info > CAKE(8)

man perldoc info pydoc3 ri search

📛 NAME

CAKE - Common Applications Kept Enhanced (CAKE)

🚀 Quick Reference

Use Case	Command	Description
Basic egress shaping with Ethernet overhead	tc qdisc add dev eth0 root cake bandwidth 100Mbit ethernet	Shape outbound to 100 Mbit/s, compensating for Ethernet framing
Ingress autorate (cellular)	tc qdisc add dev eth0 ingress cake autorate-ingress	Let CAKE estimate link capacity dynamically from incoming traffic
Set RTT for satellite	tc qdisc add dev eth0 root cake bandwidth 1Mbit satellite	Configure AQM for high-latency satellite link
Dual-srchost flow isolation (LAN egress)	tc qdisc add dev eth0 root cake bandwidth 10Mbit dual-srchost	Prevent one host from monopolising uplink regardless of flow count
Wash untrusted diffserv markings	tc qdisc add dev eth0 root cake bandwidth 100Mbit wash	Clear QoS bits (but not ECN) after priority queuing
Show real-time statistics	tc -s qdisc show dev eth0	Display per-tin stats, drops, delays, and flow counts

📋 SYNOPSIS

tc qdisc ... cake
  [ bandwidth RATE | unlimited* | autorate-ingress ]
  [ rtt TIME | datacentre | lan | metro | regional | internet* | oceanic
    | satellite | interplanetary ]
  [ besteffort | diffserv8 | diffserv4 | diffserv3* ]
  [ flowblind | srchost | dsthost | hosts | flows | dual-srchost | dual-dsthost
    | triple-isolate* ]
  [ nat | nonat* ]
  [ wash | nowash* ]
  [ split-gso* | no-split-gso ]
  [ ack-filter | ack-filter-aggressive | no-ack-filter* ]
  [ memlimit LIMIT ]
  [ fwmark MASK ]
  [ ptm | atm | noatm* ]
  [ overhead N | conservative | raw* ]
  [ mpu N ]
  [ ingress | egress* ]
  (* marks defaults)

📘 DESCRIPTION

CAKE (Common Applications Kept Enhanced) is a shaping‑capable queue discipline which uses both AQM and FQ. It combines COBALT (Codel + BLUE), a deficit‑mode shaper, and a variant of DRR++ for flow isolation. 8‑way set‑associative hashing virtually eliminates collisions. Simplified diffserv priority queuing is available, and overhead compensation for encapsulation schemes is tightly integrated.

All settings are optional; defaults are sensible for most deployments. Most people only need to set the bandwidth parameter, but reading the Overhead Compensation and Round Trip Time sections is strongly encouraged.

🔧 SHAPER PARAMETERS

CAKE uses a deficit‑mode shaper (no initial burst). It automatically bursts just enough to maintain the configured throughput.

• 🔓 unlimited (default) – No bandwidth limit.
• 🔢 bandwidth RATE – Set the shaper bandwidth. See tc(8) for RATE syntax.
• 🚦 autorate-ingress – Automatic capacity estimation based on incoming traffic. Useful for variable‑rate links (e.g., cellular). An initial bandwidth can be used as a starting estimate. Cannot estimate bandwidth of downstream links.

📦 OVERHEAD COMPENSATION PARAMETERS

The packet size Linux sees may differ from the wire size. These options make CAKE internally treat each packet as larger, accounting for encapsulation overheads. Keywords exist for common link technologies.

📐 Manual Overhead Specification

• 📦 overhead BYTES – Adds (or subtracts) bytes to each packet. Range: -64 to 256.
• 📏 mpu BYTES – Rounds each packet (with overhead) up to a minimum length. Range: 0 to 256.
• 📠 atm – Compensate for 53‑byte ATM cell framing (ADSL). Performed after the overhead parameter.
• 📡 ptm – Compensate for PTM encoding (VDSL2, 64b/65b).
• 🔕 noatm – Disable both ATM and PTM compensation.

⚡ Failsafe Overhead Keywords

Quick‑and‑dirty settings if you don’t want to read further.

• 🌿 raw (default) – No overhead compensation; Linux packet size used directly. Other overhead keywords can be added after raw to work relative to the reported size.
• 🛡️ conservative – Adds plenty of overhead (48 bytes + ATM) suitable for most real‑world links. Equivalent to overhead 48 atm.

📡 ADSL Overhead Keywords

All implicitly enable atm. Check your modem’s framing settings.

