How to gets less than 5ms in bufferbloat test? (PPPoE, SQM)


does distence between me and server affect the bufferbloat test ?
and how to calculate Per Packet Overhead (byte) in sqm and what overhead i can put in cake option my wan is pppoe-wan and i'm using the veth0 that handle downloads traffic
my openwrt is x86 with cpu is x86 intel core quad q9550 is that enough

I would say that depends on the test, but likely yes. Longer paths typically mean more data in flight and a less reactive control loop, which both can contribute to bufferbloat.

It is always the same, you need to configure the overhead such, that your shapers gross speed and overhead are a conservative model of the true bottleneck's gross rate and applicable overhead, conservative here means, that from a bufferbloat minimization perspective it is fine to under-estimate the bottleneck gross rate and to over-estimate the bottleneck per-packet-overhead. So these both typically depend on your specific bottleneck link and hence it is hard to give precise guidance from far-away, but the OpenWrt SQM detail wiki offers some guidance.

But to the circle of your question, SQM's AQM, cake and/or fq_codel are by default configured for a delay target of 5ms, so if you absolutely require below 5ms average delay, you probably need to decrease target and interval for fq_codel or rtt for cake (that is equivalent to fq_codel's interval, cake does not expose target explicitly to user-space). Keep in mind when doing so SQM will work fine for short RTT paths, but longer paths will see less throughput than expected.

1 Like