Discussion:
SQM Settings for Bonded DSL?
(too old to reply)
Rich Brown
2018-06-21 11:08:01 UTC
Permalink
Hi folks,

Our local DSL ISP (Consolidated Communications, Inc, formerly Fairpoint) recently installed a Smart/RG SR555ac (https://www.smartrg.com/sr555ac) bonded ADSL2 modem.

I seem to remember earlier messages stating that the dual queues in the DSL modem screwed up (or, de-optimized) the SQM in the router.

The web GUI does provide info on SNR, sync rates, etc. Any advice for SQM beyond the standard, "measure the no-SQM speed, then start at 5% below..."? Thanks.

Rich
Sebastian Moeller
2018-06-21 11:14:54 UTC
Permalink
Hi Rich,
Post by Rich Brown
Hi folks,
Our local DSL ISP (Consolidated Communications, Inc, formerly Fairpoint) recently installed a Smart/RG SR555ac (https://www.smartrg.com/sr555ac) bonded ADSL2 modem.
I seem to remember earlier messages stating that the dual queues in the DSL modem screwed up (or, de-optimized) the SQM in the router.
As far as I can see this should not really cause an issue (besides that latency for single packets will be limited by the fact that you have two half-total-bandwidth links) except maybe that there might be additional overhead for the bonding on the link, but I have never looked at channel bonding so this is pure spekulation. I assume here that your Modem handles the bonding transparently. I could envision that running an non-transparent load-balancer (like with mwan3 under openwrt) might introduce issues for sqm, but if all you see is one ethernet link to the modem I do not expect any major quirks (except latency not being in line with the expectancy from total bandwidth).
Post by Rich Brown
The web GUI does provide info on SNR, sync rates, etc. Any advice for SQM beyond the standard, "measure the no-SQM speed, then start at 5% below..."? Thanks.
I am curius myself and would like to ask you to keep me/the list posted on whatever you find out about the applicability of sqm to bonding.

Best Regards
Sebastian
Post by Rich Brown
Rich
_______________________________________________
Bloat mailing list
https://lists.bufferbloat.net/listinfo/bloat
Jonathan Foulkes
2018-06-21 13:16:23 UTC
Permalink
Hi Rich, Sebastian,

Most bonded modems do a good job of making the line look just like a single, higher-capacity line. The only major issues I’ve observed is when the bonded lines have some asymmetry to them (e.g. one link has weaker SNR), then the bond drops, re-synchs and continues. But occasionally, it will run for extended periods at 50% capacity (i.e. on only one of the lines).
It uses PMT protocols, so not at all like mwan3, as any one connection can achieve full throughput of the bonded set.

I have data on hundreds of bonded lines, and other than the scenarios I mentioned above about bad bonds, as far as SQM goes, it behaves just like any other DSL line of equal capacity. So all the same guidelines would apply.

I hope that helps,

Jonathan Foulkes
Post by Sebastian Moeller
Hi Rich,
Post by Rich Brown
Hi folks,
Our local DSL ISP (Consolidated Communications, Inc, formerly Fairpoint) recently installed a Smart/RG SR555ac (https://www.smartrg.com/sr555ac) bonded ADSL2 modem.
I seem to remember earlier messages stating that the dual queues in the DSL modem screwed up (or, de-optimized) the SQM in the router.
As far as I can see this should not really cause an issue (besides that latency for single packets will be limited by the fact that you have two half-total-bandwidth links) except maybe that there might be additional overhead for the bonding on the link, but I have never looked at channel bonding so this is pure spekulation. I assume here that your Modem handles the bonding transparently. I could envision that running an non-transparent load-balancer (like with mwan3 under openwrt) might introduce issues for sqm, but if all you see is one ethernet link to the modem I do not expect any major quirks (except latency not being in line with the expectancy from total bandwidth).
Post by Rich Brown
The web GUI does provide info on SNR, sync rates, etc. Any advice for SQM beyond the standard, "measure the no-SQM speed, then start at 5% below..."? Thanks.
I am curius myself and would like to ask you to keep me/the list posted on whatever you find out about the applicability of sqm to bonding.
Best Regards
Sebastian
Post by Rich Brown
Rich
_______________________________________________
Bloat mailing list
https://lists.bufferbloat.net/listinfo/bloat
_______________________________________________
Bloat mailing list
https://lists.bufferbloat.net/listinfo/bloat
Sebastian Moeller
2018-06-21 13:35:48 UTC
Permalink
Hi Jonathan,
Post by Jonathan Foulkes
Hi Rich, Sebastian,
Most bonded modems do a good job of making the line look just like a single, higher-capacity line. The only major issues I’ve observed is when the bonded lines have some asymmetry to them (e.g. one link has weaker SNR), then the bond drops, re-synchs and continues. But occasionally, it will run for extended periods at 50% capacity (i.e. on only one of the lines).
It uses PMT protocols, so not at all like mwan3, as any one connection can achieve full throughput of the bonded set.
I have data on hundreds of bonded lines, and other than the scenarios I mentioned above about bad bonds, as far as SQM goes, it behaves just like any other DSL line of equal capacity. So all the same guidelines would apply.
Thank you very much, all very useful to know.


