[Cake] Query regarding COBALT implementation in CAKE

[Cake] Query regarding COBALT implementation in CAKE

[Shefali Gupta]

[-- Attachment #1: Type: text/plain, Size: 406 bytes --]

Hello Jonathan and Dave,

We were trying to look into Linux Kernel CAKE code to understand COBALT
implementation.

We are unable to understand how over_target variable is being used to
change the dropping state, on the line number 535 of CAKE code.

We are referring to this CAKE code:
https://elixir.bootlin.com/linux/latest/source/net/sched/sch_cake.c

Thanks and regards,
Shefali Gupta
Jendaipou Palmei

[-- Attachment #2: Type: text/html, Size: 968 bytes --]

Re: [Cake] Query regarding COBALT implementation in CAKE

[Jonathan Morton]

> On 22 Jan, 2019, at 10:29 am, Shefali Gupta <shefaligups11@gmail.com> wrote:
> 
> We are unable to understand how over_target variable is being used to change the dropping state, on the line number 535 of CAKE code.

A little earlier in the code, you can see how over_target is calculated.  It's simply a boolean decision:

	over_target = sojourn > p->target &&
		      sojourn > p->mtu_time * bulk_flows * 2 &&
		      sojourn > p->mtu_time * 4;

There are three decision terms, all of which compare the sojourn time of the current packet to some threshold, and all of which must be true for over_target to be true, and thus for the Codel logic to enter dropping state.  In principle, it could be replaced by "sojourn > min(p->target, p_mtu_time * min(4, bulk_flows * 2))", but that would be harder to lay out readably in the Linux kernel's 80-column style.

The middle term is not relevant for your implementation, because you have a single-queue version of COBALT; you could consider bulk_flows as being fixed at 1, so that term is always true when the last term is true.

The last term depends on p->mtu_time which is initialised whenever Cake's shaper is reconfigured.  It refers to the time required to stuff an MTU-sized packet down the pipe.  Since your implementation is not closely integrated with a shaper and therefore you don't have an easy way to calculate mtu_time, you may eliminate this term, and instead manually configure the target to satisfy the same criterion.

You can therefore simplify if(over_target) to if(sojourn > p->target) for your purposes.

 - Jonathan Morton

Re: [Cake] Query regarding COBALT implementation in CAKE

[Jonathan Morton]

> On 22 Jan, 2019, at 11:04 am, Jonathan Morton <chromatix99@gmail.com> wrote:
> 
> In principle, it could be replaced by "sojourn > min(p->target, p_mtu_time * min(4, bulk_flows * 2))"

Correction: sojourn > max(p->target, p_mtu_time * max(4, bulk_flows * 2))

 - Jonathan Morton