Kitz ADSL Broadband Information
adsl spacer  
Support this site
Home Broadband ISPs Tech Routers Wiki Forum
 
     
   Compare ISP   Rate your ISP
   Glossary   Glossary
 
Please login or register.

Login with username, password and session length
Advanced search  

News:

Pages: 1 2 [3] 4 5

Author Topic: FTTC Line Bonding?  (Read 19783 times)

WWWombat

  • Kitizen
  • ****
  • Posts: 1674
Re: FTTC Line Bonding?
« Reply #30 on: February 08, 2016, 06:38:18 PM »

How would that work in a typical DSL scenario with PPPoE router connected by Ethernet (at least 10meg) to a DSL modem with let's say an 800K upload speed?   If the router just sends the packets as fast as they arrive and the modem drops them if the uplink is still busy, then that means no possibility of upstream QoS on the router.  Or to be more precise no possibility of prioritisation by the router.

Wouldn't you have that anyway? Where your PC, attached with a "gigabit ethernet" connection to the switch, sends too much for the switch to forward to the router, because it is only connected with a 100Mbps "fast ethernet" connection? Now your worry has moved into your switch, not your modem.

The truth is that the TCP protocol starts slow, and ramps up ... until packets start to be dropped somewhere (anywhere) along the route. It then falls back to a speed where everything was happy.

If the bottleneck is at the link between router and modem, as your example, then the router would never have been receiving stuff to send on at 10Mbps; once the modem even started to dump packets (at 800k if it is dumb; earlier if smart), TCP will have backed off at source, and the router will only end up receiving packets at (or below) 800k. A sustainable level for passing onward into the modem.

If a second PC starts trying to send, the two independent TCP stacks will fall back to around 400k.

TCP doesn't work by dumping a huge file as fast as it can into a router, and then waiting for the queue to empty. It attempts to maintain a steady-ish speed that can be sustained end-to-end.

If you want QoS to work, then all your network nodes that have any congestion control within them need to support the same QoS system; this gets used by every place that chooses what packets to dump, and what to favour. It builds on top of standard TCP congestion mechanisms.

It has always been my understanding that the Internet does not have QoS, so from your modem outbound, no QoS applies

"The internet" doesn't. But QoS exists on many, more private, networks. BT's network - which your packets traverse before arriving at your ISP and "the internet" - does employ QoS. As we go further forward, and your voice calls become a service on top of IP, you can be sure that BT will have QoS mechanisms that keep voice prioritised.

I thought it was well known that if you turn off QOS on a Huawei HG612, your upstream sync increases by 1 Mbps.

Yup. Even if you don't send any QoS information with your upstream data (and almost noone does), the HG612 would spend time looking for those QoS markers, to decide how to queue your packets, and to decide what order to drop things in. This check was enough to slow down upstream throughput. Wasted work... but work, nonetheless.
Logged

burakkucat

  • Respected
  • Senior Kitizen
  • *
  • Posts: 38300
  • Over the Rainbow Bridge
    • The ELRepo Project
Re: FTTC Line Bonding?
« Reply #31 on: February 08, 2016, 07:20:32 PM »

I thought it was well known that if you turn off QOS on a Huawei HG612, your upstream sync increases by 1 Mbps.

Do you have any links to experimental results, please? (I am well aware of the BatBoy claims in the TBB forum but would appreciate sight of something more substantial.)
Logged
:cat:  100% Linux and, previously, Unix. Co-founder of the ELRepo Project.

Please consider making a donation to support the running of this site.

Dray

  • Kitizen
  • ****
  • Posts: 2361
Re: FTTC Line Bonding?
« Reply #32 on: February 08, 2016, 07:54:35 PM »

No, I haven't tried. I just meant it was a "well known fact" :)

But as far as I'm aware, you can only control QOS on the upstream. It's up to the ISP to control it on the downstream, like Plusnet do for example.
Logged

aesmith

  • Kitizen
  • ****
  • Posts: 1216
Re: FTTC Line Bonding?
« Reply #33 on: February 08, 2016, 07:54:47 PM »

How would that work in a typical DSL scenario with PPPoE router connected by Ethernet (at least 10meg) to a DSL modem with let's say an 800K upload speed?   If the router just sends the packets as fast as they arrive and the modem drops them if the uplink is still busy, then that means no possibility of upstream QoS on the router.  Or to be more precise no possibility of prioritisation by the router.