• pppoa-vcmux – equivalent to overhead 10 atm
• pppoa-llc – equivalent to overhead 14 atm
• pppoe-vcmux – equivalent to overhead 32 atm
• pppoe-llcsnap – equivalent to overhead 40 atm
• bridged-vcmux – equivalent to overhead 24 atm
• bridged-llcsnap – equivalent to overhead 32 atm
• ipoa-vcmux – equivalent to overhead 8 atm
• ipoa-llcsnap – equivalent to overhead 16 atm

📶 VDSL2 Overhead Keywords

• pppoe-ptm – equivalent to overhead 30 ptm
  PPPoE: 2B PPP + 6B PPPoE + Ethernet: 6B dst MAC + 6B src MAC + 2B ethertype + 4B FCS + PTM: 1B SoF (S) + 1B EoF (Ck) + 2B TC‑CRC
• bridged-ptm – equivalent to overhead 22 ptm
  Ethernet: 6B dst + 6B src + 2B ethertype + 4B FCS + PTM: 1B S + 1B Ck + 2B TC‑CRC

📺 DOCSIS Cable Overhead Keyword

• docsis – equivalent to overhead 18 mpu 64 noatm. Matches the head‑end’s Ethernet‑frame metering.

🌐 Ethernet Overhead Keywords

• ethernet – Accounts for preamble, inter‑frame gap, and FCS. Equivalent to overhead 38 mpu 84 noatm.
• ether-vlan – Adds 4 bytes for an IEEE 802.1Q VLAN header. Can be repeated for multiple VLAN tags.

⏱️ ROUND TRIP TIME PARAMETERS

AQM in CAKE (COBALT) responds to queue occupancy, but response speed must match actual RTT. Very short RTT presets can starve throughput; use metro for local LANs unless you have a custom kernel.

• 🔧 rtt TIME – Manual RTT in time units.
• 🏢 datacentre – ≈ 100 µs. For 10GigE+ only.
• 🏠 lan – ≈ 1 ms. Pure Ethernet, not Wi‑Fi. Not for Internet access links.
• 🏙️ metro – ≈ 10 ms. Traffic within a single city.
• 🗺️ regional – ≈ 30 ms. A European‑sized country.
• 🌍 internet (default) – ≈ 100 ms. Suitable for most Internet traffic.
• 🌏 oceanic – ≈ 300 ms. Higher‑latency regions (e.g., Australia).
• 🛰️ satellite – ≈ 1000 ms. Geostationary satellite links.
• 🚀 interplanetary – ≈ 3600 s. Nearly disables AQM (Jupiter is 1 light‑hour away).

🔀 FLOW ISOLATION PARAMETERS

Packets from different flows are isolated in separate queues, each with its own AQM state. Set‑associative hashing minimises collisions. Choose fairness based on source, destination, 5‑tuple, or layered isolation.

• 🔓 flowblind – Disable flow isolation; single queue per tin.
• ⬆️ srchost – Flows defined by source address. Useful on ISP egress.
• ⬇️ dsthost – Flows defined by destination address. Useful on ISP ingress.
• 👥 hosts – Flows defined by source‑destination host pair (host isolation, not per‑flow).
• 🌊 flows – 5‑tuple isolation (src IP, dst IP, protocol, src port, dst port). Like SFQ / fq_codel.
• 🔼 dual‑srchost – Fairness first over source addresses, then per‑flow. Prevents one LAN host from hogging uplink.
• 🔽 dual‑dsthost – Fairness first over destination addresses, then per‑flow. Prevents one LAN host from monopolising downlink.
• ✨ triple‑isolate (default) – Fairness over source and destination addresses intelligently, then per‑flow. Works for both upstream and downstream.
• 🌐 nat – Perform NAT lookup before applying flow‑isolation rules to see true inside addresses/ports. Improves fairness behind NAT.
• 🚫 nonat (default) – No NAT lookup; uses addresses/ports visible on the interface.

🎚️ PRIORITY QUEUE PARAMETERS

Traffic is divided into “tins” based on Diffserv field. Each tin has independent flow‑isolation queues and is serviced via WRR with priority‑sharing thresholds.

• ⚖️ besteffort – All traffic in one tin, no priority queuing.
• 🔲 precedence – Legacy TOS “Precedence” interpretation. Discouraged on modern Internet.
• 🎬 diffserv4 – 4 tins: Bulk (CS1, LE – 6.25% threshold), Best Effort (100%), Video (AF4x, AF3x, CS3, AF2x, CS2, TOS4, TOS1 – 50%), Voice (CS7, CS6, EF, VA, CS5, CS4 – 25%).
• 🎯 diffserv3 (default) – 3 tins: Bulk (CS1, LE – 6.25%), Best Effort (100%), Voice (CS7, CS6, EF, VA, TOS4 – 25% with reduced Codel interval).
• 🏷️ fwmark MASK – Overrides tin selection via firewall marks. The masked & shifted result becomes the tin number. Allows firewall‑controlled policies.

🛠️ OTHER PARAMETERS

• 💾 memlimit LIMIT – Limit queue memory to LIMIT bytes. Not directly BDP‑sized; based on acceptable worst‑case memory. Default is derived from bandwidth & RTT.
• 🧽 wash – Clear all Diffserv bits (but not ECN) after priority queuing. Useful when you cannot trust upstream markings (e.g., Comcast). Best combined with besteffort on inbound.
• ✂️ split‑gso (default) – Split GSO super‑packets into their on‑wire components for lower latency. At >10 Gbps, no‑split‑gso may raise throughput by keeping larger packets.

