1. IETF87 Berlin DCTCP implementation in FreeBSD
Midori Kato
Richard Scheffenegger
Lars Eggert
Alexander Zimmermann
Keio logcwndを下げないことで、delayの収束時間およびcwndの収束時間が遅くなる。

2. DCTCP (Data Center TCP) overview
Problem statement
Large FIFO buffers in the presence of bulk transfers impacts latency-sensitive flows
DCTCP
Maintain the queue size short using ECN (Explicit Congestion Notification)
Endpoint-only mechanism
Switches/routers’ mechanism is same as standard ECN
There are two kinds of flows..
Without DCTCP
With DCTCP
throughput-sensitive flows
High
throughput
High
throughput
Short delay
Long delay
latency-sensitive flows
ECN threshold

3. DCTCP algorithm Sender side Receiver side
Maintain the fraction of ECN marked seg. for each RTT
and update average fraction
of marked seg. ( )
Adopt alpha to cwnd decrease
Mark ECE only when CE packet is received
send immediate ACK when CE state is changed (regardless of delayed ACK)
F = \frac{\mathrm{\#\ of\ marked\ ACKs}}{\mathrm{\#\ of\ ACKs}}
\alpha \leftarrow (1 - g)\ \alpha + g F
\mathrm{cwnd} \leftarrow (1 - \frac{\alpha}{2} )\mathrm{\ cwnd}
wo Delayed ACK
w Delayed ACK S R S R
Immediate
ACK
** reaction to packet loss (e.g., dup ack and timeout) is same with NewReno
CE (Congestion Experience)
ECE (ECN Echo)

4. FreeBSD implementation status [1/2]
Total # of lines: 479 lines
tcp_input.c (24 lines), tcp_output.c (12 lines),
cc.h (4 lines), cc_dctcp.c (439 lines)
Add ECN handling functions to CC module
Add DCTCP as one of CC algorithms
References for FreeBSD DCTCP
Released Linux patch [1]
DCTCP paper “Data Center TCP (DCTCP) [SIGCOMM’10]”
Tcp_input: 24 lines
Tcp_output: 12 lines
cc.h: 4 lines
Cc_dctcp: 439 lines
[1]

5. FreeBSD implementation status [2/2]
Design issues
Cwnd behavior on first ECN
Halve or Ignore cwnd
Reconsideration of alpha update interval
every RTT or more frequent (e.g., every ack)
Alpha value handling after idle time
Discard or keep alpha
Alpha calculation affected by receive buffer control
Change nothing or handle alpha as special case
Topic in this talk
In progress

6. Open issue: cwnd behavior on first ECN
Option 1: DCTCP never reduces cwnd at first congestion recovery
Option 2: halves cwnd at first congestion recovery
0.5
[KBytes]
Linux DCTCP
Linux DCTCP
[ms]
150
100 50 40 20
Benefit:
Delay converges
quickly
Font size
Time = zero
Reaction to the first congestion event
----- 会議メモ (7/24/13 11:31) -----
srtt chart
should add queueing delay and rtt line
FreeBSD DCTCP
150
100 50
FreeBSD DCTCP 40 20
Time[s]
Time[s]

7. Performance with small queue size
Experiment
- Measure transfer time for X MB and SRTT (Smoothed RTT) using flowgrind
sender
receiver
dummynet
BW: 2Mbps, delay: 20ms
qlen: 7, ecn_thresh: 5
Transfer time
Xaxis -> logscale
----- 会議メモ (7/24/13 11:31) -----
dummynet tune bw and buffer size
newreno -> sack tcp
color
red - oragange
green -> dctcp

8. Performance with large queue size
Experiment
- Measure transfer time for X MB and SRTT (Smoothed RTT) using flowgrind
sender
receiver
dummynet
BW: 2Mbps, delay: 20ms
qlen: 30, ecn_thresh: 10
Why dctcp delay is bigger than sack tcp/our dctcp is because sigcomm dctcp never halves window size, delay is on the way of window shrinking

9. Summary Introduce DCTCP implementation in FreeBSD and design issues
Plan to submit our code to FreeBSD
Future work
Interaction with RFC3168
CUBIC and DCTCP
Change architecture for coexist of DCTCP and other CC algorithms
----- 会議メモ (7/24/13 11:31) -----
switch arch ->

10. Performance evaluation
Experiment
- Measure transfer time for X KB and SRTT (Smoothed RTT) using flowgrind
sender
receiver
dummynet
BW: 2Mbps, delay: 20ms
qlen: 14, ecn_thresh: 10
Transfer time
Xaxis -> logscale
----- 会議メモ (7/24/13 11:31) -----
dummynet tune bw and buffer size
newreno -> sack tcp
color
red - oragange
green -> dctcp

11. Microscope view of Sack TCP with ECN

12. Microscope view of Linux DCTCP

13. Microscope view of FreeBSD DCTCP

14. TCP(RFC3168) and DCTCP Question TCP (RFC 3168) DCTCP sender
When is cwnd reduced?
Reduce cwnd to half as soon as it receives ECE seg.
Reduce cwnd using the proportion of marked bytes
When CWR is marked?
Only when the host receives CWR
When the host reduces cwnd
Receiver (wo Delayed ACK)
When is ECE marked?
Keep sending segment with ECE until the host receives CWR
Send a segment with ECE only when the packet is marked CE (*)
Receiver (w Delayed ACK)
(*) +
Send immediate ack when CE state has changed