Skip to contentSkip to conversation

Multiple connections

LARTC [email protected] 8 participants 9 messages Sep 25, 2003 – Jun 10, 2004

lore
Share & tools
lore
#0 lore raw

Hi:

I have an 'advanced-networking' problem. I want to have two interfaces to

the internet, apart from the internal one to the LAN.

I came across this page:

http://www.redhat.com/mirrors/LDP/HOWTO/Adv-Routing-HOWTO/lartc.rpdb.multipl
e-links.html
<http://www.redhat.com/mirrors/LDP/HOWTO/Adv-Routing-HOWTO/lartc.rpdb.multip
le-links.html>

and Julian's patches for the kernel published here:

http://www.ssi.bg/~ja/

and the How-to: http://www.ssi.bg/~ja/nano.txt

So, I downloaded the patch:

routes-2.4.22-9.diff (December 14, 2001), and applied it to my 2.4 kernel...

The kernel re-compile part was fine, following the instructions in :

http://tldp.org/HOWTO/Kernel-HOWTO/

The issue is when I have to create a new initrd file.. it appears to create

a vmlinuz, not a bzImage as the kernel-howto document says.

So the new kernel does not boot...

Questions:

Have anyone done this?

I mean, having two connections to the Internet, using RH8.0 or RH 9.0, and

having them selecting (balancing) which interface to use to send packets

out, and receiving in either seamlessly and with NAT ?

I think this functionality should be built in and available, instead of

having to patch the kernel?

What would you guys recommend?

As for the initrd part, any suggestions? Is there an official RedHat

"re-build kernel" document?

Thanks before-hand,

Carlos

#1 lore raw

Hello again,

I am having a hard time splitting the bandwidth among my friends in my LAN.

I use HTB and iptables. As far as I see HTB works great when it comes to limiting the bandwidth (ceil parameter),

but when it comes to splitting the bandwidth as equal as possible (through the rate parameter). it is rather useless.

For instance I have a 128kbps Internet connection. I give the same rate to all my 10 "clients" (somewhere around 12kbps each).

If 2 of my clients are downloading from the internet one file each, the splitting works. and one client gets 64kbps and the other 64kbps.

If one of those 2 clients are downloading by means of a Download Manager (like FlashGet or Download Accelerator) every gets messed up.

The one downloading with multiple connections gets 90-95% of the bandwidth and the other client gets almost nothing.

And of course, most of the time a client will have multiple connections to the Internet, not only one.

Can you tell me if I am the only person using HTB that encountered this problem? Do you have an idea how to fix this?

Thanks in advance,

Vlad Mihai

#2 lore raw
In reply to [email protected]
Quoted 10 lines

I am having a hard time splitting the bandwidth among my friends in my LAN.

I use HTB and iptables. As far as I see HTB works great when it comes to limiting the bandwidth (ceil parameter),

but when it comes to splitting the bandwidth as equal as possible (through the rate parameter). it is rather useless.

disagree here... she's fantastic

Quoted 5 lines

For instance I have a 128kbps Internet connection. I give the same rate to all my 10 "clients" (somewhere around 12kbps each).

12kbps is too high for 10 clients, u shld accomodate burst, am sure ur
probably breaking the ceil parameter

Quoted 4 lines

If 2 of my clients are downloading from the internet one file each, the splitting works. and one client gets 64kbps and the other 64kbps.

i believe perturb is 10secs

Quoted 3 lines

If one of those 2 clients are downloading by means of a Download Manager (like FlashGet or Download Accelerator) every gets messed up.

there is sth wrong, for example i have 20 systems each having 4kbps and
DAP running on one doesnt exceed the 4kbps dedication, if there browsing at the same time, there browsing alone gets slow cause the've used up almost all there allocation

Quoted 4 lines

The one downloading with multiple connections gets 90-95% of the bandwidth and the other client gets almost nothing.

u sure ur using sfq ?

Quoted 10 lines

And of course, most of the time a client will have multiple connections to the Internet, not only one.

