TC(8)

🎀 NAME 🎀

okay so the manual page for tc (the traffic control tool) has a whole section on this one: drr - deficit round robin scheduler 💖

🧚 SYNOPSIS 🧚

tc qdisc ... add drr [ quantum bytes ]

💅 DESCRIPTION 💅

okay bestie, the Deficit Round Robin Scheduler is a classful queuing discipline 👑 and it's giving a more flexible upgrade to Stochastic Fairness Queuing (SFQ) ✨

unlike SFQ, there are no built-in queues 🙈 you have to add your own classes and then set up filters so packets get sorted into them. this is lowkey super useful if you want, say, RED qdiscs with different settings for specific traffic 🦋 also, there is no default class. if a packet can't be classified, it gets dropped 💀

🌈 ALGORITHM 🌈

every class gets its own deficit counter, and it starts out equal to quantum 🫶

DRR keeps an (internal) active list of classes whose qdiscs are non-empty, and this list is what drives dequeuing 🔥 a packet gets dequeued from the class sitting at the head of the list if the packet size is smaller than or equal to the deficit counter. if the counter is too small, it gets bumped up by quantum and the scheduler moves on to the next class in the active list. the way that it just keeps cycling, no cap, iconic 💜

🍓 PARAMETERS 🍓

• quantum — the number of bytes a flow is allowed to dequeue before the scheduler moves on to the next class 🎀 it defaults to the MTU of the interface, and the minimum value is 1 ✨

🦋 EXAMPLE & USAGE 🦋

okay so to attach to device eth0 and use the interface MTU as the quantum:

# tc qdisc add dev eth0 handle 1 root drr

next up, adding two classes 🌸

# tc class add dev eth0 parent 1: classid 1:1 drr
# tc class add dev eth0 parent 1: classid 1:2 drr

you also need to add at least one filter to classify packets, otherwise nothing gets sorted 🙈

# tc filter add dev eth0 protocol .. classid 1:1

bestie, real talk: like SFQ, DRR is only useful when it owns the queue 🚨 it's a pure scheduler and does not delay packets. attaching non-work-conserving qdiscs like tbf to it does not make sense, because other qdiscs in the active list will also go inactive until the dequeue operation succeeds. to make sure it owns the queue, embed DRR within another qdisc like HTB or HFSC 💖

if you want to mimic SFQ behavior, you can assign packets to the attached classes using the flow filter 🧚

tc qdisc add dev .. drr

for i in .. 1024;do
     tc class add dev .. classid $handle:$(print %x $i)
     tc qdisc add dev .. fifo limit 16
done

tc filter add .. protocol ip .. $handle flow hash keys
src,dst,proto,proto-src,proto-dst divisor 1024 perturb 10

📚 SOURCE 📚

• M. Shreedhar and George Varghese "Efficient Fair Queuing using Deficit Round Robin", Proc. SIGCOMM 95. 🫶

🙈 NOTES 🙈

heads up, this implementation does not drop packets from the longest queue on overrun 😭 that's because limits get handled by the individual child qdiscs instead ✨

🔗 SEE ALSO 🔗

tc(8), tc-htb(8), tc-sfq(8) 💜

👑 AUTHOR 👑

sched_drr was written by Patrick McHardy, who is absolutely iconic 🌈

bestie, that's the whole thing! 💖 this page is from iproute2, dated January 2010, and it's the TC(8) manual 🎀 stay fair, stay fast, and keep those queues flowing 💅🫶✨