Hi!
Can somebody tell what is the best way for load balancing with one ISP.
Here is the scenario (there is a LAN behind the wireless clients):
linux wireless client--------\
---Access Point (Orinoco
AP1000)----linux router----ISP
linux wireless client--------/
The two wireless clients are connected to the two different wireless
interfaces of the Access Point.
They could be two PC''s or one PC with two wi-fi cards (I don''t
think it
matters). Since the configuration is in use now it is not recommended to
make very long lasting experiments:).
The idea is to balance (increase) the speed for the PC''s behind the
wireless client(s). Single wireless client achieves 5Mbit/s, but is
shaped to 4 Mbit/s since at 5Mbit there is big lag, so a PC in the LAN
when using static routes can''t use more than 4Mbit even the other
channel is free.
Has somebody some experience with similar configuration, and what is the
best solution:
(t)eql, multipath routing, some variant of bonding, something else?
Is it at all possible to use bonding in such situation?
_______________________________________________
LARTC mailing list / LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/
Евгени Гечев wrote:> Hi! > Can somebody tell what is the best way for load balancing with one ISP. > Here is the scenario (there is a LAN behind the wireless clients): > > linux wireless client--------\ > ---Access Point (Orinoco > AP1000)----linux router----ISP > linux wireless client--------/ > > The two wireless clients are connected to the two different wireless > interfaces of the Access Point. > They could be two PC''s or one PC with two wi-fi cards (I don''t think it > matters). Since the configuration is in use now it is not recommended to > make very long lasting experiments:). > The idea is to balance (increase) the speed for the PC''s behind the > wireless client(s). Single wireless client achieves 5Mbit/s, but is > shaped to 4 Mbit/s since at 5Mbit there is big lag, so a PC in the LAN > when using static routes can''t use more than 4Mbit even the other > channel is free.I would probably use CBQ + [e]SFQ. You can define classes for both linux wireless clients on the linux router which borrow traffic between each other and use [e]SFQ to equal connection speed inside these classes; or the [e]SFQ part may run on the linux wireless clients> Has somebody some experience with similar configuration, and what is the > best solution: > (t)eql, multipath routing, some variant of bonding, something else?multipath routing ... where you find multipath routes on that picture ? bonding ... it''s applied in PC1 | NIC | - - - - - - - - | NIC | PC2 | NIC | - - - - - - - - | NIC | enviroment.> Is it at all possible to use bonding in such situation? > > > _______________________________________________ > LARTC mailing list / LARTC@mailman.ds9a.nl > http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/ > >_______________________________________________ LARTC mailing list / LARTC@mailman.ds9a.nl http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/
Maybe my question is not very clear, I''ll try to explain again. I use imq+htb+wrr+esfq, but it has nothing to do with the problem. The limit is the physical line, a PC behind wireless clients can''t use more than 4 mbit due to physical restriction, it just can''t use the other wireless channel. The shaper on the router is 4Mbit for each wireless client, or 2*4Mbit=8Mbit total. So, I need a way to combine the channels, and it cannot be done with classes. Ivo Vachkov wrote:> Евгени Гечев wrote: > >> Hi! >> Can somebody tell what is the best way for load balancing with one ISP. >> Here is the scenario (there is a LAN behind the wireless clients): >> >> linux wireless client--------\ >> ---Access Point (Orinoco >> AP1000)----linux router----ISP >> linux wireless client--------/ >> >> The two wireless clients are connected to the two different wireless >> interfaces of the Access Point. >> They could be two PC''s or one PC with two wi-fi cards (I don''t think >> it matters). Since the configuration is in use now it is not >> recommended to make very long lasting experiments:). >> The idea is to balance (increase) the speed for the PC''s behind the >> wireless client(s). Single wireless client achieves 5Mbit/s, but is >> shaped to 4 Mbit/s since at 5Mbit there is big lag, so a PC in the >> LAN when using static routes can''t use more than 4Mbit even the >> other channel is free. > > > I would probably use CBQ + [e]SFQ. You can define classes for both > linux wireless clients on the linux router which borrow traffic > between each other and use [e]SFQ to equal connection speed inside > these classes; or the [e]SFQ part may run on the linux wireless clients > >> Has somebody some experience with similar configuration, and what is >> the best solution: >> (t)eql, multipath routing, some variant of bonding, something else? > > multipath routing ... where you find multipath routes on that picture ?route1: linux router-->AP-->wireless client1-->PC route2: linux router-->AP-->wireless client2-->PC The PC is in the LAN behind wireless clients.