one more bit is enough yong xia, rpi lakshmi subramanian, ucb ion stoica, ucb shiv kalyanaraman, rpi...
TRANSCRIPT
![Page 1: One More Bit Is Enough Yong Xia, RPI Lakshmi Subramanian, UCB Ion Stoica, UCB Shiv Kalyanaraman, RPI SIGCOMM’ 05, Philadelphia, PA 08 / 23 / 2005](https://reader036.vdocument.in/reader036/viewer/2022062511/551beea4550346af588b6577/html5/thumbnails/1.jpg)
One More Bit Is EnoughOne More Bit Is Enough
Yong Xia, RPILakshmi Subramanian, UCB
Ion Stoica, UCBShiv Kalyanaraman, RPI
SIGCOMM’05, Philadelphia, PA
08 / 23 / 2005
![Page 2: One More Bit Is Enough Yong Xia, RPI Lakshmi Subramanian, UCB Ion Stoica, UCB Shiv Kalyanaraman, RPI SIGCOMM’ 05, Philadelphia, PA 08 / 23 / 2005](https://reader036.vdocument.in/reader036/viewer/2022062511/551beea4550346af588b6577/html5/thumbnails/2.jpg)
VCP 2
Motivation #1: TCP does not perform well in high b/w
bottleneck utilization
round-trip delay (ms)
TCP
50%
100%
bandwidth (Mbps)
bottleneck utilization
TCP
70%
100%
![Page 3: One More Bit Is Enough Yong Xia, RPI Lakshmi Subramanian, UCB Ion Stoica, UCB Shiv Kalyanaraman, RPI SIGCOMM’ 05, Philadelphia, PA 08 / 23 / 2005](https://reader036.vdocument.in/reader036/viewer/2022062511/551beea4550346af588b6577/html5/thumbnails/3.jpg)
VCP 3
Motivation #1: Why TCP does not scale?
TCP uses binary congestion signals, such as loss or one-bit Explicit Congestion Notification (ECN)
time
congestionwindow Multiplicative Decrease (MD)
Additive Increase (AI)
congestion
slow start
congestion avoidance
AI with a fixed step-size can be very slow for large bandwidth
slow!
![Page 4: One More Bit Is Enough Yong Xia, RPI Lakshmi Subramanian, UCB Ion Stoica, UCB Shiv Kalyanaraman, RPI SIGCOMM’ 05, Philadelphia, PA 08 / 23 / 2005](https://reader036.vdocument.in/reader036/viewer/2022062511/551beea4550346af588b6577/html5/thumbnails/4.jpg)
VCP 4
Motivation #2: XCP scales
XCP decouples efficiency control and fairness control
x
sender receiverrouter
C
spare bandwidth -- MIMD
DATA
-- AIMD rate & rtt
ACKrate
But, XCP needs multiple bits (128 bits in its current IETF draft) to carry the congestion-related information from/to network
![Page 5: One More Bit Is Enough Yong Xia, RPI Lakshmi Subramanian, UCB Ion Stoica, UCB Shiv Kalyanaraman, RPI SIGCOMM’ 05, Philadelphia, PA 08 / 23 / 2005](https://reader036.vdocument.in/reader036/viewer/2022062511/551beea4550346af588b6577/html5/thumbnails/5.jpg)
VCP
Goal
5
Design a TCP-like scheme that:
requires a small amount of congestion information (e.g., 2 bits)
scales across a wide range of network scenarios
![Page 6: One More Bit Is Enough Yong Xia, RPI Lakshmi Subramanian, UCB Ion Stoica, UCB Shiv Kalyanaraman, RPI SIGCOMM’ 05, Philadelphia, PA 08 / 23 / 2005](https://reader036.vdocument.in/reader036/viewer/2022062511/551beea4550346af588b6577/html5/thumbnails/6.jpg)
VCP
Key Observation
6
Fairness is not critical in low-utilization region
Use Multiplicative Increase (MI) for fast convergence onto efficiency in this region
Handle fairness in high-utilization region
