Need to throttle bandwidth to Guest/Shared WIFI (eth0.4), retain all bandwidth for personal WIFI (eth0.1). eth1 is WAN has no associated interface.
I picked up below code from config file (WINSCP).
Appreciate experts to check the below SQM setup is correct. Presently the DSLreport I can see huge drop in internet speed as the graph develops. Youtube buffer bloat happens some of the days.
The average net downlaod speed is around 45mbps though ISP is 120mbps (downlload).
Upload speed provided by ISP is 40 mbps
Help my understand this, did you hook up two different APs (one for guest and one for private) to different ports of your router and use VLANs to differentially shape traffic by VLAN, or did you create two SSIDs on a combined wifi router?
Instantiate SQM on the true wan interface with correct settings for your link, for both upstream and downstream (with the appropriate bell's and whistles), to handle ISP bufferbloat. You could simply try per-internal-IP-fairness, which generally works well for home networks (and if you have few enough guest users, might already be good enough, as no guest will be able to monopolize all bandwidth).
Instantiate additional SQM instance(s) for the guest networks. If all guest come via wlan1.1, I would try to instantiate SQM on that interface, also, I would configure SQM for both upstream and downstream (and also for per-internal-IP fairness, to treat all guests equally). Instantiating a shaper on a bridge is always a bit dicy, especially if you want differential speeds for up and downlink.
no dedicated shaper for the private WLAN, unless you want to be generous to your guests (but then you might be better off, with just SQM on WAN and per-internal-IP fairness).
If I would see the guest traffic as a real problem, I would probably use a dedicated secondary router for the guests, with NAT, in which case all guest traffic will look like coming from one internal IP, and hence all guests will only get 1/(total number of internal hosts active) of the available internet speed.
I would recommend you try 1 and see whether that is not already good enough and only go to 2 if you are unhappy
Thanks Moeller for quick response
Will try the options suggested and monitor performace for a week or so.
Learning from all experts like you and this forum in particular, I am no way into networking or IT, just a passion to experiment. Appreciate your knowledge and good luck!
What if the instantiation is on VLAN say eth0.3 or eth0.4. Is it same issues as a bridge.
Without going of option 4, possibly a modified option 2 with a secondary router. Just a calrification for my information.
"
BTW implemented option 1 on WAN eth1. With "per-internal-IP-fairness" and "link layer set as = ethernet overhead 44pckt.
Thanks, will monitor it for couple of days.
In theory that should work, but there is another thread here that seems to indicate that there might be issues in sqm's hotplug handling for vlan interfaces...[quote="ranish, post:6, topic:74456"]
Without going of option 4, possibly a modified option 2 with a secondary router. Just a calrification for my information.
[/quote]
Sure, in that case the secondary router would be acting as AP and not do NAT, you could easily instantiate sqm on the APs port that you connect to your main router.
If you post the output of tc -s qdisc and cat /etc/config/sqm I will try to offer comments on potential changes/improvements.
Let's start with that and only go into the others if you decide to switch
/etc/config/sqm looks decent, IMHO the only things worth looking at are:
The bandwidth settings, what values do you robustly and repeatble measure against a close by speedtest server (maybe your ISP's or ookla's speedtest.net, and netflix's fast.com). As first approximation, I would set the achivable net goodput of a reliable speedtest as the gross speed for the shaper.
Encapsulation
That looks like there is some additional encapsulation in play, typically the 1440 value should be 1500 for unadorned ethernet, or 1492 if PPPoE is used, no idea what your ISP uses these 60 bytes for... (this is relevant, as you might need to account for that depending on how your ISP controlls your access speed). Could you post the content of the "Share Your Results" box of https://www.speedguide.net/analyzer.php to get some idea about the encapsulation? Question: Does your ISP also offer IPv6 and what IPv4 address gets assigned to your wan interface? Something starting with 10. ?
The minimal sizes:
indicate that on that interface you see smaller packets than likely can/will be used, so please add the following stanzas to your /etc/config/sqm for option on:
Mmmh, okay that does not look like an IPv6 tunnel is used by your ISP. But the smaller than expected MTU/MSS to remote sites like speedguide indicates a tunnel somewhere along the path but a tunnel with apparently working pMTU discovery, so very unlikely to be your problem....
@ranish, it seems you found a solution that you are happy with, great!
Could I convince you to sort of close this thread with a short description of your initial challenge and the solution you opted for in the end? That way others finding this thread will have an easier time learning from your experience. Thanks!
This solution worked. i.e have the SQM only in eth1 (Wan). No noticable bufferbloat observed.
By instantiating on wlan 1-1 (guest WiFi) the throttling perculated even to my private WiFI on Wan1. So as suggested by Moeller0 this is indeed not producing desired dividents.
Did not try this option yet, will try in coming weeks and report the findings.
Yes, this should work for keeping wan bufferbloat under control, glad it works. Whether that is sufficiently fair to also solve your guest network issue is another question, but it should at least improve things over the starting condition to buy you time to research alternative options.
I am always amazed how well cake's per-internal-IP fairness mode actually behaves. Sure it is not a perfect match for everybody, but it avoids the rather annoying condition that one/a few unresponsive users/applications can make an internet klink pretty much unusable for the rest.
It should help, especially when parents have online meetings + simultaneouly kids have online classes and the guest had unwittingly let the torrents flow into his dirty pc. Had video meeting disruptions in the past, hope that is settled once and for all.
