Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[GigabitEthernet]

I've for a while had an issue with BT Broadband whereby I get terrible ping spikes when I'm torrenting.

Initially I believed this to be a bufferbloat issue so I employed SQM through OpenWrt but although this has reduced the ping spikes from 100s of ms to say up to 50ms, I'm still seeing the problem.

I've had a chat with some people on the OpenWrt forum and SQM should really stop these issues, so they concluded it's likely congestion.

My own tests seem to confirm this, if I cap my download speed to below 20Mb, the issue seems to go away. This is no good as that's more than half of my bandwidth but it does seem to indicate the issue is congestion.

Is there anything I can do about this, or is the solution to change ISP?

[burakkucat]

Is there anything I can do about this, or is the solution to change ISP?

If your observations are due to service provider congestion then the answers to your two part question, quoted above, will be "no" and "yes".

[Alex Atkin UK]

I'd be more inclined to believe that torrenting is maxing out your routers CPU as its a big reason I switched to x86 hardware in the first place.

The more concurrent connections you have open, the more your router has to do.  On older routers you could fill the NAT connection table just running a single torrent.

SQM also dramatically increases the routers load as it means more work for every packet of data, so that may be why SQM didn't help.

[GigabitEthernet]

Alex, I am using an X86 system with a J4105 and 8GB of RAM, could this still be too slow?

[Alex Atkin UK]

That does seem unlikely, that's a pretty beefy CPU by router standards.

Refreshing my memory on your connection I can't imagine your speed is likely to stress even a consumer router with SQM.  I setup a TP-Link WDR-3600 some time ago on my friends Virgin connection and it maxed out around 130Mbit with SQM enabled, although I expect torrents would have brought that down had we tried that.

I wonder if there are any other settings that could potentially be a problem?

I think when I ran on OpenWRT x86 I had to increase the connection table size?  I think the default is 16384 max, my current state table on pfSense is 11301/395000 with 90 torrents active but not really doing anything, so mostly DHT traffic I expect.

[bogof]

An issue I had many moons ago with Intel based x86 PCs for routing duties was that interrupt moderation settings on the NIC (which are designed to help facilitate huge transfers with lower CPU loads) would harm ping times under some load conditions.   Sometimes there can be options for tuning this in the NIC drivers.

[Weaver]

That’s a good suggestion from bogof.

[GigabitEthernet]

I've swapped back to a ZyXEL NBG6617 for now still running OpenWrt, unfortunately the issue remains. This router should be enough with SQM and torrents, no?

[GigabitEthernet]

An issue I had many moons ago with Intel based x86 PCs for routing duties was that interrupt moderation settings on the NIC (which are designed to help facilitate huge transfers with lower CPU loads) would harm ping times under some load conditions.   Sometimes there can be options for tuning this in the NIC drivers.

Do you have any ideas how I would tackle this in OpenWrt?

[bogof]

First you need to workout what the ethernet driver is, and then work out what options that driver has for tuning and how to pass them to it (it's a long time since I could be bothered with OpenWRT!).  The idea behind interrupt moderation (which I think was the Intel NIC name for it, though I think others also had it) was that when high amounts of traffic were coming through to limit the max rate at which interrupts were fired.  This would benefit throughput but at the cost of response time.

Having said that, I'm not sure it will make a huge difference to you even if you find such options.  But worth a try if you can find the setting.

[GigabitEthernet]

@bogof, okay I think I've disabled it all.

https://forum.openwrt.org/t/how-to-make-ethtool-setting-persistent-on-br-lan/6433 - the stuff mentioned here seems to match up with what you're talking about.

[bogof]

I don't know if they're the right parameters or not for what I meant, but it might be they're even more appropriate..  This was a howto I found:
https://community.mellanox.com/s/article/understanding-interrupt-moderation
It's probably almost a decade since I last messed around with it and it was all different then.  Used to have to carve the config bits out of raw silicon and attach them to the config pins... :-p

[GigabitEthernet]

Thanks @bogof and all of you for your help.

I've made as many of the suggested changes to OpenWrt as possible, I will monitor the connection for a while and report back. In the meantime, would appreciate any more suggestions you may have.

[Alex Atkin UK]

Trouble is, if its happening with an actual consumer router build too, its probably not a NIC problem.  Although its kinda hard to say as obviously a router with its CPU tapped out could exhibit the same symptoms.

Did you check the CPU load when on the  ZyXEL NBG6617 and was that with SQM on and off?  I wouldn't expect it to be struggling but you never can tell with these things.

Are you throttling torrents to less than your maximum upload speed actually on the torrent client?  That's a pretty good idea in general.

[GigabitEthernet]

So I am quite sure there is congestion within BT's core network, please correct me if I am wrong.

What this graph shows is the latency over the many hops between me and 1.1.1.1 (DNS):



I experienced a latency increase at about 14:25 this afternoon, you can see that on the hops within my control, i.e. between BT HUB (hop 2) and 172.x (hop 3 - the cabinet?) there is no latency change.

But deeper into the core network you can see an increase in latency starting around hop 5, by now we're well into BT's network.

No torrents were running at this time.