> > bonding ... it''s applied in > > PC1 | NIC | - - - - - - - - | NIC | PC2 > | NIC | - - - - - - - - | NIC |Yes. Adding one more ethernet card on the linux router leads to similar picture.> > enviroment. > >> Is it at all possible to use bonding in such situation? >> >> >> _______________________________________________ >> LARTC mailing list / LARTC@mailman.ds9a.nl >> http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/ >> >> > > >_______________________________________________ LARTC mailing list / LARTC@mailman.ds9a.nl http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/
Evgeni Gechev wrote:> Maybe my question is not very clear, I''ll try to explain again. > I use imq+htb+wrr+esfq, but it has nothing to do with the problem. > The limit is the physical line, a PC behind wireless clients can''t use > more than 4 mbit due to physical restriction, it just can''t use the > other wireless channel. The shaper on the router is 4Mbit for each > wireless client, or 2*4Mbit=8Mbit total. So, I need a way to combine the > channels, and it cannot be done with classes.Maybe you need this? http://lartc.org/howto/lartc.rpdb.multiple-links.html -- pagarbiai, Artūras Šlajus _______________________________________________ LARTC mailing list / LARTC@mailman.ds9a.nl http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/
Evgeni Gechev wrote:> Maybe my question is not very clear, I''ll try to explain again. > I use imq+htb+wrr+esfq, but it has nothing to do with the problem. > The limit is the physical line, a PC behind wireless clients can''t use > more than 4 mbit due to physical restriction, it just can''t use the > other wireless channel. The shaper on the router is 4Mbit for each > wireless client, or 2*4Mbit=8Mbit total. So, I need a way to combine the > channels, and it cannot be done with classes.Hmm ... You''re right. I thought you''re asking about shaper configuration ... One way to combine both channels is to run a link between them and start link level routing protocol (OSPF for example) but the configuration becomes too expencive and somewhat unstable. The other way to increase speed is to double the capacity of both links and then bond them or use another devices (probably run 802.11a devices at 54Mbps - for more info mail me offlist).> Ivo Vachkov wrote: > >> Евгени Гечев wrote: >> >>> Hi! >>> Can somebody tell what is the best way for load balancing with one ISP. >>> Here is the scenario (there is a LAN behind the wireless clients): >>> >>> linux wireless client--------\ >>> ---Access Point (Orinoco >>> AP1000)----linux router----ISP >>> linux wireless client--------/ >>> >>> The two wireless clients are connected to the two different wireless >>> interfaces of the Access Point. >>> They could be two PC''s or one PC with two wi-fi cards (I don''t think >>> it matters). Since the configuration is in use now it is not >>> recommended to make very long lasting experiments:). >>> The idea is to balance (increase) the speed for the PC''s behind the >>> wireless client(s). Single wireless client achieves 5Mbit/s, but is >>> shaped to 4 Mbit/s since at 5Mbit there is big lag, so a PC in the >>> LAN when using static routes can''t use more than 4Mbit even the >>> other channel is free. >> >> >> >> I would probably use CBQ + [e]SFQ. You can define classes for both >> linux wireless clients on the linux router which borrow traffic >> between each other and use [e]SFQ to equal connection speed inside >> these classes; or the [e]SFQ part may run on the linux wireless clients >> >>> Has somebody some experience with similar configuration, and what is >>> the best solution: >>> (t)eql, multipath routing, some variant of bonding, something else? >> >> >> multipath routing ... where you find multipath routes on that picture ? > > > route1: linux router-->AP-->wireless client1-->PC > route2: linux router-->AP-->wireless client2-->PC > > The PC is in the LAN behind wireless clients. > >> >> bonding ... it''s applied in >> >> PC1 | NIC | - - - - - - - - | NIC | PC2 >> | NIC | - - - - - - - - | NIC | > > > Yes. Adding one more ethernet card on the linux router leads to similar > picture. > >> >> enviroment. >> >>> Is it at all possible to use bonding in such situation? >>> >>> >>> _______________________________________________ >>> LARTC mailing list / LARTC@mailman.ds9a.nl >>> http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/ >>> >>>_______________________________________________ LARTC mailing list / LARTC@mailman.ds9a.nl http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/