[LARTC] /proc/net/dev counters

[Piotr Chytla]

Hi 

Maybe my problem is litle offtopic to this list , but maybe someone had
something similar like this , and have some good solution .

Ok, I've have router with four intel e1000 pci-x(2x100Mhz/2x133Mhz) nics that 
push 
about 200Mbit/s , 
and I'm using  nload for realtime traffic monitoring. Everything was great 
until I've updated kernel to 2.6.17.13 . After update nload is showing 
some crazy values .

I've tried to get counters from /proc/net/dev (like nload). 

[EMAIL PROTECTED]:~# while true; do cat /proc/net/dev|grep eth0; sleep 1; done
  eth0:1013729758 5729322500 18821020 0  0 0 
1572910106 3860638290000 0   0  0
  eth0:1055515817 5730043720 18821020 0  0 0 
1601291109 3860694606000 0   0  0 
  eth0:1055515817 5730043720 18821020 0  0 0 
1601291109 3860694606000 0   0  0
  eth0:1097729573 5730764320 18821020 0  0 0 
1629536436 3860751311000 0   0  0
  eth0:1097729573 5730764320 18821020 0  0 0 
1629536436 3860751311000 0   0  0
  eth0:1139487258 5731484690 18821020 0  0 0 
1658034498 3860807633000 0   0  0
  eth0:1139487258 5731484690 18821020 0  0 0 
1658034498 3860807633000 0   0  0
  eth0:1181148113 5732200760 18821020 0  0 0 
1685931047 3860863738000 0   0  0

Between second and third second counter increase of 0 bytes, 

1055515817-1055515817=0

the same for fourth and fifth ( on mrtg graphs I've about 140Mbit/s incoming
traffic - 17.5 MB/s ).

On other router I've kernel 2.6.14.7 , e1000/2xe100 32bit nics , and
interfece counters in /proc/net/dev are looking good:

  eth0:2101453114 3324697664 3604 3441 344192  0   1347064 
867323793 3458783738000 0   0  0
  eth0:2107451373 3324705778 3604 3441 344192  0   1347066 
874688831 3458792480000 0   0  0
  eth0:2113352907 3324713495 3604 3441 344192  0   1347066 
881546913 3458800862000 0   0  0
  eth0:2119707847 3324721929 3604 3441 344192  0   1347069 
889060588 3458809975000 0   0  0
  eth0:2125601276 3324729816 3604 3441 344192  0   1347070 
896165072 3458818569000 0   0  0

2113352907-2107451373=5901534 (traffic on intefece about 40Mbit/s = 5MB/s) - 
seems good.

Both routers have CONFIG_HZ=1000, two cpus, and bofh are using
e1000/e100 driver with enabled NAPI. 

Besides I've similar thing on laptop, 2.6.14 , Broadcom BCM4401-B0 nic (b44) :)

Why interface counters are increasing so slowy? this behavior can be
changed? or in other way this is normal behavior or some bug , feature?



/pch

-- 
Dyslexia bug unpatched since 1977 ...
exploit has been leaked to the underground.
___
LARTC mailing list
LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc

RE: [LARTC] Problem with two providers:Need to route packets on the interface on which they arrives.

[Oscar Mechanic]

Hi

Yep, I understand so looking at this. What I describe below
is the simplest way I can think of doing it. Else you should
look at connmark or even the wonderful ebtables.

modem A IP 192.168.0.254 next hop 192.168.0.253
modem B IP 172.16.30.254 next hop 172.16.30.254

Server address is 10.0.0.12 and 10.0.0.13

if packet comes from MODEM A DNAT to server address 10.0.0.12
if packet comes from MODEM B DNAT to server address 10.0.0.13

Req from modem A responce ACK from sever 10.0.0.12 comes back and it
chooses default route modem A 

Req from modem B responce ACK from server 10.0.0.13 comes back and it 
chooses route from table 120

Cause you have added rule
ip rule add from 10.0.0.13 lookup 120
ip route add default via 172.16.30.254 table 120
-or-
ip route add from 10.0.0.13 via 172.16.30.254 (but I like tables so I
suggest above)

Some may look at this as a bad solution as it creates hidden solution
specific info.

