Hello.
I have a benchmark setup where I try to shape the traffic of
3000 concurrent tcp connection using HTB and hash tables.
The machine doing the shaping is running Debian 4.0 with Linux kernel
2.6.18 and has two 1 Gbps ethernet interfaces (in and out).
During the testing everything is going fine for couple of minutes but
after that kernel''s ring buffer starts to fill of
"htb: too many events !" messages and total throughput drops to 1/3 of
original speed.
The code of htb_do_events(struct htb_sched *q, int level) function in
net/sched/sch_htb.c don''t really comment it''s self at least
not for me :).
So in what kind of situations the for loop inside
htb_do_events(struct htb_sched *q, int level) function
does not return in 500 (magic number?) rounds and kernel warning
"htb: too many events !" will be printed?
Any ideas?
BR
--
Tuure Vartiainen