Can you tell me if I am the only person using HTB that encountered this problem? Do you have an idea how to fix this?

havent had the problem, wont mind taking alook at that script...

Quoted 7 lines

Thanks in advance,

Vlad Mihai

K

_________________________________________________________________
From the hottest toys to tips on keeping fit this winter, you’ll find a range of helpful holiday info here.
http://special.msn.com/network/happyholidays.armx

_______________________________________________
LARTC mailing list / [email protected]
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/

#3 lore raw
In reply to [email protected]

sfq generally creates a queue per connection. What you want is a simple modification to allow per source (for outgoing traffic) ip address allocation.

Quoted 8 lines
Quoted 3 lines

The one downloading with multiple connections gets 90-95% of the bandwidth and the other client gets almost nothing.

u sure ur using sfq ?

Quoted 4 lines

And of course, most of the time a client will have multiple connections to the Internet, not only one.

_______________________________________________
LARTC mailing list / [email protected]
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/

#4 lore raw
In reply to [email protected]
Quoted 7 lines · Chijioke Kalu
Quoted 7 lines

I am having a hard time splitting the bandwidth among my friends in my LAN.
I use HTB and iptables. As far as I see HTB works great when it comes to limiting the bandwidth (ceil parameter),
but when it comes to splitting the bandwidth as equal as possible (through
the rate parameter). it is rather useless.

Quoted 6 lines

disagree here... she's fantastic

Quoted 3 lines

For instance I have a 128kbps Internet connection. I give the same rate to
all my 10 "clients" (somewhere around 12kbps each).

12kbps is too high for 10 clients, u shld accomodate burst, am sure ur
probably breaking the ceil parameter

I have the same configuration and settings for 10 clients i have down here. Each clients get its proper rate. I agree that sometimes client get rate a little bit higher than the CEIL parameters, but that because SFQ able to burst. As far as i monitored my bandwidth manager, the rate wont get higher than 1 kilobit for each class i defined.

Regards,
Rio Martin.

_______________________________________________
LARTC mailing list / [email protected]
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/

#5 lore raw
In reply to [email protected]

try using SFQ.. or some similar wrr,esfq

Quoted 19 lines

I am having a hard time splitting the bandwidth among my friends in my LAN.

I use HTB and iptables. As far as I see HTB works great when it comes to limiting the bandwidth (ceil parameter),

but when it comes to splitting the bandwidth as equal as possible (through the rate parameter). it is rather useless.

For instance I have a 128kbps Internet connection. I give the same rate to all my 10 "clients" (somewhere around 12kbps each).

If 2 of my clients are downloading from the internet one file each, the splitting works. and one client gets 64kbps and the other 64kbps.

Quoted 8 lines

If one of those 2 clients are downloading by means of a Download Manager (like FlashGet or Download Accelerator) every gets messed up.

The one downloading with multiple connections gets 90-95% of the bandwidth and the other client gets almost nothing.

And of course, most of the time a client will have multiple connections to the Internet, not only one.

...

_______________________________________________
LARTC mailing list / [email protected]
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/

#6 lore raw
In reply to [email protected]

I hope I am not that pushy, but I cannot figure this out.

If 2 of my clients are downloading from the internet one file each, the splitting works and one client gets 64kbps and the other 64kbps. If one of those 2 clients are downloading by means of a Download Manager (like Flash Get or Download Accelerator) every gets messed up. The one downloading with multiple connections gets 90-95% of the bandwidth and the other client gets almost nothing.

In my opinion if I have 2 clients downloading at the same time, the sum of all the connections from each other would have to be 64kbbit. Please tell me what I have to change to make sure that the splitting works ok.
I am sure for you this is a piece of cake but for me it's a nightmare :)

This is the script that I use.