Now if you want to make this a better solution look at connmark and
mark. And ip rule add fwmark XX lookup X. I know this would work but 
you will have to figure it out.

Also you have ipt_ROUTE but I have little success with this.


 
On Sat, 2006-10-14 at 19:21 +0200, KORTA wrote:
> I made a picture
> If a packet arrives from provider B, I want that the packet will be routed
> by the same provider even if the default route is provider A.
> 
> You gave me the syntax to route from source address,
> I just want to know if there is possible to route packets depends on which
> interface they arrives.
> Tx
> 
> 
> -Message d'origine-
> De : Oscar Mechanic [mailto:[EMAIL PROTECTED] 
> Envoyé : samedi 14 octobre 2006 13:56
> À : KORTA
> Cc : lartc@mailman.ds9a.nl
> Objet : Re: [LARTC] Problem with two providers:Need to route packets on the
> interface on which they arrives.
> 
> There is a simple way a hard way to do this. You could use
> connmark in iptables. And then use ip rule & routes to set
> route based on that. As I am not going to replicate this to test
> I wont try and guess commands.
> 
> Easiest configure 2 IP's on server.
> 
> DNAT like 
> iptables -t nat -I PREROUTING 1 -i ethA -j DNAT --to-destination
> <10.0.0.A> 
> iptables -t nat -I PREROUTING 1 -i ethB -j DNAT --to-destination
> <10.0.0.B>
> 
> ip rule add from 10.0.0.B lookup 120
> ip route add default via  table 120
> 
> Thats the easiest I can think of.
> 
> 
> 
> On Sat, 2006-10-14 at 13:10 +0200, KORTA wrote:
> > Hello,
> > 
> >  
> > 
> > i would like to know how to resolve a problem.
> > 
> >  
> > 
> > I have a debian router with 3 interfaces (LAN, and two internet
> > providers (Provider A, Provider B)).
> > 
> >  
> > 
> > The default route is configured to use the provider A
> > 
> >  
> > 
> > The problem is that,
> > 
> > When an external connection arrives from provider B to an internal
> > server (with nat),  the packet is routed to the default route:
> > 
> > I explain:
> > 
> > - A packet arrives from provider B in direction of a internal
> > server
> > 
> > - The router performs nat operation
> > 
> > - The internal server generates a response
> > 
> > - The router routes the packet on the interface Provider A
> > 
> >  
> > 
> > Consequently, The connection cannot been established 
> > 
> >  
> > 
> > I want to know if its possible to configure my debian router to route
> > packets to the interface on which packet arrives. In the example,
> > packets should have been routed by the interface connected to provider
> > B
> > 
> >  
> > 
> > If yes, do you know how to do that ?
> > 
> >  
> > 
> > Thanks.
> > 
> >  
> > 
> >  
> > 
> > 
> > ___
> > LARTC mailing list
> > [EMAIL PROTECTED]://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc

___
LARTC mailing list
LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc

RE: [LARTC] Problem with two providers:Need to route packets on the interface on which they arrives.

[KORTA]

I made a picture
If a packet arrives from provider B, I want that the packet will be routed
by the same provider even if the default route is provider A.

You gave me the syntax to route from source address,
I just want to know if there is possible to route packets depends on which
interface they arrives.
Tx


-Message d'origine-
De : Oscar Mechanic [mailto:[EMAIL PROTECTED] 
Envoyé : samedi 14 octobre 2006 13:56
À : KORTA
Cc : lartc@mailman.ds9a.nl
Objet : Re: [LARTC] Problem with two providers:Need to route packets on the
interface on which they arrives.

There is a simple way a hard way to do this. You could use
connmark in iptables. And then use ip rule & routes to set
route based on that. As I am not going to replicate this to test
I wont try and guess commands.

Easiest configure 2 IP's on server.

