dslreports logo
site
 
    All Forums Hot Topics Gallery
spc

spacer




how-to block ads


Search Topic:
uniqs
2564
share rss forum feed


hackybag

@rr.com

Big downloads ruin voice quality despite QOS

I have Tomato limiting a 1.0Mbps upload to 900Kbps with a priority of VoIP (by device MAC address) > DNS > HTTP(S) 0-512KB/s > Everything else.

Voice quality is excellent with concurrent web browsing, small downloads, and even Hulu/ Youtube. What really causes dropped syllables and static is downloads approaching the capacity of the line. Usually in the teen Mbps.

What am I doing wrong?


Bink
Villains... knock off all that evil

join:2006-05-14
Castle Rock, CO
kudos:4

Ideally you want QoS in both directions (though inbound is less effective than outbound).


gweidenh

join:2002-05-18
Houston, TX
kudos:3
reply to hackybag

I think you may have answered your own question. Limiting outbound traffic will have no affect on inbound traffic. You would need to apply QoS settings to your inbound traffic as well so that large downloads are throttled.

I am not familiar with Tomato, but have been told it can do this.


wl08

join:2010-09-01
reply to hackybag

1. Find out your device's mac address
2. Go go to QOS->Classification setting on your tomato router.
3. Add a rule like this:

From xx:xx:xx:xx:xx:xx TCP/UDP Highest Voip

Save, Done!


lorennerol
Premium
join:2003-10-29
Seattle, WA

1 recommendation

reply to hackybag

Spent months trying to work around the downstream QoS issue with an expensive business grade router. Never got a satisfactory solution and ended up ditching the SIP trunks in favor of a PRI. You just can't control downstream traffic beyond your router, which is where the bottleneck is.


gweidenh

join:2002-05-18
Houston, TX
kudos:3

pfsense works great for downstream QoS



hackybag

@rr.com
reply to wl08

That's exactly what I've done



VexorgTR

join:2012-08-27
Sheffield Lake, OH
kudos:1

How much bandwidth do you have in the first place?



hackybag

@rr.com
reply to hackybag

I don't know to be honest. It's advertised as 10/1 but speed tests burst to 20-30. Sustained downloads are usually 6-10 with the fastest downloads around 14-17.



VexorgTR

join:2012-08-27
Sheffield Lake, OH
kudos:1
Reviews:
·voip.ms
·Callcentric
·callwithus

Hmm... and how's the upload check out?

If a test bursts that fast, that could actually be part of the problem perhaps. VOIP likes things very steady. Steady beats speed for many VOIP apps.

On a site like Speedtest.net, I favor a flat speed test graph.

Try pingtest.net too... that may give us a clue if the ISP isn't steady.



mgraves1
Premium
join:2004-04-05
Houston, TX
reply to hackybag

As some have described, QoS may not provide a complete solution in both directions. This is where traffic shaping comes in. You can carve off bandwidth reserved solely for voip traffic. both m0n0wall and pfsense do this easily.


Cloneman

join:2002-08-29
Montreal
kudos:4
Reviews:
·TekSavvy DSL
·Bell Fibe

1 edit
reply to hackybag

You're doing it wrong - but that's okay, pretty sure everyone in the world is doing it wrong.

Tomato's default downstream QoS does not set a global maximum and is therefore completely useless the moment you have more than 2 classes on the downstream side

You need a version of tomato that implements proper downstream QoS, which uses a global limit for inbound.

Latest Toastman is the version i've tested to work. I believe other firmwares, like recent shibby versions, also implement Tiomo's QoS rules which are the key to proper downstream bandwidth management.

You can read some of my findings here:
»Cloneman's Tomato QoS Tips for adsl, vdsl2, and cable

Later in thread I highlight how to test your setup with visualware voip tester



hackybag

@rr.com
reply to hackybag

inbound qos not perfect

 
Click for full size
Upload is a consistent 0.83Mbps (capped by QOS).

I think having a desktop firewall is a too much complexity and cost. I got a RT-N16 to handle routing.

After capping download to 14000k, voice quality during downloads has improved from unusable to mediocre. There's a random hissing noise that varies from quiet to loud to quiet and voices sound quiet. Calls break up during the first ten seconds. Latency has decreased from 100-1000ms to the usual 50ms. Slowing down the download ameliorates these issues, and results in crystal clear voice.

Should I increase the ATA's network jitter buffer from low to medium? Or decrease the inbound cap to 13000Kbps?

wl08

join:2010-09-01

Try cap other services to 80%-90%, VOIP can set to 100% or 95%.

I never have any issue, even with heavy Usenet downloading.


gweidenh

join:2002-05-18
Houston, TX
kudos:3
reply to hackybag

Try decreasing the download limit to 10000kbps.



hackybag

@rr.com
reply to hackybag

Re: Big downloads ruin voice quality despite QOS

I tried cutting the download to 10000, lowering the others to 80% and raising VoIP to 95%, and doing both. Firing up a big download still results in audio glitching.


gweidenh

join:2002-05-18
Houston, TX
kudos:3

Cut it down to 5000

you gotta find a point where your internet connection is solid, then start working your way back up.



jubangy
Premium
join:2005-03-26
Corry, PA

I found that changing udp timeout settings unreplied to 10 and assured to 300 seems to have fixed my echoes and 1 way audio issues with tomato and voip



dracuda

join:2013-02-26
Nepean, ON
reply to hackybag

Just for thoroughness, in your pap2 config have you changed the default RPT packet size from 0.03 to 0.02? This is often mentioned as a remedy for choppy voice. Parameter is located in the SIP tab.



dracuda

join:2013-02-26
Nepean, ON
reply to hackybag

BTW, you didn't mention (or I didn't see it) what type of downloads cause the problem. If it's bittorrent downloads that are causing the headache, I found that it is very necessary to limit the upload on that type of traffic, or your download will quickly become saturated with, err, a torrent of acks and swarm traffic, even on a high speed link. Granted, I've never had a download link speed above 7Mbps, but as soon as I limited upload on torrent traffic to ~160Kbs (YMMV), I still got full speed download and voice traffic never missed a beat. Basically, it's helping to take you out of the buffer bloat equation, and pushing the it upstream, where the equipment is better able to handle it.

How to classify BT on Tomato? Basically it involves classification by exclusion. You make rules for all the stuff you want to classify, except BT. For the most part, it passes through those rules and ends up in your default class, which you have appropriately limited.


kaila

join:2000-10-11
Lincolnshire, IL
reply to hackybag

I hope you get this nailed, but in the past I've made improvements but never achieved perfect success with QoS on various consumer routers using either DD-WRT or Tomato when there was high network load.

The best QoS solutions that I've tried and found to work reliably well with VoIP are pfSense and Mikrotik. The downside of course are the learning curves, but if you have the time and inclination, it works very well.


lifespeed

join:2009-09-08
reply to hackybag

The short answer is you don't have the setup right. It is also true that very few routers do inbound QoS correctly, if at all. Yes, it exists and does work despite all the naysayers with limited understanding claiming "you can't control incoming".

I only have direct experience with Draytek Vigor 2130, an excellent consumer router with bidirectional QoS. This is an inexpensive yet high-performance consumer router you can get for around $100. Other options which I am not familiar with are PFsense and monowall. I have heard from Tomato and DD-WRT proponents, but I get the impression these are fragmented solutions that run on antiquate hardware.

Good luck sorting it out. It CAN be done.
--
Lifespeed


mdseuss

join:2012-05-27
Worcester, MA

1 recommendation


To be a pedantic naysayer, you CANNOT do true QoS without control of BOTH ENDS of the link that has the bottleneck (the consumer spigot).

Yes you can achieve nearly wonderful things with traffic shaping and "consumer grade Qos" settings, but you are really just herding reasonably well behaved applications to avoid them stepping on each other.


lorennerol
Premium
join:2003-10-29
Seattle, WA

said by mdseuss:

To be a pedantic naysayer, you CANNOT do true QoS without control of BOTH ENDS of the link that has the bottleneck (the consumer spigot).

Yes you can achieve nearly wonderful things with traffic shaping and "consumer grade Qos" settings, but you are really just herding reasonably well behaved applications to avoid them stepping on each other.

+1

If your ISP has you throttled to 5mbit, the bottleneck is on their gear, not at your router.

lifespeed

join:2009-09-08
reply to mdseuss

said by mdseuss:

To be a pedantic naysayer, you CANNOT do true QoS without control of BOTH ENDS of the link that has the bottleneck (the consumer spigot).

You do have control of both ends of the spigot. The router, with proper QoS implementation, delays ACK of incoming packets causing the sending server to slow down.

I have had this setup for years and have tested it with large FTP transfers in both directions, bittorrent and other downloads. I know it works. It sounds like you have never done this, at least correctly with capable hardware. So who should we believe? Somebody who failed to get it to work, or somebody who has?

I am becoming bored with having this same discussion over and over again trying to inform the (apparently) unwashed masses. Like I said, this is a commonly held misconception that many people subscribe to.
--
Lifespeed

mdseuss

join:2012-05-27
Worcester, MA

1 recommendation


Sorry, read some RFCs

You are stuck in "consumer grade" folklore.


lifespeed

join:2009-09-08

said by mdseuss:

Sorry, read some RFCs

You are stuck in "consumer grade" folklore.

LOL. First, the Draytek 2130 is a SMB-grade router at a consumer price, not your typical drek you find at the big box store.

Second, I do not speak about folklore, but from empirical evidence and testing. You are just throwing around wild accusations, while I have experience and hard data to back up my statements.

Just because you have never seen it work, don't know how to figure it out, or don't have the right hardware doesn't mean it doesn't exist.
--
Lifespeed

lifespeed

join:2009-09-08

1 edit
reply to mdseuss

said by mdseuss:

You are stuck in "consumer grade" folklore.

said by Cloneman:

You're doing it wrong - but that's okay, pretty sure everyone in the world is doing it wrong . . . You need a version of tomato that implements proper downstream QoS, which uses a global limit for inbound.

Latest Toastman is the version i've tested to work. I believe other firmwares, like recent shibby versions, also implement Tiomo's QoS rules which are the key to proper downstream bandwidth management.

You can read some of my findings here:
»Cloneman's Tomato QoS Tips for adsl, vdsl2, and cable

Later in thread I highlight how to test your setup with visualware voip tester

said by kaila:

The best QoS solutions that I've tried and found to work reliably well with VoIP are pfSense and Mikrotik. The downside of course are the learning curves, but if you have the time and inclination, it works very well.

If you read this thread, you will notice a couple other posters using REAL routers that have had the same success I have seen. But why listen to actual experiences when you can just make unsupported statements about inbound QoS not working?
--
Lifespeed

lifespeed

join:2009-09-08
reply to lorennerol

said by lorennerol:

If your ISP has you throttled to 5mbit, the bottleneck is on their gear, not at your router.

Your router can decide which traffic has priority to use the available 5 Mbits in your example. If you have the right router.

In almost all cases your ISP network is not saturated or overloaded, only the last mile link to your house is limited in speed. A good router manages this in both directions, given proper configuration.
--
Lifespeed

mdseuss

join:2012-05-27
Worcester, MA

1 recommendation


Doesn't matter what kind of router you have. If don't control BOTH ENDS (and hence the middle) of the "thing that needs QoS", then you aren't doing REAL QoS.

This isn't "wild accusations". Just simple networking and perhaps queue theory.