the matter od queueuing mechanism.

Command, tc qdisc xxx, has the parameter.

No, burst is related to specific queueing mechanisms, not tc command http://linux.die.net/man/8/tc

Then you need adjust sc parameter of each class.

Quote:

Originally Posted by nini09 (Post 5530638)

Then you need adjust sc parameter of each class.

No, it't not the matter of hfsc, because it was moved from my previous system intact. I would rather change system buffers, but I like ask someone if it will help.