DNAT like 
iptables -t nat -I PREROUTING 1 -i ethA -j DNAT --to-destination
<10.0.0.A> 
iptables -t nat -I PREROUTING 1 -i ethB -j DNAT --to-destination
<10.0.0.B>

ip rule add from 10.0.0.B lookup 120
ip route add default via  table 120

Thats the easiest I can think of.



On Sat, 2006-10-14 at 13:10 +0200, KORTA wrote:
> Hello,
> 
>  
> 
> i would like to know how to resolve a problem.
> 
>  
> 
> I have a debian router with 3 interfaces (LAN, and two internet
> providers (Provider A, Provider B)).
> 
>  
> 
> The default route is configured to use the provider A
> 
>  
> 
> The problem is that,
> 
> When an external connection arrives from provider B to an internal
> server (with nat),  the packet is routed to the default route:
> 
> I explain:
> 
> - A packet arrives from provider B in direction of a internal
> server
> 
> - The router performs nat operation
> 
> - The internal server generates a response
> 
> - The router routes the packet on the interface Provider A
> 
>  
> 
> Consequently, The connection cannot been established 
> 
>  
> 
> I want to know if its possible to configure my debian router to route
> packets to the interface on which packet arrives. In the example,
> packets should have been routed by the interface connected to provider
> B
> 
>  
> 
> If yes, do you know how to do that ?
> 
>  
> 
> Thanks.
> 
>  
> 
>  
> 
> 
> ___
> LARTC mailing list
> [EMAIL PROTECTED]://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc
<>
___
LARTC mailing list
LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc

Re: [LARTC] Ethernet packet loss - frame errors

[Justin Schoeman]

Stephen Hemminger wrote:

On Fri, 13 Oct 2006 10:26:14 +0200
Justin Schoeman <[EMAIL PROTECTED]> wrote:


Hi all,

I have the following problem. A Linux box configured as a bridge. One 
interface connects to the router via a crossover cable, the other 
connects to a switch via the cable that used to go to the router.


Now I get the following:

[EMAIL PROTECTED] net]# ifconfig eth3
eth3  Link encap:Ethernet  HWaddr 00:03:2D:07:61:5D
   UP BROADCAST MULTICAST  MTU:1500  Metric:1
   RX packets:1969134 errors:0 dropped:0 overruns:0 frame:176459
   TX packets:2186662 errors:0 dropped:0 overruns:0 carrier:0
   collisions:0 txqueuelen:1000
   RX bytes:492595115 (469.7 Mb)  TX bytes:579725462 (552.8 Mb)
   Interrupt:17

eth3 is the interface that connects to the switch.  As you can see, 1 in 
10 rx packets are framing errors.


What are possible causes for this? The cable is a constant, so is not 
likely to cause problems, so what else can possibly cause such a high 
number of frame errors?


Any help appreciated.

Thanks,

Justin



What speed and driver?


sky 2 driver, hardcoded for 100Mbps full duplex.

-justin
___
LARTC mailing list
LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc

Re: [LARTC] Problem with two providers:Need to route packets on the interface on which they arrives.

[Oscar Mechanic]

There is a simple way a hard way to do this. You could use
connmark in iptables. And then use ip rule & routes to set
route based on that. As I am not going to replicate this to test
I wont try and guess commands.

Easiest configure 2 IP's on server.

DNAT like 
iptables -t nat -I PREROUTING 1 -i ethA -j DNAT --to-destination
<10.0.0.A> 
iptables -t nat -I PREROUTING 1 -i ethB -j DNAT --to-destination
<10.0.0.B>

ip rule add from 10.0.0.B lookup 120
ip route add default via  table 120

Thats the easiest I can think of.