@Rich, could I ask you to try to run the https://github.com/moeller0/ATM_overhead_detector on your shiny bonded link? I am really curious how this will cope with such a link (I expect no issues, but the proof and the pudding thing still applies).

Also I see that bonding might have additional overhead that might not be per packet so will not be picked up by ATM_overhead_detector, so you might need to aim a bit lower with the shaper settings than usually.

Best Regards
Sebastian
Post by Jonathan Foulkes
I hope that helps,
Jonathan Foulkes
Post by Sebastian Moeller
Hi Rich,
Post by Rich Brown
Hi folks,
Our local DSL ISP (Consolidated Communications, Inc, formerly Fairpoint) recently installed a Smart/RG SR555ac (https://www.smartrg.com/sr555ac) bonded ADSL2 modem.
I seem to remember earlier messages stating that the dual queues in the DSL modem screwed up (or, de-optimized) the SQM in the router.
As far as I can see this should not really cause an issue (besides that latency for single packets will be limited by the fact that you have two half-total-bandwidth links) except maybe that there might be additional overhead for the bonding on the link, but I have never looked at channel bonding so this is pure spekulation. I assume here that your Modem handles the bonding transparently. I could envision that running an non-transparent load-balancer (like with mwan3 under openwrt) might introduce issues for sqm, but if all you see is one ethernet link to the modem I do not expect any major quirks (except latency not being in line with the expectancy from total bandwidth).
Post by Rich Brown
The web GUI does provide info on SNR, sync rates, etc. Any advice for SQM beyond the standard, "measure the no-SQM speed, then start at 5% below..."? Thanks.
I am curius myself and would like to ask you to keep me/the list posted on whatever you find out about the applicability of sqm to bonding.
Best Regards
Sebastian
Post by Rich Brown
Rich
_______________________________________________
Bloat mailing list
https://lists.bufferbloat.net/listinfo/bloat
_______________________________________________
Bloat mailing list
https://lists.bufferbloat.net/listinfo/bloat
Rich Brown
2018-06-21 14:09:39 UTC
Permalink
Hi folks,

It's great to get all this advice from people who know more than I do :-) (I should note that the bonded DSL seemed to be working as expected using the no-sqm speed minus 5% procedure.)

Once all the solstice celebrations this weekend settle down (Happy Summer!), I'll pull together some data from the DSL modem and the ATM Overhead Detector and share it with the list.

