hi, I have strange problems with HTB and several hundred classes flat structure i.e. root |--50kbps |--30kbps |--50kbps |--80kbps |--100kbps .... several hundred classes like this Ceil is the same as rate. The machine get no more than 2-3% average cpu(2.4Ghz pentium). What happens is that from time to time the traffic got "stalled". I tried numerous things to solve the problem - cable-wiring, different cards the problem still persist. My kernel is 2.4.20, as seen in the list I got the sch_htb.c from 2.4.22 kernel yestarday it seems that it behave better now but still have traffic starvation. I was on the devik page, but cant understand which patch is applied and which not..!?hope that 2.4.22 is the last..cant make many experiments cause it is working server. Nothing suspicious in the logs. Does someone have a setup with hundered of channels ! I spotted the problem 'cause a machine on the one side of the qos box does a snmp queries (mrtg) to the router on the other side and I see big holes into the graphics which mean that packets are droped for some reason.. then i found that there is a problem.. your ideas very appricated tia
Hello raptor, Friday, October 17, 2003, 1:41:46 PM, you wrote: I have before similar problem, for me it is looks problem - when traffic coming to default class (possible with small rate). I am doing simple solution, last with last-in-check (highest number) priority - filter with police drop. Try this, if this help on your case, maybe we need make bugreport about this. r> hi, r> I have strange problems with HTB and several hundred classes flat structure i.e. r> root r> |--50kbps r> |--30kbps r> |--50kbps r> |--80kbps r> |--100kbps r> .... several hundred classes like this r> Ceil is the same as rate. The machine get no more than 2-3% average cpu(2.4Ghz pentium). r> What happens is that from time to time the traffic got "stalled". r> I tried numerous things to solve the problem - cable-wiring, different cards the problem still persist. r> My kernel is 2.4.20, as seen in the list I got the sch_htb.c from 2.4.22 r> kernel yestarday it seems that it behave better now but still have traffic starvation. r> I was on the devik page, but cant understand which patch is applied and r> which not..!?hope that 2.4.22 is the last..cant make many experiments r> cause it is working server. r> Nothing suspicious in the logs. r> Does someone have a setup with hundered of channels ! r> I spotted the problem 'cause a machine on the one side of the qos box does a snmp queries (mrtg) to the router on the other side and I see big r> holes into the graphics which mean that packets are droped for some reason.. then i found that there is a problem.. r> your ideas very appricated r> tia r> _______________________________________________ r> LARTC mailing list / LARTC@mailman.ds9a.nl r> http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/ -- Best regards, Denis mailto:admin@ams-sat.com
U mean to place a class at the end with police to drop packets with highest priority ? Is that the way ?! I have such class but it is not with highest priority... Now I'm certain this is a problem with HTB... I have dumped the snmp traffic and what is seen is that as u know mrtg via crontab makes queries every five minutes... there is 6 queries ... the graphics show problems in the first three graphics and no problems at second third.. what happens is the following : time 0 sec snmp query for interface 1, (nothing is returned) snmp query for interface 1, (nothing is returned) snmp query for interface 1, (nothing is returned) snmp query for interface 1, (nothing is returned) snmp query for interface 1, (nothing is returned) timeout snmp query for interface 2, (nothing is returned) .... five times... timeout snmp query for interface 3, (nothing is returned) .... five times... timeout snmp query for interface 4, (response recieved) snmp query for interface 5, (response recieved) snmp query for interface 6, (response recieved) aprox time passed 27sec i have shaping only on downstream.. So this is ~30sec downstream traffic starvation atleast 5-10 times per hour... it is a real problem !! PLS HELP ME... | |Friday, October 17, 2003, 1:41:46 PM, you wrote: | |I have before similar problem, for me it is looks problem - when |traffic coming to default class (possible with small rate). | |I am doing simple solution, last with last-in-check (highest number) |priority - filter with police drop. Try this, if this help on your |case, maybe we need make bugreport about this. | | |r> hi, | |r> I have strange problems with HTB and several hundred classes flat structure i.e. | |r> root |r> |--50kbps |r> |--30kbps |r> |--50kbps |r> |--80kbps |r> |--100kbps |r> .... several hundred classes like this | |r> Ceil is the same as rate. The machine get no more than 2-3% average cpu(2.4Ghz pentium). |r> What happens is that from time to time the traffic got "stalled". |r> I tried numerous things to solve the problem - cable-wiring, different cards the problem still persist. |r> My kernel is 2.4.20, as seen in the list I got the sch_htb.c from 2.4.22 |r> kernel yestarday it seems that it behave better now but still have traffic starvation. |r> I was on the devik page, but cant understand which patch is applied and |r> which not..!?hope that 2.4.22 is the last..cant make many experiments |r> cause it is working server. |r> Nothing suspicious in the logs. | |r> Does someone have a setup with hundered of channels ! |r> I spotted the problem 'cause a machine on the one side of the qos box does a snmp queries (mrtg) to the router on the other side and I see big |r> holes into the graphics which mean that packets are droped for some reason.. then i found that there is a problem.. | |r> your ideas very appricated |r> tia
On Sun, 19 Oct 2003 00:23:16 +0300 raptor <raptor@tvskat.net> wrote: |U mean to place a class at the end with police to drop packets with highest priority ? Is that the way ?! |I have such class but it is not with highest priority... | oopsi I mean to do this : root |--50kbps |--30kbps |--50kbps |--80kbps |--100kbps .... several hundred classes like this |--- class(rate 10kbps, prio 7, default) --> drop Is that what u mean ?! tia