On Sat, 2006-10-14 at 13:10 +0200, KORTA wrote:
> Hello,
> 
>  
> 
> i would like to know how to resolve a problem.
> 
>  
> 
> I have a debian router with 3 interfaces (LAN, and two internet
> providers (Provider A, Provider B)).
> 
>  
> 
> The default route is configured to use the provider A
> 
>  
> 
> The problem is that,
> 
> When an external connection arrives from provider B to an internal
> server (with nat),  the packet is routed to the default route:
> 
> I explain:
> 
> - A packet arrives from provider B in direction of a internal
> server
> 
> - The router performs nat operation
> 
> - The internal server generates a response
> 
> - The router routes the packet on the interface Provider A
> 
>  
> 
> Consequently, The connection cannot been established 
> 
>  
> 
> I want to know if its possible to configure my debian router to route
> packets to the interface on which packet arrives. In the example,
> packets should have been routed by the interface connected to provider
> B
> 
>  
> 
> If yes, do you know how to do that ?
> 
>  
> 
> Thanks.
> 
>  
> 
>  
> 
> 
> ___
> LARTC mailing list
> [EMAIL PROTECTED]://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc

___
LARTC mailing list
LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc

Re: [LARTC] Intel or AMD is better processor for router (800+ users)

[Marek Kierdelewicz]

> The author highly recommends disabling IRQ balancing in the kernel 
> config, but does not clarify what this does. I tried googling but
> didn't find much info. What does it exactly do, and why is disabling
> it recommended/required?

I haven't seen irq balance option in kernels for some time now. 

> It seems to me, IRQ balancing does not allow to specify interrupts
> per device but assigns them automatically on the run, correct?

It seems so. As I mentioned earlier I havn't seen irq balance option in
recent kernels. Static irq2cpu assignment works well (even on HT
processors).

> While searching the web, I found reports about big performance
> increases in 3D rendering due to disabling the feature. Can this be
> true and why?

I don't have a clue. I can say one thing - static irq2cpu assignment
worx4me on linux routers hauling 400+kpps. Without it only one
core/processor would be used.

cheers,
Marek Kierdelewicz
___
LARTC mailing list
LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc

[LARTC] Problem with two providers:Need to route packets on the interface on which they arrives.

[KORTA]

Hello,

 

i would like to know how to resolve a problem.

 

I have a debian router with 3 interfaces
(LAN, and two internet providers (Provider A, Provider B)).

 

The default route is configured to use the
provider A

 

The problem is that,

When an external connection arrives from
provider B to an internal server (with nat),  the packet is routed to the
default route:

I explain:

- 
A packet arrives from provider
B in direction of a internal server

- 
The router performs nat
operation

- 
The internal server generates a
response

- 
The router routes the packet on
the interface Provider A

 

Consequently, The connection cannot been
established 

 

I want to know if its possible to configure
my debian router to route packets to the interface on which packet arrives. In
the example, packets should have been routed by the interface connected to
provider B

 

If yes, do you know how to do that ?

 

Thanks.

 

 






___
LARTC mailing list
LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc

Re: [LARTC] Two outbound internet links, using one network interface

[Dashamir Hoxha]

Dashamir Hoxha wrote:

Pio Mendez wrote:


 


>Pio Mendez wrote:
>>PREROUTING chain is not traversed by local traffic, but OUTPUT
>>chain does.
>
>I think that OUTPUT is traversed after routing decision is 
taken, so

>it is still the same problem.


I'm using OUTPUT chain in production environment to balance squid
box traffic between 2 ISP, so I'm sure that you can reroute output
packets using mangle OUTPUT chain.

After traversing mangle and nat OUTPUT chains there is another
routing process. Please check this diagram:

http://www.imagestream.com/~josh/PacketFlow.png



Pio Mendez is right. I have just tested it and it works.

If I use:
iptables -t nat -A POSTROUTING -o eth0 -j MASQUERADE
instead of:
iptables -t nat -A POSTROUTING -o eth0 -m mark --mark 0x2 -j SNAT 
--to-source $IP2

iptables -t nat -A POSTROUTING -o eth0 -j SNAT --to-source $IP1
it seems not to work. So, maybe I didn't test it properly and actualy
it doesn't work. Anyway, it is not so important.

