Bufferbloat and the difference between Huawei and ECI modems

For the past few years i've been using OpenWrt to successfully combat bufferbloat on my VDSL2 connection (see my blog post for my setup, including UK-specific tweaks. Although i'm now using SQM-scripts instead of QOS-scripts).

In all that time i've been using the ECI modem that was supplied to me when i first got VDSL2. With the advent of G.INP being introduced (and because my Zen line stats seem to have been broken since February) i decided to try out the Huawei HG612 i've had in a box for the past couple of years. I updated it to the latest hacked firmware and after swapping it in for the ECI i experienced some weirdness.

After swapping it in i did a simple bufferbloat test of running a continuous ping to bbc.co.uk while running a speedtest. During the upload portion of the speedtest my pings jumped from 18ms to around 100ms. I was only able to get them back to around 21ms by dropping my upload speed to 5000kbps (was 5500kbps using the ECI modem. The Huawei sync stats showed 5626 for the upload while my last Zen upload sync stats showed 6628 using the ECI) in SQM-scripts.

I then decided to run the recommended netperf-wrapper tests.

Here's my RRUL graph for the Huawei HG612. I'm not too clued up on how to read these but i think ideally you want the average line to be as smooth as possible. With the Huawei the download speeds are all over the place and i suspect i'd have to limit the download speed in SQM-scripts even more to improve things (it was set at 90% of the sync speed as reported by the Huawei).

Keeping the exact same values in SQM scripts i swapped the ECI modem back in and reran the test. Here's the RRUL graph. Much smoother, with almost half the latency.

So i'm sticking with the ECI modem even if it lacks access to the line stats and G.INP (which i don't think my cabinet has yet anyway). I was also able to bump my upload speed limit back to 5500kbps with no effect on latency.

Line Stat details:

As reported by Zen (last updated 26th February): Upstream=6628, Downstream=33878, Sync Rate=35000

As reported by Huawei: Upstream=5626, Downstream=34998

Speed limits set in SQM-scripts: Download Speed=31500kbps, Upload Speed=5000kbps

Which cabinet is it?

I suspect it's an ECI cabinet, it certainly looks like the example pics i've seen.

It's cabinet P10 at SDRYD.

Yes, they're all ECI at Ryde, except for the new ones planted this year by BDUK which are Huawei.

I'm not surprised that ECI modems perform better on ECI cabs than Huawei modems do.

Edited by deleted (Thu 09-Apr-15 13:30:47)

Except that, if Zen's stats are to be believed, both modems sync at the same down rate. And i'm limiting the down speed on the router to 90% of that.

So why does the Huawei exhibit such large peaks and troughs in the download graph? I do expect the ECI modem to perform better on ECI cabs but these results just seem weird.

I'll run another test with the Huawei tomorrow and limit my speed some more to see if that improves responsiveness.

I think your line may have been banded by the Openreach DLM. There is a 35Mbps one.

It's a possibility as the line has always been bad.

But why would that cause the differences between the two different modems?

From the graphs you can see the Huawei is all over the place with large peaks and troughs in the download traffic, induced latency of an additional 80ms and the upload flows not being prioritised correctly.

Whereas the ECI is fairly smooth with download traffic, has only an additional 20ms of induced latency and the upload traffic is bring prioritised correctly.

If both modems are syncing at the same speed and i'm actively limiting the speed to below the DLM why are the results so drastically different? Is the Huawei really that terrible when connected to an ECI cab?

Pass. Sorry.

I cant answer your question but nice work on your router and especially for making a guide to others can copy.

Thanks, i wrote it up because i couldn't find anything for recent versions of OpenWrt describing how to VLAN Wifi across multiple routers/access points.

I finally managed to get around to doing some more tests this morning.

After my initial tests with the Huawei and then putting the ECI back in place i noticed that my Zen line stats had finally updated. Where the Huawei indicated that its upstream sync was 5626kbps the Zen stats indicated that the ECI was synced at 6536kbps. So there's a substantial difference in upstream syncs although the downstream remains the same.

In my router i set the following values in SQM and kept them the same throughout the tests:

Basic Settings:
Interface name: pppoe-wan
Download speed (kbit/s) (ingress): 31000

Queue Discipline:
Queueing discipline: fq_codel
Queue setup script: simple_pppoe.qos

Link Layer Adaptation:
Which link layer to account for: Ethernet with Overhead
Per Packet Overhead (byte): 8

Under basic settings i adjusted the "Upload speed (kbit/s) (egress):" value for each test.

Here are the RRUL graphs

@5500kbps

Huawei, ECI.

Download traffic using the Huawei jumps all over the place, and the upload flows are all mashed together. There is around 60ms of induced latency!

With the ECI modem the download traffic is fairly smooth, the upload flows are properly prioritised and there's only around 10ms of induced latency.

@5000kbps

Huawei, ECI.

No change in the Huawei, it's still bad.

@4500kbps

Huawei, ECI.

Finally the Huawei shows some improvements. Downloads are fairly smooth, upload flows are prioritised and induced latency is down to 10-15ms.

As i have to limit the upload speed to a meg lower when using the Huawei to minimise my latency it corroborates what the Huawei and Zen syncing stats are saying; the ECI is syncing at a meg higher on the upstream.

So i'll be sticking with the ECI modem for my connection and, as i've a spare one of these too, i might try opening up the ECI and flashing a more recent OpenWrt onto it to gain direct access (a few people have had some success at this).

Edited by deleted (Sun 12-Apr-15 09:58:44)