What is "per packet overhead"

hello everyone
I'm using nano station m5 using pppoe (username+password) working in router mode and connected via cat6 cable to Archer router (openwrt) through wan port
running sqm but i don't know what i should put in the per packet overhead field !!!
need help please

I started at 8, then went to 34 for ptm, then 44 for cable, then 50 for VLAN adder. I am back to 8 for pppoe, for PPO.

1 Like

This is almost guaranteed to be too low :wink:

Have a look at the discussion in https://openwrt.org/docs/guide-user/network/traffic-shaping/sqm-details#sqmlink_layer_adaptation_tab. We generally recommend to err on the side of too large, which for most users not using a VPN means that 44 Bytes is a decent starting point (case in point, I use 50 Bytes, but I have to deal with a specific situation of rate limiting fast-ethernet with VLAN and PPPoE, so 38+4+8=50, because my vdsl2-modem only has fast ethernet, but the dsl sync rate is 116 Mbps, which in turn means the true ethernet-overhead becomes relevant and con not be ignored anymore).
The link above contains a pointer how to use smaller packet sizes to confirm whether the selected per-packet-overhead is okay. If that leaves questions, please ask here.


I get fairly good results with all the values, around 41msec bufferbloat on dslreports; it seems to be dominated by a good egress value. I haven't yet pinned down the best PPO figure. I'll set PPO back to 34. My sync is around 26 Mbps for a 25/10 connection; it will be going up next week to maybe 52 for a 50/10 connection. I had the RG501 modem untag the packets from VLAN 35 off the Bell line. There are a few config pages for DSL on the RG501 dealing with VPI/VCI; I haven't found any instructions how to set this up yet.

1 Like

I fully believe that, as the link discusses (well the text behind the link...) for any given packet size one can make up for an underestimation of either shaper rate or PPO by a complimentary overestimation of the other. Speedtests (especially unidirectional ones) typically use packets as large as the network path allows, which is okay in that most bulk transfers will do the same, but it does not help figuring out whether the configured PPO is safe at all packet sizes. The simplest solution is to restrict the packet-size (e.g. by MSS clamping) and to repeat a bufferbloat test, if PPO was too small bufferbloat will noticeably increase.... (There is one complication though, cake will also show increased latency under load if cake does not get enough CPU cycles when it needs to, so either perform these tests with HTB=fq_codel (which will show less utilization instead of more latency) and/or monitor the CPU load on the router somewhow during the test).

These rates indicate that this can not be ADSL[1|2|2+] anymore, and since VDSL1 is exceedingly rare it must be VDSL2. And that gives us a theoretical likely lower bound estimate of the overhead of 34 bytes with a shaper instantiated on pppoe-wan.

And VPI/VCI are only important/relevant for ADSL/ATM, so I guess you can safely ignore thosse with your vdsl link.

1 Like

I'm pretty sure it is VDSL2. The SR501 doesn't have a VPI/VCI setting for ptm. For the WAN bridge service, however, VPI/VCI shows up for an untagging option on the SR501. If I don't untag the packets on the modem, I can use a Banana Pi or Raspberry Pi to untag them by using eth0.35 as the WAN interface. It is nice to see that a VCI of 35 from the modem translates to a VLAN of 35 in the router.

so my friend in my case what should i put in PPO field ?

Just try 44 at first. If you get good performance... 95% chance you're done

1 Like