$TC qdisc add dev $LAN_IFACE root handle 1: htb default 30 $TC class add dev $LAN_IFACE parent 1: classid 1:1 htb rate 128kbit ceil 128kbit burst 80k

#--------default---------
$TC class add dev $LAN_IFACE parent 1:1 classid 1:30 htb rate 1kbit ceil 128kbit burst 80k
$TC qdisc add dev $LAN_IFACE parent 1:30 handle 300: sfq perturb 10

#-----clients...-------------

$TC class add dev $LAN_IFACE parent 1:1 classid 1:11 htb rate 8kbit ceil 128kbit burst 80k
$FILTER_IP 192.168.0.121 flowid 1:11
$TC qdisc add dev $LAN_IFACE parent 1:11 handle 110: sfq perturb 10

$TC class add dev $LAN_IFACE parent 1:1 classid 1:13 htb rate 8kbit ceil 96kbit burst 80k
$FILTER_IP 192.168.0.123 flowid 1:13
$TC qdisc add dev $LAN_IFACE parent 1:13 handle 130: sfq perturb 10

etc...

_______________________________________________
LARTC mailing list / [email protected]
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/

#7 lore raw
In reply to [email protected]
Quoted 40 lines · Mihai Vlad

I hope I am not that pushy, but I cannot figure this out.

If 2 of my clients are downloading from the internet one file each, the splitting works and one client gets 64kbps and the other 64kbps. If one of those 2 clients are downloading by means of a Download Manager (like Flash Get or Download Accelerator) every gets messed up. The one downloading with multiple connections gets 90-95% of the bandwidth and the other client gets almost nothing.

In my opinion if I have 2 clients downloading at the same time, the sum of all the connections from each other would have to be 64kbbit. Please tell me what I have to change to make sure that the splitting works ok.
I am sure for you this is a piece of cake but for me it's a nightmare :)

This is the script that I use.

$TC qdisc add dev $LAN_IFACE root handle 1: htb default 30 $TC class add dev $LAN_IFACE parent 1: classid 1:1 htb rate 128kbit ceil 128kbit burst 80k

#--------default---------
$TC class add dev $LAN_IFACE parent 1:1 classid 1:30 htb rate 1kbit ceil 128kbit burst 80k
$TC qdisc add dev $LAN_IFACE parent 1:30 handle 300: sfq perturb 10

#-----clients...-------------

$TC class add dev $LAN_IFACE parent 1:1 classid 1:11 htb rate 8kbit ceil 128kbit burst 80k
$FILTER_IP 192.168.0.121 flowid 1:11
$TC qdisc add dev $LAN_IFACE parent 1:11 handle 110: sfq perturb 10

Quoted 4 lines

$TC class add dev $LAN_IFACE parent 1:1 classid 1:13 htb rate 8kbit ceil 96kbit burst 80k
$FILTER_IP 192.168.0.123 flowid 1:13
$TC qdisc add dev $LAN_IFACE parent 1:13 handle 130: sfq perturb 10

I don't know if it's the cause if your problems, but burst 80k is too high and your rate is too low. Burst is made for small burst of a few k, not 80k. And htb needs a minimal burst before htb can work properly.

Stef

Signature · 8 lines
[email protected]
 "Using Linux as bandwidth manager"
     http://www.docum.org/
     #lartc @ irc.openprojects.net

_______________________________________________
LARTC mailing list / [email protected]
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/
#8 lore raw

[LARTC] multiple connections

In reply to [email protected]

Hello,

I have an issue with multiple connections to the Internet. I tried following the steps described in [1] but things are not working properly. I would like the network setup as follows:

______
| |- ppp0 -- Dynamic IP (PPPoE on eth2)

  Internal---- eth0 | GW |
                    |____|- eth1 -- Static IP -> Static's GW

From [1], the steps I did were:

a. ip route flush table 4
b. ip route show table main | grep -Ev ^default \
| while read ROUTE ; do
ip route add table 4 $ROUTE
done
c. ip route add table 4 default via <Static IP>

 d. iptables -t mangle -A PREROUTING -p tcp --dport 22 -s \
          <Internal Net>/24 -j MARK --set-mark 4

e. iptables -t nat -A POSTROUTING -o ppp0 -j MASQUERADE f. iptables -t nat -A POSTROUTING -o eth1 -j SNAT --to-source \
<Static IP>
g. ip rule add fwmark 4 table 4
h. ip route flush cache

Now if I try to connect to (say) a web server, everything is fine: it goes out the PPPoE connection just fine. When I try to SSH to the machine (the same box as the web query) I never get the password prompt.

