Will the manipulation of TOS affect throughput/delay/etc ? I have tried to set TOS of packets going through a particular source port to have Maximum Throughput. But however, it doesn''t seem to have any effect on the throughput. Experiment: 3 TCP full streams from server into client: source port 42010 -> TOS set to Maximum throughput (0x01 0x08) source port 42011 -> normal source port 42012 -> normal However, the throughput is identical for each of the stream. Am i missing something here?
On Sat, Mar 24, 2001 at 01:30:46AM +0800, Rick Goh wrote:> Will the manipulation of TOS affect throughput/delay/etc ? > > I have tried to set TOS of packets going through a particular source port to have Maximum Throughput. But however, it doesn''t seem to have any effect on the throughput. > > Experiment: > 3 TCP full streams from server into client: > source port 42010 -> TOS set to Maximum throughput (0x01 0x08) > source port 42011 -> normal > source port 42012 -> normal > > However, the throughput is identical for each of the stream.I think only the ''minimum delay'' feature is useful. Many routers (Linux included) will then insert your packets in front of the queue. Note however that it is very important that your UPSTREAM router supports this! Support on your own gateway is not enough. Regards, bert -- http://www.PowerDNS.com Versatile DNS Services Trilab The Technology People ''SYN! .. SYN|ACK! .. ACK!'' - the mating call of the internet
On Mar 24 2001, bert hubert wrote:> I think only the ''minimum delay'' feature is useful. Many routers > (Linux included) will then insert your packets in front of the > queue.Ahh... Another possible reason for packets arriving out of order... []s, Roger... -- =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- Rogerio Brito - rbrito@iname.com - http://www.ime.usp.br/~rbrito/ =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
On Sun, Mar 25, 2001 at 12:57:20AM -0300, Rogerio Brito wrote:> On Mar 24 2001, bert hubert wrote: > > I think only the ''minimum delay'' feature is useful. Many routers > > (Linux included) will then insert your packets in front of the > > queue. > > Ahh... Another possible reason for packets arriving out of > order...No. All packets within a session have the same TOS flags. Linux has a stable three-band priority queue. Regards, bert -- http://www.PowerDNS.com Versatile DNS Services Trilab The Technology People ''SYN! .. SYN|ACK! .. ACK!'' - the mating call of the internet
On Sun, 25 Mar 2001, bert hubert wrote:> On Sun, Mar 25, 2001 at 12:57:20AM -0300, Rogerio Brito wrote: > > On Mar 24 2001, bert hubert wrote: > > > I think only the ''minimum delay'' feature is useful. Many routers > > > (Linux included) will then insert your packets in front of the > > > queue. > > > > Ahh... Another possible reason for packets arriving out of > > order... > > No. All packets within a session have the same TOS flags. Linux has a stable > three-band priority queue.Define session? One single TCP connection? If that is what you mean, then you are dead wrong. OpenSSH for one sets the TOS flags for a TCP connection *after* it has logged in successfully. This provided some weird behaviour in interacting with multipath uplinks, as suddenly the (already set up) connection got routed out a different interface. Doei, Arthur. -- /\ / | arthurvl@sci.kun.nl | Work like you don''t need the money /__\ / | A friend is someone with whom | Love like you have never been hurt / \/__ | you can dare to be yourself | Dance like there''s nobody watching
On Mon, Mar 26, 2001 at 01:47:25PM +0200, Arthur van Leeuwen wrote:> Define session? One single TCP connection? If that is what you mean, then > you are dead wrong. OpenSSH for one sets the TOS flags for a TCP connection > *after* it has logged in successfully. This provided some weird behaviour in > interacting with multipath uplinks, as suddenly the (already set up) > connection got routed out a different interface.But is this a problem? -- http://www.PowerDNS.com Versatile DNS Services Trilab The Technology People ''SYN! .. SYN|ACK! .. ACK!'' - the mating call of the internet
On Mon, 26 Mar 2001, bert hubert wrote:> On Mon, Mar 26, 2001 at 01:47:25PM +0200, Arthur van Leeuwen wrote: > > > Define session? One single TCP connection? If that is what you mean, then > > you are dead wrong. OpenSSH for one sets the TOS flags for a TCP connection > > *after* it has logged in successfully. This provided some weird behaviour in > > interacting with multipath uplinks, as suddenly the (already set up) > > connection got routed out a different interface. > > But is this a problem?No, just like out-of-order transmission and receipt of IP packets isn''t a problem. You just have to be aware of it, and take measures accordingly. In the case of multipath uplink routing it may become a problem if IP packets are masqueraded on each uplink, to a *different* IP address, such as when you have ADSL connectivity from 3 providers... Doei, Arthur. -- /\ / | arthurvl@sci.kun.nl | Work like you don''t need the money /__\ / | A friend is someone with whom | Love like you have never been hurt / \/__ | you can dare to be yourself | Dance like there''s nobody watching
On Mon, Mar 26, 2001 at 02:07:05PM +0200, Arthur van Leeuwen wrote: [snip]> the case of multipath uplink routing it may become a problem if IP packets > are masqueraded on each uplink, to a *different* IP address, such as when > you have ADSL connectivity from 3 providers...sick sick sick Just because you can doesn''t mean you should.