![Page 7: One More Bit Is Enough Yong Xia, RPI Lakshmi Subramanian, UCB Ion Stoica, UCB Shiv Kalyanaraman, RPI SIGCOMM’ 05, Philadelphia, PA 08 / 23 / 2005](https://reader036.vdocument.in/reader036/viewer/2022062511/551beea4550346af588b6577/html5/thumbnails/7.jpg)
VCP 7
Variable-structure congestion Control Protocol (VCP)
sender receiver
x
router
Routers signal the level of congestion End-hosts adapt the control algorithm accordingly
traffic rate link capacity
(11)
(10)
(01)
code loadfactor
region
low-load
high-load
overload
control
Multiplicative Decrease (MD)
Additive Increase (AI)
Multiplicative Increase (MI)
range of interest
ACK
2-bit ECN
0
1
scale-free
2-bit ECN
![Page 8: One More Bit Is Enough Yong Xia, RPI Lakshmi Subramanian, UCB Ion Stoica, UCB Shiv Kalyanaraman, RPI SIGCOMM’ 05, Philadelphia, PA 08 / 23 / 2005](https://reader036.vdocument.in/reader036/viewer/2022062511/551beea4550346af588b6577/html5/thumbnails/8.jpg)
VCP 8
VCP vs. ECN
sender receiver
x
router
code loadfactor
region
overload
control
Multiplicative Decrease (MD)
0
1
ECN doesn’t differentiate between low-load and high-load regions
TCP AQM / ECN
(11)
(10)
(01)low-load
high-loadAdditive Increase (AI)
Multiplicative Increase (MI)
ACK
2-bit ECN
VCP
(1)
(0)underloadAdditive Increase (AI)
ACK
1-bit ECN
![Page 9: One More Bit Is Enough Yong Xia, RPI Lakshmi Subramanian, UCB Ion Stoica, UCB Shiv Kalyanaraman, RPI SIGCOMM’ 05, Philadelphia, PA 08 / 23 / 2005](https://reader036.vdocument.in/reader036/viewer/2022062511/551beea4550346af588b6577/html5/thumbnails/9.jpg)
VCP 9
An illustration example
MI tracks available bandwidth exponentially fast
time (sec)
After high utilization is attained, AIMD provides fairness
link utilization
flow cwnd (pkt)
9 flows join and then leave
MI AIMD
![Page 10: One More Bit Is Enough Yong Xia, RPI Lakshmi Subramanian, UCB Ion Stoica, UCB Shiv Kalyanaraman, RPI SIGCOMM’ 05, Philadelphia, PA 08 / 23 / 2005](https://reader036.vdocument.in/reader036/viewer/2022062511/551beea4550346af588b6577/html5/thumbnails/10.jpg)
VCP 10
VCP vs. ECN
TCP+RED/ECN
cwnd
utilization
time (sec)
VCP
utilization
cwnd
time (sec)
![Page 11: One More Bit Is Enough Yong Xia, RPI Lakshmi Subramanian, UCB Ion Stoica, UCB Shiv Kalyanaraman, RPI SIGCOMM’ 05, Philadelphia, PA 08 / 23 / 2005](https://reader036.vdocument.in/reader036/viewer/2022062511/551beea4550346af588b6577/html5/thumbnails/11.jpg)
VCP 11
VCP key ideas and properties
Achieve high efficiency, low loss, and small queue Fairness model is similar to TCP:
Long flows get lower bandwidth than in XCP (proportional vs. max-min fairness)
Fairness convergence much slower than XCP
Use network link load factor as the congestion signal Decouple efficiency and fairness controls in different load regions
overload
high-load
low-load
router
fairness control
efficiency control MI
AIMD
end-host
![Page 12: One More Bit Is Enough Yong Xia, RPI Lakshmi Subramanian, UCB Ion Stoica, UCB Shiv Kalyanaraman, RPI SIGCOMM’ 05, Philadelphia, PA 08 / 23 / 2005](https://reader036.vdocument.in/reader036/viewer/2022062511/551beea4550346af588b6577/html5/thumbnails/12.jpg)
VCP 12
Major design issues
At the router How to measure and encode the load factor?