Dashamir


Now the script becomes something like this:

-8<--
ip link set eth0 up
ip address flush eth0
ip address add $IP1 dev eth0
ip address add $IP2 dev eth0

route add to default via $GATEWAY1

ip route flush table 2
ip route show table main | grep -Ev ^default \
| while read ROUTE ; do ip route add table 2 $ROUTE ; done
ip route add table 2 default via $GATEWAY2

ip rule del fwmark 2 table 2   2>/dev/null
ip rule add fwmark 2 table 2

iptables -t mangle -N MARK-RULES
iptables -t mangle -A PREROUTING -j MARK-RULES
iptables -t mangle -A OUTPUT -j MARK-RULES

PORT_LIST="22 53"
for PORT in $PORT_LIST
do
iptables -t mangle -A MARK-RULES -m tcp -p tcp -dport $PORT -j MARK 
--set-mark 0x2

done

iptables -t nat -A POSTROUTING -o eth0 -m mark --mark 0x2 -j SNAT 
--to-source $IP2

iptables -t nat -A POSTROUTING -o eth0 -j SNAT --to-source $IP1
8<-
___
LARTC mailing list
LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc




___
LARTC mailing list
LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc

Re: [LARTC] HTB has 2 bucket?

[Thossapron Apinyapanha]

thank you for your reply 
 
i'm doubt about 
before i known , there have two bucket.
my assumption is ...
first bucket for keep token sending with rate guarantee
and this bucket size are same as burst size
and in every minute ... number of token will create at 
least rate token but max with burst size
 
but after true is ... HTB have 2 bucket 
My assumption are true for htb technique to handle "burst traffic"???
first bucket size for keep token is burst size for handle when burst traffic coming.
every minute, number of token will create at least rate token for guarantee bandwidth
but max with burst size and support burst (i'm not sure if max token are burst size, at first time
to handle burst traffic. htb will use all token in bucket but when next second??? Bucket are empty?
so max bucket size may be bigger than burst size??)
 
when class use banwidth exceed rate, class will borrow banwidth from parent class
so class will use ctoken that keep in another bucket(bucket for keeping ctoken and
ctoken are come from only parent' token)
i don't understand when burst traffic coming
how htb choose between 2 bucket are use?
i see picture from http://linux-ip.net/traffic-control/htb-class.png
don't understand. why must checking in ctoken before token 
Is it true? because burst traffic will use token more than token in first Bucket.
so it will skip to check ctoken. Is it max enough, it will use all token in first+bucket
and ctoken in secoond Bucket
 
because for theory htb to handle burst traffic,
htb allow to sending with burst rate until average rate equal to burst threadhold,
htb will change rate to rate guarantee, 
 
if  use only token in first bucket to handle burst traffic
i think this mechanism will use token and among sending with burst rate
if number of available token are qual or more than little bit num of rate token 
for guarantee bandwidth, it will change to sending with rate guarantee?
(i'm not sure for my assumtion, If my assumption are wrong. can you tell me the trueth?)
 
advise me please,
 
thank you
 
 
- Original Message From: Martin A. Brown <[EMAIL PROTECTED]>To: Thossapron Apinyapanha <[EMAIL PROTECTED]>Cc: lartc lartc Sent: Friday, October 13, 2006 7:56:02 PMSubject: Re: [LARTC] HTB has 2 bucket?
-BEGIN PGP SIGNED MESSAGE-Hash: SHA1Greetinsg Thossapron,: in HTB use 2 bucket for manage 2 rate??? first bucket -> keep : token for sending with rate second bucket -> keep ctoken for : sending with ceil rate Is it true?? may be i'm misunderstand : about token/bucket thoeryYes, there are two different buckets used.  One bucket is for tokens, another bucket is for ctokens.  Brief picture of association of parameters:  rate:  burst, tokens  ceil:  cburst, ctokensSee the upper right corner of this diagram [0].  In particular, I should warn you that the SFQ qdisc in this diagram is the one which is granted the dequeue opportunity, so although packets mostly flow from left to right in this diagram, the SFQ is displayed to the left of the HTB rate/ceil buckets, even though logically this is
 reversed.Good luck,- -Martin[0] http://linux-ip.net/traffic-control/htb-class.png- -- Martin A. Brownhttp://linux-ip.net/-BEGIN PGP SIGNATURE-Version: GnuPG v1.4.2 (GNU/Linux)Comment: pgf-0.72 (http://linux-ip.net/sw/pine-gpg-filter/)iD8DBQFFL4zmHEoZD1iZ+YcRAm1mAJ42tQy4cRL88JnuwR2/YR3zrRoTOACfbLtuccrh3V/7eBzDlpRvWTgOtZs==RqAV-END PGP SIGNATURE-
___
LARTC mailing list
LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc

Re: [LARTC] HTB has 2 bucket?