On Mon, Mar 26, 2001 at 09:40:04AM -0500, Gregory Maxwell wrote:> On Mon, Mar 26, 2001 at 02:07:05PM +0200, Arthur van Leeuwen wrote: > [snip] > > the case of multipath uplink routing it may become a problem if IP packets > > are masqueraded on each uplink, to a *different* IP address, such as when > > you have ADSL connectivity from 3 providers... > > sick sick sick > Just because you can doesn''t mean you should.I would try to run three vtund tunnels to the same endpoint in that case. Shit this bad just *needs* to be encapsulated. Regards, bert -- http://www.PowerDNS.com Versatile DNS Services Trilab The Technology People ''SYN! .. SYN|ACK! .. ACK!'' - the mating call of the internet
Of course it does! :P In our dormitory rooms each person is intended to use one 10-Base-T port, but we''re going to run a box that connects to both and balances the traffic across both. We''re also going to set up similar boxes in our friends'' rooms, and balance the *services* between them. We''re going to squeeze out every ounce of bandwidth we can. It''s going to be fun, to say the least. :-) On Mon, 26 Mar 2001, Gregory Maxwell wrote:> On Mon, Mar 26, 2001 at 02:07:05PM +0200, Arthur van Leeuwen wrote: > [snip] > > the case of multipath uplink routing it may become a problem if IP packets > > are masqueraded on each uplink, to a *different* IP address, such as when > > you have ADSL connectivity from 3 providers... > > sick sick sick > Just because you can doesn''t mean you should. > > _______________________________________________ > LARTC mailing list / LARTC@mailman.ds9a.nl > http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://ds9a.nl/2.4Routing/ >
> On Sun, 25 Mar 2001, bert hubert wrote: > > No. All packets within a session have the same TOS flags. Linux has a stable > > three-band priority queue.On Mon, Mar 26, 2001 at 01:47:25PM +0200, Arthur van Leeuwen wrote:> > Define session? One single TCP connection? If that is what you mean, then > you are dead wrong. OpenSSH for one sets the TOS flags for a TCP connection > *after* it has logged in successfully. This provided some weird behaviour in > interacting with multipath uplinks, as suddenly the (already set up) > connection got routed out a different interface. >I don''t know about 2.4, but on 2.2 I disable to "routing based on tos" in all of my compiles. I haven''t noticed any routes changing based on tos in my route cache.