At the end-host When to switch from MI to AI? What MI / AI / MD parameters to use? How to handle heterogeneous RTTs?
control
Multiplicative Decrease (MD)Additive Increase (AI)
Multiplicative Increase (MI)
(11)(10)
(01)
code load region
low-load
high-loadoverload
![Page 13: One More Bit Is Enough Yong Xia, RPI Lakshmi Subramanian, UCB Ion Stoica, UCB Shiv Kalyanaraman, RPI SIGCOMM’ 05, Philadelphia, PA 08 / 23 / 2005](https://reader036.vdocument.in/reader036/viewer/2022062511/551beea4550346af588b6577/html5/thumbnails/13.jpg)
VCP 13
Design issue #1: measuring and encoding load factor
Calculate the link load factor
demandload_
factor capacity =
link_bandwidth * t
arrival_traffic + queue_size =
t = 200ms most rtt
The load factor is quantized and encoded into the two ECN bits
information loss affects fairness
capacity
demand
![Page 14: One More Bit Is Enough Yong Xia, RPI Lakshmi Subramanian, UCB Ion Stoica, UCB Shiv Kalyanaraman, RPI SIGCOMM’ 05, Philadelphia, PA 08 / 23 / 2005](https://reader036.vdocument.in/reader036/viewer/2022062511/551beea4550346af588b6577/html5/thumbnails/14.jpg)
VCP 14
Design issue #2: setting MI / AI / MD parameters (, , )
overload
high-load
low-load MI
AIMD
![Page 15: One More Bit Is Enough Yong Xia, RPI Lakshmi Subramanian, UCB Ion Stoica, UCB Shiv Kalyanaraman, RPI SIGCOMM’ 05, Philadelphia, PA 08 / 23 / 2005](https://reader036.vdocument.in/reader036/viewer/2022062511/551beea4550346af588b6577/html5/thumbnails/15.jpg)
VCP
Design issue #2: setting MI / AI / MD parameters (, , )
Q: load factor transition point * for MI AI?
load factor
100%
80%
safety margin
0%
0.875
15
TCP: = 1.0 VCP: =
0.06STCP: =
0.01
MI
AI
MD
k (1 *) / *
where k = 0.25 (for stability)
80%
20%
1.0 = =
=
* =
![Page 16: One More Bit Is Enough Yong Xia, RPI Lakshmi Subramanian, UCB Ion Stoica, UCB Shiv Kalyanaraman, RPI SIGCOMM’ 05, Philadelphia, PA 08 / 23 / 2005](https://reader036.vdocument.in/reader036/viewer/2022062511/551beea4550346af588b6577/html5/thumbnails/16.jpg)
VCP
Design issue #3: Handling RTT heterogeneity for MI/AI
Scale to prevent MI from overshooting capacity when RTT is small
16
rtt < t
overshoot s
if rtt < t
rtt = t
t t
time
spare capacity
1
![Page 17: One More Bit Is Enough Yong Xia, RPI Lakshmi Subramanian, UCB Ion Stoica, UCB Shiv Kalyanaraman, RPI SIGCOMM’ 05, Philadelphia, PA 08 / 23 / 2005](https://reader036.vdocument.in/reader036/viewer/2022062511/551beea4550346af588b6577/html5/thumbnails/17.jpg)
VCP
VCP scales across b/w, rtt, num flows
Evaluation using extensive ns2 simulations
17
150Mbps, 80ms, 50 forward flows and 50 reverse flows
![Page 18: One More Bit Is Enough Yong Xia, RPI Lakshmi Subramanian, UCB Ion Stoica, UCB Shiv Kalyanaraman, RPI SIGCOMM’ 05, Philadelphia, PA 08 / 23 / 2005](https://reader036.vdocument.in/reader036/viewer/2022062511/551beea4550346af588b6577/html5/thumbnails/18.jpg)
VCP
VCP achieves high efficiency
18
bottleneck utilization
TCP
VCP
XCP
bandwidth (Mbps)
bottleneck utilization
TCP
VCP
XCP
round-trip delay (ms)
![Page 19: One More Bit Is Enough Yong Xia, RPI Lakshmi Subramanian, UCB Ion Stoica, UCB Shiv Kalyanaraman, RPI SIGCOMM’ 05, Philadelphia, PA 08 / 23 / 2005](https://reader036.vdocument.in/reader036/viewer/2022062511/551beea4550346af588b6577/html5/thumbnails/19.jpg)
VCP
VCP minimizes packet loss rate
19
TCPVCPXCP
round-trip delay (ms)
packet loss rate
TCP
VCPXCP
bandwidth (Mbps)
packet loss rate