🎯 OVERRIDING CLASSIFICATION WITH TC FILTERS

CAKE allows overriding its internal classification via standard tc filters. Tin assignment uses skb→priority, and flow hashing uses classid.

🏷️ Tin override

The major number of the priority must match the qdisc handle; the minor is the tin index. Example: classify ICMP as ‘bulk’ (tin 1).

# tc qdisc replace dev eth0 handle 1: root cake diffserv3
# tc filter add dev eth0 parent 1: protocol ip prio 1 \
  u32 match icmp type 0 0 action skbedit priority 1:1

🔢 Flow hash override

Set the classid to override flow hashing. The major number is the host hash (used in host‑isolation modes), the minor is the flow hash (used in flows mode). Both can be set; missing ones are computed normally. Host isolation assigns a host ID to flow queues – the same flow cannot have multiple host IDs. Overriden host ID is used as both source and destination host.

# tc qdisc replace dev eth0 handle 1: root cake
# tc filter add dev eth0 parent 1: protocol ip prio 1 \
  u32 match icmp type 0 0 classid 0:1

💡 EXAMPLES

Setup and initial statistics:

# tc qdisc delete root dev eth0
# tc qdisc add root dev eth0 cake bandwidth 100Mbit ethernet
# tc -s qdisc show dev eth0
qdisc cake 1: root refcnt 2 bandwidth 100Mbit diffserv3 triple‑isolate rtt 100.0ms noatm overhead 38 mpu 84
 Sent 0 bytes 0 pkt (dropped 0, overlimits 0 requeues 0)
 backlog 0b 0p requeues 0
 memory used: 0b of 5000000b
 capacity estimate: 100Mbit
 min/max network layer size:        65535 /       0
 min/max overhead‑adjusted size:    65535 /       0
 average network hdr offset:            0

                  Bulk  Best Effort        Voice
 thresh       6250Kbit      100Mbit       25Mbit
 target          5.0ms        5.0ms        5.0ms
 interval      100.0ms      100.0ms      100.0ms
 pk_delay          0us          0us          0us
 av_delay          0us          0us          0us
 sp_delay          0us          0us          0us
 pkts                0            0            0
 bytes               0            0            0
 way_inds            0            0            0
 way_miss            0            0            0
 way_cols            0            0            0
 drops               0            0            0
 marks               0            0            0
 ack_drop            0            0            0
 sp_flows            0            0            0
 bk_flows            0            0            0
 un_flows            0            0            0
 max_len             0            0            0
 quantum           300         1514          762

After some traffic:

# tc -s qdisc show dev eth0
qdisc cake 1: root refcnt 2 bandwidth 100Mbit diffserv3 triple‑isolate rtt 100.0ms noatm overhead 38 mpu 84
 Sent 44709231 bytes 31931 pkt (dropped 45, overlimits 93782 requeues 0)
 backlog 33308b 22p requeues 0
 memory used: 292352b of 5000000b
 capacity estimate: 100Mbit
 min/max network layer size:           28 /    1500
 min/max overhead‑adjusted size:       84 /    1538
 average network hdr offset:           14

                  Bulk  Best Effort        Voice
 thresh       6250Kbit      100Mbit       25Mbit
 target          5.0ms        5.0ms        5.0ms
 interval      100.0ms      100.0ms      100.0ms
 pk_delay        8.7ms        6.9ms        5.0ms
 av_delay        4.9ms        5.3ms        3.8ms
 sp_delay        727us        1.4ms        511us
 pkts             2590        21271         8137
 bytes         3081804     30302659     11426206
 way_inds            0           46            0
 way_miss            3           17            4
 way_cols            0            0            0
 drops              20           15           10
 marks               0            0            0
 ack_drop            0            0            0
 sp_flows            2            4            1
 bk_flows            1            2            1
 un_flows            0            0            0
 max_len          1514         1514         1514
 quantum           300         1514          762

🔗 SEE ALSO

tc(8), tc‑codel(8), tc‑fq_codel(8), tc‑htb(8)

✍️ AUTHORS

CAKE principal author: Jonathan Morton. Contributors: Tony Ambardar, Kevin Darbyshire‑Bryant, Toke Høiland‑Jørgensen, Sebastian Moeller, Ryan Mounce, Dean Scarff, Nils Andreas Svee, Dave Taht.

Manual page: Loganaden Velvindron. Corrections to: <netdev AT vger.ker‑nel.org>.

📛 NAME 🚀 Quick Reference 📋 SYNOPSIS 📘 DESCRIPTION 🔧 SHAPER PARAMETERS 📦 OVERHEAD COMPENSATION PARAMETERS ⏱️ ROUND TRIP TIME PARAMETERS 🔀 FLOW ISOLATION PARAMETERS 🎚️ PRIORITY QUEUE PARAMETERS 🛠️ OTHER PARAMETERS 🎯 OVERRIDING CLASSIFICATION WITH TC FILTERS 💡 EXAMPLES 🔗 SEE ALSO ✍️ AUTHORS