"John Anthony Kazos Jr." wrote:> > Of course it does! :P In our dormitory rooms each person is intended to > use one 10-Base-T port, but we''re going to run a box that connects to both > and balances the traffic across both. We''re also going to set up similar > boxes in our friends'' rooms, and balance the *services* between them. > We''re going to squeeze out every ounce of bandwidth we can. > > It''s going to be fun, to say the least. :-)Arrrrrggggggghhhhhhh! Folks like you make bandwidth throlling by folks like me a necessity, to preserve what we have for the other campus users. We have seen P2P activities drive us to 80+ percent utilization at times, with Napster-likes being the bulk thereof. We''ve had servers and RPGs cause similar load problems, especially when we have a truly talented designer/developer pop up. This isn''t a tirade against the students, more a plea. We''re seeing bandwidth demand accellerate beyond our (as network managers'') ability to increase supply. Part of this is that bandwidth isn''t free... nor even cheap... and we have to live on budgets based on what we anticipated 18 or so months ago, and the management above us in the food chain subsequently cut. If we guessed wrong in either direction, our credibility was screwed and we can''t get an increase next year... Think about it. This is a cool _ONE_TIME_ experiment, but if it ramps up, expect your name in the student newspaper as one of the reasons the bandwidth for your dorm was cut to 56kbps. TTFN, gerry -- Gerry Creager | Never ascribe to Malice that AATLT | which can adequately be Texas A&M University | explained by Stupidity. 979.458.4020 (Phone) | -- Lazarus Long 979.847.8578 (Fax)
I figure that, in the ideal situation, bandwidth throttling would occur on a drop-by-drop basis. (That''s the only way I can see to do it that can''t be abused, like the TOS fields can, to bring it back to that discussion. ;-)) In that ideal situation, what I plan to do (and, just to un-fluff the feathers of sys-admins everywhere, at a not--continually-maximal bandiwdth utilization) is to "reclaim" (read: fully utilize) any bandwidth wasted by one person not fully using theirs, just as space is wasted if you have two partitions with a filing system in each, which can''t share files, rather than a single partition, or two filing systems which can combine their unused space. And by the sharing of services (not bandwidth; services) amongst different rooms, a greater amount of idleness is maintained on each, so peaks are taken care of handily, and usage is balanced over the entire network. (In the *truly* ideal scenario, all users of the entire network across campus, across a state, in the world, ad infinitum, would share the same logical bandwidth with no waste. The "real" world is a subset, or an approximation, of the ideal world.) On Mon, 26 Mar 2001, Gerry Creager wrote:> "John Anthony Kazos Jr." wrote: > > > > Of course it does! :P In our dormitory rooms each person is intended to > > use one 10-Base-T port, but we''re going to run a box that connects to both > > and balances the traffic across both. We''re also going to set up similar > > boxes in our friends'' rooms, and balance the *services* between them. > > We''re going to squeeze out every ounce of bandwidth we can. > > > > It''s going to be fun, to say the least. :-) > > Arrrrrggggggghhhhhhh! Folks like you make bandwidth throlling by folks > like me a necessity, to preserve what we have for the other campus > users. We have seen P2P activities drive us to 80+ percent utilization > at times, with Napster-likes being the bulk thereof. We''ve had servers > and RPGs cause similar load problems, especially when we have a truly > talented designer/developer pop up. > > This isn''t a tirade against the students, more a plea. We''re seeing > bandwidth demand accellerate beyond our (as network managers'') ability > to increase supply. Part of this is that bandwidth isn''t free... nor > even cheap... and we have to live on budgets based on what we > anticipated 18 or so months ago, and the management above us in the food > chain subsequently cut. If we guessed wrong in either direction, our > credibility was screwed and we can''t get an increase next year... > > Think about it. This is a cool _ONE_TIME_ experiment, but if it ramps > up, expect your name in the student newspaper as one of the reasons the > bandwidth for your dorm was cut to 56kbps. > > TTFN, gerry >
On Mon, 26 Mar 2001 15:25:47 -0800, you wrote:>I don''t know about 2.4, but on 2.2 I disable to "routing based on tos" in >all of my compiles. I haven''t noticed any routes changing based on tos in >my route cache.Where is located exactly this option?? =-=-=-=-=-=-=-=-=-=-=-=-=-=-=- ** RoMaN SoFt / LLFB ** roman@madrid.com http://pagina.de/romansoft ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
On Tue, Mar 27, 2001 at 02:02:44PM +0200, RoMaN SoFt / LLFB !! wrote:> On Mon, 26 Mar 2001 15:25:47 -0800, you wrote: > > >I don''t know about 2.4, but on 2.2 I disable to "routing based on tos" in > >all of my compiles. I haven''t noticed any routes changing based on tos in > >my route cache. > > Where is located exactly this option?? >It''s under network options. It shows after you enable policy routing. Can''t be more specific, no time.. Mike