[Thossapron Apinyapanha]

thank you for your reply 
 
i'm doubt about 
before i known , there have two bucket.
my assumption is ...
first bucket for keep token sending with rate guarantee
and this bucket size are same as burst size
and in every minute ... number of token will create at 
least rate token but max with burst size
 
but after true is ... HTB have 2 bucket 
My assumption are true for htb technique to handle "burst traffic"???
first bucket size for keep token is burst size for handle when burst traffic coming.
every minute, number of token will create at least rate token for guarantee bandwidth
but max with burst size and support burst (i'm not sure if max token are burst size, at first time
to handle burst traffic. htb will use all token in bucket but when next second??? Bucket are empty?
so max bucket size may be bigger than burst size??)
 
when class use banwidth exceed rate, class will borrow banwidth from parent class
so class will use ctoken that keep in another bucket(bucket for keeping ctoken and
ctoken are come from only parent' token)
i don't understand when burst traffic coming
how htb choose between 2 bucket are use?
i see picture from http://linux-ip.net/traffic-control/htb-class.png
don't understand. why must checking in ctoken before token 
Is it true? because burst traffic will use token more than token in first Bucket.
so it will skip to check ctoken. Is it max enough, it will use all token in first+bucket
and ctoken in secoond Bucket
 
because for theory htb to handle burst traffic,
htb allow to sending with burst rate until average rate equal to burst threadhold,
htb will change rate to rate guarantee, 
 
if  use only token in first bucket to handle burst traffic
i think this mechanism will use token and among sending with burst rate
if number of available token are qual or more than little bit num of rate token 
for guarantee bandwidth, it will change to sending with rate guarantee?
(i'm not sure for my assumtion, If my assumption are wrong. can you tell me the trueth?)
 
advise me please,
 
thank you
 
 
- Original Message From: Martin A. Brown <[EMAIL PROTECTED]>To: Thossapron Apinyapanha <[EMAIL PROTECTED]>Cc: lartc lartc Sent: Friday, October 13, 2006 7:56:02 PMSubject: Re: [LARTC] HTB has 2 bucket?
-BEGIN PGP SIGNED MESSAGE-Hash: SHA1Greetinsg Thossapron,: in HTB use 2 bucket for manage 2 rate??? first bucket -> keep : token for sending with rate second bucket -> keep ctoken for : sending with ceil rate Is it true?? may be i'm misunderstand : about token/bucket thoeryYes, there are two different buckets used.  One bucket is for tokens, another bucket is for ctokens.  Brief picture of association of parameters:  rate:  burst, tokens  ceil:  cburst, ctokensSee the upper right corner of this diagram [0].  In particular, I should warn you that the SFQ qdisc in this diagram is the one which is granted the dequeue opportunity, so although packets mostly flow from left to right in this diagram, the SFQ is displayed to the left of the HTB rate/ceil buckets, even though logically this is
 reversed.Good luck,- -Martin[0] http://linux-ip.net/traffic-control/htb-class.png- -- Martin A. Brownhttp://linux-ip.net/-BEGIN PGP SIGNATURE-Version: GnuPG v1.4.2 (GNU/Linux)Comment: pgf-0.72 (http://linux-ip.net/sw/pine-gpg-filter/)iD8DBQFFL4zmHEoZD1iZ+YcRAm1mAJ42tQy4cRL88JnuwR2/YR3zrRoTOACfbLtuccrh3V/7eBzDlpRvWTgOtZs==RqAV-END PGP SIGNATURE-
___
LARTC mailing list
LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc