Do I need to use QoS? Will it lower pings?

Hi, we have bt infinity 2 80mb and we are getting a new router that supports QoS. Do I need to use QoS for gaming to get lower pings or what? Will it improve my ping?
Would you recommend it and in what situations would you advise that we use it?
Thanks,
Ben

Ping uses the ICMP protocol to get a response from a remote device. It is not actually the ping you want to lower (as it has nothing to do with games) it is the latency of the game itself.

It is the games protocols that you would need to set the priority of for QoS (and different games will use different protocols).

If your line is not heavily utilised (ie is not running at full throttle doing downloads, streaming, etc) then QoS will make no difference. If you are maxing your connection then QoS may help to lower latency on gaming traffic - but a better method would be not to max the line whilst you are doing latency sensitive activities.

QoS on your router will make no difference if it is the network between your router and the gaming server that is causing the latency increase - ie it can only prioritise packets within your own network. Some ISPs (like PlusNet) will also use QoS in their own network to give priority within the portion they control (but again, doesn't affect anything once it leaves the ISP network).

Edited by ian72 (Fri 11-Jul-14 09:11:12)

As ian72 says, if you're not maxing your connection while gaming, it probably won't make a difference.

If you are maxing your connection and want to retain responsiveness of your connection you should make sure the QOS your router runs is underpinned by fq_codel (hint: it doesn't. The only router that has a 1st-party firmware supporting fq_codel so far is the Ubiquiti Edgerouter, and it isn't enabled by default. 3rd-party open-source firmwares like OpenWrt do support fq_codel).

See Bufferbloat for more information.

In a home environment: Only if someone is constantly fully maxing out the connection and you are seeing lag in games.

Otherwise, no I would not implement QoS in a home environment 99% of the time.

If you have a working qos system (such as the ones in openwrt, cerowrt, ipfire, edgerouter, and multiple other fq_codel enabled systems) I'd recomend having it on all the time. I note that the first shipping 1st party fq_codel enabled product is actually streamboost, which is quite good, I'm told, and in multiple high end gaming routers today. In all cases you have to do a measurement and then derive the correct settings.

Peak loads happen all the time. A single web page load can saturate your connection, briefly, and mess up your interactive traffic. A systained load, such as that from a big upload or things from bittorrent, will also mess up your interactive traffic badly. Even a mild amount of jitter induced by other traffic can make a game unplayable.

As for measuring with ping, ping is just a proxy for (dns,voip,new flows, etc) getting a feel for how well interactive traffic will perform under whatever other load you are running with. Holding jitter of ping-like packets below 30ms is needed for good voip and gaming performance. Increased latency leads to talking overrunning each other, with about 250ms total latency being worst case desirable performance for voip in particular (you start stepping on each other's conversation in an otherwise natural pause).

The classic "bufferbloat" test is one flow up, one down, and one ping. We do a variety of more complex and useful test in the rrul test suite which is available as part of "netperf-wrapper" on github.

As for examples of the benefits of using a good aqm/packet scheduling system such as those in the products above:

https://www.bufferbloat.net/projects/codel/wiki/RRUL...

I believe my BQM shows fq_codel in action.

Yesterday my speeds were set incorrectly in the QOS scripts (a holdover from when my line sync was much higher than it is now), such that they exceeded my actual speed.

At a little after 6pm i downloaded 2GB of files. You can see my average latency spiked from 25ms to 70ms, with maximum latency hitting 110ms.

That evening i adjusted my speeds in the QOS settings to 90% of actual speed.

This afternoon i downloaded 4GB of files and you can see the difference. Average latency only increased to 30ms, with maximum latency never exceeding 70ms.

And this is with the default OpenWrt qos-scripts package. I understand you submitted a patch to include Cerowrt's SQM stuff in OpenWrt with improvements which would be particularly useful for UK users (our FTTC/FTTP ISPs use PPPoE almost exclusively), i hope it makes it in before the BB freeze.

Big fan of all the work you and the other Cerowrt developers have done in making the Internet responsive as well as pushing forward things like IPv6 and DNSSEC in home routers. Thanks!

Do you have a "before" BDM? I realize that life without fq_codel is often annoying, but spending 24 hours without it in the name of science shouldn't be too bad. Testing before/after with videoconferencing with up/downloads going on is worthwhile...

The sqm system can be fairly easily overlaid over any openwrt. The package is all scripts, so you can install the latest "binary" package from the cerowrt distribution. It probably has some dependencies on other kernel modules (so if you try to run it and you get errors, you are missing some other package)

Yes, I hope to push sqm up to openwrt (or get qos-scripts further improved), but time has been lacking of late. Volunteers to get it fully openwrtable desired... there is a ton of code related to doing experiements with other systems in there, that can get ripped out, etc. I found a bug with dscp squashing today...

A problem we have is most of the low end routers have insufficient cpu to shape at much higher than 50mbits, and thus we keep looking for devices (such as the edgerouter) that can get past 100. More and more of these are appearing, but it's still early days.

Lastly:

Thank you for your interest and enthusiasm! If it wasn't for the enthusiasm of everyone that tries this stuff (and gets their config right), I'd have gone off to doing something else long ago.

Edited by deleted (Sun 13-Jul-14 15:59:57)

In reply to a post by dtaht:

Do you have a "before" BDM? I realize that life without fq_codel is often annoying, but spending 24 hours without it in the name of science shouldn't be too bad. Testing before/after with videoconferencing with up/downloads going on is worthwhile...

I don't have any before fq_codel BQMs (only been using the BQM for the past week, and fq_codel for the last year), but i'll add it to my todo list.

I noticed that the BQM "snapshot" i previously posted actually didn't "snap" at that point in time and continued on until 00:00, thus moving the relevant bit off the edge of the graph.

Here's a BQM from when my settings were incorrect. Average latency spikes to 70ms from the baseline of 25ms at around 06:00 and 18:00 during large downloads.

Here's a BQM after i adjusted my settings. At first the QOS settings were around 95% of actual speed, and a large (4GB) download occurred at around 13:00. Average latency only increased to around 32ms. I then decreased the settings to around 90% of actual speed and large downloads at 18:00 and 19:30 saw the average latency only increase to around 28ms.

A problem we have is most of the low end routers have insufficient cpu to shape at much higher than 50mbits, and thus we keep looking for devices (such as the edgerouter) that can get past 100. More and more of these are appearing, but it's still early days.

Have you looked into the TP-Link WDR4900? It's running a PPC processor rather than the more common MIPS and mine seems pretty beefy.

I haven't done any shaping tests, but when i bought it last year i did a few basic iperf tests. With the default OpenWrt config of DHCP+NAT it did 394Mbps WAN-to-LAN (416Mbps after splitting the wired and wireless into separate networks).

I do plan to do some more testing the next time i update my router (probably when BB final gets released), so i'll add some shaping tests to my todo list too.

Ran without fq_codel all day yesterday and here's the BQM. Large latency spikes whenever downloads take place.

Running with fq_codel-based QOS makes a definite improvement in the responsiveness of my connection, no matter what i'm doing.