Using tcpdump I get the following results. This is listening on eth1 as I try to SSH to the destination from an internal box (using lynx to connect to the same destination results in a web page):

tcpdump: listening on eth1

07:13:12.614674 <Static IP>.37662 >  <Dest IP>.ssh: S \
    2808907073:2808907073(0) win 5840 <mss1460,sackOK,timestamp \
    611570059 0,nop,wscale 0> (DF)
07:13:12.649772 <Dest IP>.ssh > <Static IP>.37662: S \
    2414052745:2414052745(0) \
    ack 2808907074 win 65535 <mss 1400,nop,wscale \
    0,nop,nop,timestamp 2742813 611570059> (DF)
07:13:15.609403 <Static IP>.37662 >  <Dest IP>.ssh: S \
    2808907073:2808907073(0) win 5840 <mss 1460,sackOK,timestamp 61 \
    1570359 0,nop,wscale 0> (DF)
07:13:15.643437 <Dest IP>.ssh > <Static IP>.37662: S \
    2414052745:2414052745(0) \
    ack 2808907074 win 65535 <mss 1400,nop,wscale \
    0,nop,nop,timestamp 2743112 611570359> (DF)
07:13:18.634659 <Dest IP>.ssh > <Static IP>.37662: S \
    2414052745:2414052745(0) ack \
    2808907074 win 65535 <mss 1400,nop,wscale 0,nop,nop,timestamp \
    2743412 611570359> (DF)

This is what the destination sees (not the same transaction):

tcpdump: listening on fxp0

07:15:59.917179 <Static IP>.37663 > <Dest IP>.ssh: S 30 \
    01400670:3001400670(0) win 5840 <mss 1400,sackOK,timestamp
    6115867860,nop,wscale 0> (DF)
07:15:59.917319 <Dest IP>.ssh > <Static IP>.37663: S 65 \
    5604264:655604264(0) ack 3001400671 win 65535 <mss \
    1452,nop,wscale0,nop,nop,timestamp 2759543 611586786> (DF)
07:16:02.911250 <Static IP>.37663 > <Dest IP>.ssh: S 30 \
    01400670:3001400670(0) win 5840 <mss 1400,sackOK,timestamp \
    6115870860,nop,wscale 0> (DF)
07:16:02.911369 <Dest IP>.ssh > <Static IP>.37663: S 65 \
    5604264:655604264(0) ack 3001400671 win 65535 <mss \
    1452,nop,wscale0,nop,nop,timestamp 2759842 611587086> (DF)
07:16:05.905034 <Dest IP>.ssh > <Static IP>.37663: S 65 \
    5604264:655604264(0) ack 3001400671 win 65535 <mss \
    1452,nop,wscale0,nop,nop,timestamp 2760142 611587086> (DF)

Also, I don't get a echo response back from the static IP. If I ping the static's GW I get answers, but not the actual static IP. The echo requests are gettng there though:

07:35:41.966769 <Dest IP> > <Static IP>: icmp: echo request 07:35:42.977156 <Dest IP> > <Static IP>: icmp: echo request 07:35:43.992579 <Dest IP> > <Static IP>: icmp: echo request 07:35:44.997944 <Dest IP> > <Static IP>: icmp: echo request 07:35:46.003377 <Dest IP> > <Static IP>: icmp: echo request

No responses come back though.

Any suggestions?

[1] http://linux-ip.net/html/adv-multi-internet.html

Signature · 7 lines
David Magda <dmagda at ee.ryerson.ca>, http://www.magda.ca/
Because the innovator has for enemies all those who have done well under
the old conditions, and lukewarm defenders in those who may do well 
under the new. -- Niccolo Machiavelli, _The Prince_, Chapter VI
_______________________________________________
LARTC mailing list / [email protected]
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/