Sqm Overhead for Fiber

Installing and Using OpenWrt
1

Hello everyone, so i need help/clarification on setting up SQM for my conncetion. Ive read alot about the issue on this forum but theres no one definitive answer. What overhead should i choose?

Conncetion details
ISP: PTCL Flash Fiber
Country: Pakistan
Technology: FTTH GPON
Speed: 100mbps (i get roughly 110-115 download and 125-130mbps upload)

Setup
The actual fiber lands up in my home and gets connected to the ONT on which ive setup bridge mode and a vlan 10 with priority 0. The bridged port then gets connected to my openwrt router which does the pppoe dialup.
Ive have the cake bandwidth limits set at 92000kbps for both directions.

the overhead confusion
These are the tests that i did for deductions however i feel like they all just work? The only difference i feel is in game (i play pubg mobile) and i feel like the bullet connection get altered a bit based on overhead value.

  1. No Overhead: DL peak 3.97 ms, UL peak 6.09 ms, DL jitter 0.31 ms, UL jitter 0.61 ms, speeds 78.2↓ / 84.3↑ Mbps
  2. Ethernet 12: DL peak 3.66 ms, UL peak 4.12 ms, DL jitter 0.35 ms, UL jitter 0.38 ms, speeds 77.4↓ / 84.2↑ Mbps
  3. Ethernet 25: DL peak 3.17 ms, UL peak 5.43 ms, DL jitter 0.27 ms, UL jitter 0.55 ms, speeds 77.9↓ / 82.9↑ Mbps
  4. Ethernet 35: DL peak 3.72 ms, UL peak 7.17 ms, DL jitter 0.32 ms, UL jitter 1.02 ms, speeds 77.4↓ / 80.4↑ Mbps
  5. Ethernet 39: DL peak 3.17 ms, UL peak 6.57 ms, DL jitter 0.26 ms, UL jitter 0.93 ms, speeds 77.0↓ / 81.0↑ Mbps
  6. Ethernet 44: DL peak 2.94 ms, UL peak 7.49 ms, DL jitter 0.24 ms, UL jitter 0.85 ms, speeds 76.7↓ / 81.5↑ Mbps
  7. ATM 44 : DL peak 2.80 ms, UL peak 4.35 ms, DL jitter 0.23 ms, UL jitter 0.51 ms, speeds 69.5↓ / 75.8↑ Mbps
  8. Ethernet 50: DL peak 3.10 ms, UL peak 7.13 ms, DL jitter 0.23 ms, UL jitter 0.78 ms, speeds 76.4↓ / 80.9↑ Mbps

(5 speed tests were performed for each overhead value and the results above are an average of the same)

2

For GPON + PPPoE + VLAN I would avoid ATM; that is for DSL and it explains the big speed drop. A practical starting point is Ethernet with an overhead around 34–44 bytes, then tune the bandwidth rather than chasing tiny jitter differences between overhead values. Your latency numbers are already very close, so I’d keep the setting that gives stable latency and acceptable throughput, and leave some headroom below the real 110/125 rates.

3

IIRC it's 5-10%

4

I would not read too much into PUBG feel from the overhead number alone. Overhead mainly makes CAKE account bytes correctly; if it is a bit high you lose some throughput, if it is too low you can still get queueing under load. Pick ethernet, not ATM, then test with a bufferbloat test while the line is busy. If 92/92 is stable, you can try raising upload/download in small steps until latency starts climbing, then back off a little.

5

This is what I could never understand: SQM expects "physical link" throughput while speed tests measure "goodput" and then we are recommended to take 80..90% of "good put" and put it in place that expects "physical link" throughput. It is like apples and oranges to me.

I have a similar connection (GPON+PPPoE+VLAN) with higher speeds (533/533 MBps as reported by speedtest-go). My SQM is configured with higher values and I am getting consistent 4..5 ms pings to 8.8.8.8 with download/upload saturated.

I tested without SQM, and I am seeing 100+ ms latency, so SQM really helps. I am getting solid 500/500 MBps with SQM.

config queue 'eth1'
	option enabled '1'
	option interface 'pppoe-wan'
	option download '550000'
	option upload '550000'
	option qdisc 'cake'
	option script 'piece_of_cake.qos'
	option qdisc_advanced '1'
	option ingress_ecn 'ECN'
	option egress_ecn 'ECN'
	option qdisc_really_really_advanced '1'
	option itarget 'auto'
	option etarget 'auto'
	option linklayer 'ethernet'
	option use_mq '0'
	option overhead '56'
	option squash_dscp '1'
	option squash_ingress '1'
	option iqdisc_opts 'dual-dsthost nat ingress mpu 84'
	option eqdisc_opts 'dual-srchost nat mpu 84'

Anyone know what I am missing here?

6

5 to 6 ms is the increase in latency from idle or what? If so, try to reduce the throughput to 480000. Why did you choose that overhead value, any specofoc reason? Also, no need to specify the nat option as these days, it is automatically applied.

7

No increase. It stays constant no matter the load.

I found it here on this forum for "FTTH/PPPoE/VLAN" setups, but in the end what mattered more is how I came up with SQM limits. I just kept on increasing download/upload until latency spiked under load and then took 80..90% of that. Works like a charm now.

8

I was reading online and found a page that says for gpon the minimum packet size is 53 and the maximum is 1518. Doesn't that make mpu 53 and overhead 26 (1518-1492 = 26)

The smallest packet size is 53 bytes, and the largest is 1518 bytes.
https://www.simplifinetworks.com/blog/the-basics-of-pon-epon-gpon