Wouldn't you have that anyway? Where your PC, attached with a "gigabit ethernet" connection to the switch, sends too much for the switch to forward to the router, because it is only connected with a 100Mbps "fast ethernet" connection? Now your worry has moved into your switch, not your modem.

When I say QoS I'm referring as I said specifically to queuing and prioritisation, congestion management techniques.  It is a first principle that these need to be applied at the point of congestion, where the decision about which packets to send and which to drop is being made.   For example I might configure the router to reserve bandwidth for voice or voice signalling traffic.   Then when the link becomes fully utilised, those voice packets will be forwarded in preference to others.  However in the PPPoE router/modem scenario the router will never see it's Ethernet becoming fully utilised, so that selective treatment won't come into play and the forward or drop decision will be made by the modem.

If you want QoS to work, then all your network nodes that have any congestion control within them need to support the same QoS system; this gets used by every place that chooses what packets to dump, and what to favour. It builds on top of standard TCP congestion mechanisms.

In an ideal world yes, however if the drop probability is significantly less further along the path then the sort of techniques I described above are effective.
Logged

aesmith

  • Kitizen
  • ****
  • Posts: 1216
Re: FTTC Line Bonding?
« Reply #34 on: February 08, 2016, 07:59:35 PM »

By the way, my reason for asking in the first place was to do with traffic allocation between links that are load sharing on a packet by packet basis.  If PPPoE had it's own flow control then each output queue on the FB would only empty at the speed that it's modem was forwarding.  Hence load sharing wouldn't need the relative bandwidths hard coded.
Logged

d2d4j

  • Kitizen
  • ****
  • Posts: 1103
Re: FTTC Line Bonding?
« Reply #35 on: February 08, 2016, 08:12:35 PM »

Hi

If dray is willing to reverse his QoS setting, I will reverse my QoS, which should answer the theory of QoS turned off/on give any benefit

I have only just joined online logging with tony, so if dray is willing, I just need a time/date to complete this test and leave running until for say 2 days, and you can see the results

Many thanks

John
Logged

Dray

  • Kitizen
  • ****
  • Posts: 2361
Re: FTTC Line Bonding?
« Reply #36 on: February 08, 2016, 08:23:05 PM »

Sorry, from my somewhat faulty memory, I think it only has an effect if the HG612 is being used as a router. In bridge mode, the QOS settings are ignored.
Logged

WWWombat

  • Kitizen
  • ****
  • Posts: 1674
Re: FTTC Line Bonding?
« Reply #37 on: February 08, 2016, 08:32:42 PM »

Do you have any links to experimental results, please? (I am well aware of the BatBoy claims in the TBB forum but would appreciate sight of something more substantial.)

Turning QoS off did have a beneficial impact on upstream speeds for me.

I've found a graph from my SamKnows box when I was on a 40/10 package with Plusnet.

I had an equally good result on the 80/20 package, but I can't find a graph for that ... and my SamKnows account got reset when they sent me a new box, so I lost all the old results.
Logged

d2d4j

  • Kitizen
  • ****
  • Posts: 1103
Re: FTTC Line Bonding?
« Reply #38 on: February 08, 2016, 09:45:31 PM »

Hi

If it helps, and as wombat is using samknows as the guide, I might fully turn off QoS on the hg612, and let it run for a few days then post the result.