Best,

Rich
Post by Sebastian Moeller
Hi Jonathan,
Post by Jonathan Foulkes
Hi Rich, Sebastian,
Most bonded modems do a good job of making the line look just like a single, higher-capacity line. The only major issues I’ve observed is when the bonded lines have some asymmetry to them (e.g. one link has weaker SNR), then the bond drops, re-synchs and continues. But occasionally, it will run for extended periods at 50% capacity (i.e. on only one of the lines).
It uses PMT protocols, so not at all like mwan3, as any one connection can achieve full throughput of the bonded set.
I have data on hundreds of bonded lines, and other than the scenarios I mentioned above about bad bonds, as far as SQM goes, it behaves just like any other DSL line of equal capacity. So all the same guidelines would apply.
Thank you very much, all very useful to know.
@Rich, could I ask you to try to run the https://github.com/moeller0/ATM_overhead_detector on your shiny bonded link? I am really curious how this will cope with such a link (I expect no issues, but the proof and the pudding thing still applies).
Also I see that bonding might have additional overhead that might not be per packet so will not be picked up by ATM_overhead_detector, so you might need to aim a bit lower with the shaper settings than usually.
Best Regards
Sebastian
Post by Jonathan Foulkes
I hope that helps,
Jonathan Foulkes
Post by Sebastian Moeller
Hi Rich,
Post by Rich Brown
Hi folks,
Our local DSL ISP (Consolidated Communications, Inc, formerly Fairpoint) recently installed a Smart/RG SR555ac (https://www.smartrg.com/sr555ac) bonded ADSL2 modem.
I seem to remember earlier messages stating that the dual queues in the DSL modem screwed up (or, de-optimized) the SQM in the router.
As far as I can see this should not really cause an issue (besides that latency for single packets will be limited by the fact that you have two half-total-bandwidth links) except maybe that there might be additional overhead for the bonding on the link, but I have never looked at channel bonding so this is pure spekulation. I assume here that your Modem handles the bonding transparently. I could envision that running an non-transparent load-balancer (like with mwan3 under openwrt) might introduce issues for sqm, but if all you see is one ethernet link to the modem I do not expect any major quirks (except latency not being in line with the expectancy from total bandwidth).
Post by Rich Brown
The web GUI does provide info on SNR, sync rates, etc. Any advice for SQM beyond the standard, "measure the no-SQM speed, then start at 5% below..."? Thanks.
I am curius myself and would like to ask you to keep me/the list posted on whatever you find out about the applicability of sqm to bonding.
Best Regards
Sebastian
Post by Rich Brown
Rich
_______________________________________________
Bloat mailing list
https://lists.bufferbloat.net/listinfo/bloat
_______________________________________________
Bloat mailing list
https://lists.bufferbloat.net/listinfo/bloat
Arie
2018-06-21 11:15:08 UTC
Permalink
I have perfect results treating my bonded VDSL like a single connection, so
like you said, measure the unshaped speed, start a few % below.
Post by Rich Brown
Hi folks,
Our local DSL ISP (Consolidated Communications, Inc, formerly Fairpoint)
recently installed a Smart/RG SR555ac (https://www.smartrg.com/sr555ac)
bonded ADSL2 modem.
I seem to remember earlier messages stating that the dual queues in the
DSL modem screwed up (or, de-optimized) the SQM in the router.
The web GUI does provide info on SNR, sync rates, etc. Any advice for SQM
beyond the standard, "measure the no-SQM speed, then start at 5% below..."?
Thanks.
Rich
_______________________________________________
Bloat mailing list
https://lists.bufferbloat.net/listinfo/bloat
David Lang
2018-06-21 18:10:24 UTC
Permalink
Is that router running openwrt? or are you having to operate downstream of it?

I've got a bonded DSL line (10/2) and the current router has a 2.6 kernel, and
has horrible buffer bloat, but I have not been able to track down something I
can use to replace it.

David Lang
Post by Rich Brown
Hi folks,
Our local DSL ISP (Consolidated Communications, Inc, formerly Fairpoint) recently installed a Smart/RG SR555ac (https://www.smartrg.com/sr555ac) bonded ADSL2 modem.
I seem to remember earlier messages stating that the dual queues in the DSL modem screwed up (or, de-optimized) the SQM in the router.
The web GUI does provide info on SNR, sync rates, etc. Any advice for SQM beyond the standard, "measure the no-SQM speed, then start at 5% below..."? Thanks.
Rich
_______________________________________________
Bloat mailing list
https://lists.bufferbloat.net/listinfo/bloat
Rich Brown
2018-06-22 03:05:12 UTC
Permalink
My DSL modem is running the native/factory OS.

A separate OpenWrt modem (WRT3200ACM with davidc502's firmware) that's running SQM handles traffic from the laptops/phones/desktops.

Rich
Post by David Lang
Is that router running openwrt? or are you having to operate downstream of it?
I've got a bonded DSL line (10/2) and the current router has a 2.6 kernel, and has horrible buffer bloat, but I have not been able to track down something I can use to replace it.
David Lang
Post by Rich Brown
Hi folks,
Our local DSL ISP (Consolidated Communications, Inc, formerly Fairpoint) recently installed a Smart/RG SR555ac (https://www.smartrg.com/sr555ac) bonded ADSL2 modem.
I seem to remember earlier messages stating that the dual queues in the DSL modem screwed up (or, de-optimized) the SQM in the router.
The web GUI does provide info on SNR, sync rates, etc. Any advice for SQM beyond the standard, "measure the no-SQM speed, then start at 5% below..."? Thanks.
Rich
_______________________________________________
Bloat mailing list
https://lists.bufferbloat.net/listinfo/bloat
Loading...