Re: [PATCH 5/5] tc: man: add man page for drr scheduler

[Patrick McHardy <kaber@trash.net>]

Florian Westphal wrote:

Thanks Florian. A few minor suggestions below.

> diff --git a/man/man8/tc-drr.8 b/man/man8/tc-drr.8
> new file mode 100644
> index 0000000..5fcb0eb
> --- /dev/null
> +++ b/man/man8/tc-drr.8
> @@ -0,0 +1,89 @@
> +.TH TC 8 "January 2010" "iproute2" "Linux"
> +.SH NAME
> +drr \- deficit round robin scheduler
> +.SH SYNOPSIS
> +.B tc qdisc ... add drr
> +.B [ quantum
> +bytes
> +.B ]
> +
> +.SH DESCRIPTION
> +
> +The Deficit Round Robin Scheduler is a classful queuing discipline as
> +a more flexible replacement for Stochastic Fairness Queuing.
> +
> +Unlike SFQ, there are no built-in queues \-\- you need to add classes
> +and then set up filters to classify packets accordingly.
> +This can be useful e.g. for using RED qdiscs with different settings for particular
> +traffic. There is no default class \-\- if a packet cannot be classified,
> +it is dropped.
> +
> +.SH ALGORITHM
> +Each class is assigned a deficit counter, initialized to
> +.B quantum.
> +
> +DRR maintains an (internal) ''active'' list of classes whose qdiscs are
> +non-empty.  This list is used for dequeuing.  A packet is dequeued from
> +the class at the head of the list if the packet size is smaller or equal
> +to the deficit counter.  If the counter is too small, it is increased by
> +.B quantum
> +and the scheduler moves on to the next class in the active list.
> +
> +
> +.SH PARAMETERS
> +.TP
> +quantum
> +Amount of bytes a flow is allowed to dequeue before the scheduler moves to the next.

next class or next flow?

> +Defaults to the MTU of the interface. The minimum value is 1.
> +
> +.SH EXAMPLE & USAGE
> +
> +To attach to device eth0, using the interface MTU as its quantum:
> +.P
> +# tc qdisc add dev eth0 handle 1 root drr
> +.P
> +Adding two classes:
> +.P
> +# tc class add dev eth0 parent 1: classid 1:1 drr
> +# tc class add dev eth0 parent 1: classid 1:2 drr
> +.P
> +You also need to add at least one filter to classify packets.
> +.P
> +# tc filter add dev eth0 protocol .. classid 1:1
> +.P
> +
> +Like SFQ, DRR is only useful when it owns the queue \-\- it is a pure scheduler and does
> +not delay packets.  Attaching non-work-conserving qdiscs like tbf to it does not make
> +sense \-\- other qdiscs in the active list will also become inactive until the dequeue
> +operation succeeds.  Embed DRR within another qdisc like HTB or HFSC to ensure it owns the queue.
> +.P
> +You can mimic SFQ behavior by assigning packets to the attached classes using the
> +hash filter:

flow filter

> +
> +.B tc qdisc add dev .. drr
> +.B for i in .. 1024;do
> +.B   tc class add dev ..  classid $handle:$(print %x $i)
> +.B   tc qdisc add dev .. fifo limit 16
> +.B done
> +.B tc filter add .. protocol ip .. $handle flow hash keys src,dst divisor 1024 perturb 10

To mimic SFQ, the filter needs to use "keys
src,dst,proto,proto-src,proto-dst".

> +
> +
> +.SH SOURCE
> +.TP
> +o
> +M. Shreedhar and George Varghese "Efficient Fair
> +Queuing using Deficit Round Robin", Proc. SIGCOMM 95.
> +
> +.SH NOTES
> +
> +This implementation does not drop packets from the longest queue on overrun,
> +as limits are handled by the individual child qdiscs.
> +
> +.SH SEE ALSO
> +.BR tc (8),
> +.BR tc-htb (8),
> +.BR tc-sfq (8)
> +
> +.SH AUTHOR
> +sched_drr was written by Patrick McHardy.
> +