On samknows, our upload has been fairly stable at just under 17mb, so any difference should show

Would this be of any use to anyone

Myself, it does not matter either way, but please be aware we are a heavy user, well I think so, last month with uploaded over 50gb and downloaded over 350gb, and that is with some days missing from dslstats

Dray, sorry I never thought of that, but wombat post appears to show differently unless wombat uses it as a router and not bridge mode. If router, then I suspect it cannot handle both, whereas our draytek can

I did consider if you were meaning an external speedtest as apposed to syncd speed, but if so, we use helweb from speedguide.net, as this is our server and we know it's setup/network etc

Many thanks

John
Logged

Dray

  • Kitizen
  • ****
  • Posts: 2361
Re: FTTC Line Bonding?
« Reply #39 on: February 08, 2016, 09:48:58 PM »

As I said, faulty memory so I'd appreciate a refresh from someone else as I couldn't test it at the moment.
Logged

tbailey2

  • Kitizen
  • ****
  • Posts: 1245
Re: FTTC Line Bonding?
« Reply #40 on: February 08, 2016, 10:22:16 PM »


This might be worth reading from 2012 (Kitz link):

Re disabling QOS on HG612
Logged
Tony
My Books!
Plusnet 80/20 - DSLstats - HG612/TG582n - ECI

d2d4j

  • Kitizen
  • ****
  • Posts: 1103
Re: FTTC Line Bonding?
« Reply #41 on: February 08, 2016, 10:42:16 PM »

Hi

Many thanks and not read that before thank you

I am talking about hg612 and HH as we use draytek, but interestingly, I did not post, but the QoS on the hg612 does not appear to have profiles setup, either for LAN or WAN, other then FQP, so I'm not sure if turned on, it actually does any QoS

If I do any tests, please let me know the tests, but we have a range of IP address, which some must remain working with the exception of quick breaks, so can do most tests.

If using samknows as baseline, I can post present upload graphs, with QoS as standard on, then turn it off and let it run for a few days, and you can see the result

If you want me to do this, just let me know and I'll organise it, along with any other test which is feasible to complete without bringing our IPs offline

Many thanks

John
Logged

burakkucat

  • Respected
  • Senior Kitizen
  • *
  • Posts: 38300
  • Over the Rainbow Bridge
    • The ELRepo Project
Re: FTTC Line Bonding?
« Reply #42 on: February 08, 2016, 11:31:06 PM »


This might be worth reading from 2012 (Kitz link):

Re disabling QOS on HG612

Hmm . . . nine posts to that thread and one third of them were mine!

I have absolutely no recollection of that thread.  :paperbag:
Logged
:cat:  100% Linux and, previously, Unix. Co-founder of the ELRepo Project.

Please consider making a donation to support the running of this site.

burakkucat

  • Respected
  • Senior Kitizen
  • *
  • Posts: 38300
  • Over the Rainbow Bridge
    • The ELRepo Project
Re: FTTC Line Bonding?
« Reply #43 on: February 08, 2016, 11:36:57 PM »

If using samknows as baseline, I can post present upload graphs, with QoS as standard on, then turn it off and let it run for a few days, and you can see the result

If you want me to do this, just let me know and I'll organise it, along with any other test which is feasible to complete without bringing our IPs offline

Thank you. I think that would be a worthwhile experiment and it would be interesting to see if the result is similar to that which WWWombat obtained.

As for other tests . . . perhaps others will make suggestions, please.  ;)
Logged
:cat:  100% Linux and, previously, Unix. Co-founder of the ELRepo Project.

Please consider making a donation to support the running of this site.

gt94sss2

  • Kitizen
  • ****
  • Posts: 1281
Re: FTTC Line Bonding?
« Reply #44 on: February 09, 2016, 12:01:56 AM »

I thought it was well known that if you turn off QOS on a Huawei HG612, your upstream sync increases by 1 Mbps.

Do you have any links to experimental results, please? (I am well aware of the BatBoy claims in the TBB forum but would appreciate sight of something more substantial.)

I turned my QOS off last month. According to my Samknows box, the upload increased from 8.6/7MB to about 9.3MB
Logged
Pages: 1 2 [3] 4 5
 

anything