TCP
VCP
XCP
TCP
VCP
XCP
bottleneck utilization bottleneck utilization
![Page 20: One More Bit Is Enough Yong Xia, RPI Lakshmi Subramanian, UCB Ion Stoica, UCB Shiv Kalyanaraman, RPI SIGCOMM’ 05, Philadelphia, PA 08 / 23 / 2005](https://reader036.vdocument.in/reader036/viewer/2022062511/551beea4550346af588b6577/html5/thumbnails/20.jpg)
VCP
VCP comparisons
20
Compared to TCP+AQM/ECN Same architecture (end-hosts control, routers signal) Router congestion detection: queue-based load-based Router congestion signaling: 1-bit 2-bit ECN End-host adapts (MI/AI/MD) according to the ECN feedback End-host scales its MI/AI parameters with its RTT
Compared to XCP Decouple efficiency/fairness control across load regions Functionality primarily placed at end-hosts, not in routers
![Page 21: One More Bit Is Enough Yong Xia, RPI Lakshmi Subramanian, UCB Ion Stoica, UCB Shiv Kalyanaraman, RPI SIGCOMM’ 05, Philadelphia, PA 08 / 23 / 2005](https://reader036.vdocument.in/reader036/viewer/2022062511/551beea4550346af588b6577/html5/thumbnails/21.jpg)
VCP
Theoretical results
21
Assumptions: One bottleneck of infinite buffer space is shared by synchronous
flows that have identical RTTs; The exact value of load factor is echoed back.
Theorem for the VCP fluid model: It is globally stable with a unique and fair equilibrium, if k 0.5; The equilibrium is max-min fair for general topologies; The equilibrium is optimal by achieving all the design goals.
VCP protocol differs from the model in fairness.
![Page 22: One More Bit Is Enough Yong Xia, RPI Lakshmi Subramanian, UCB Ion Stoica, UCB Shiv Kalyanaraman, RPI SIGCOMM’ 05, Philadelphia, PA 08 / 23 / 2005](https://reader036.vdocument.in/reader036/viewer/2022062511/551beea4550346af588b6577/html5/thumbnails/22.jpg)
VCP
Conclusions
22
With a few minor changes over TCP + AQM / ECN, VCP is able to approximate the performance of XCP
High efficiency Low persistent bottleneck queue Negligible congestion-caused packet loss Reasonable (i.e., TCP-like) fairness
![Page 23: One More Bit Is Enough Yong Xia, RPI Lakshmi Subramanian, UCB Ion Stoica, UCB Shiv Kalyanaraman, RPI SIGCOMM’ 05, Philadelphia, PA 08 / 23 / 2005](https://reader036.vdocument.in/reader036/viewer/2022062511/551beea4550346af588b6577/html5/thumbnails/23.jpg)
VCP
Future work
23
How do we get there, incrementally? End-to-end VCP TCP-friendliness Incentive
Extensions Applications: short-lived data traffic, real-time traffic Environment: wireless channel
Security Robust signaling, e.g., ECN nonce
![Page 24: One More Bit Is Enough Yong Xia, RPI Lakshmi Subramanian, UCB Ion Stoica, UCB Shiv Kalyanaraman, RPI SIGCOMM’ 05, Philadelphia, PA 08 / 23 / 2005](https://reader036.vdocument.in/reader036/viewer/2022062511/551beea4550346af588b6577/html5/thumbnails/24.jpg)
VCP
The end
24
Thanks!
![Page 25: One More Bit Is Enough Yong Xia, RPI Lakshmi Subramanian, UCB Ion Stoica, UCB Shiv Kalyanaraman, RPI SIGCOMM’ 05, Philadelphia, PA 08 / 23 / 2005](https://reader036.vdocument.in/reader036/viewer/2022062511/551beea4550346af588b6577/html5/thumbnails/25.jpg)
VCP
Design issue #3: Handling RTT heterogeneity for MI/AI
25
TCP throughput is biased against flows with large RTT
packet_sizerate
rtt sqrt( loss_rate )
VCP scales for fair rate sharing (regardless of RTT)
rate = cwnd / rtt
![Page 26: One More Bit Is Enough Yong Xia, RPI Lakshmi Subramanian, UCB Ion Stoica, UCB Shiv Kalyanaraman, RPI SIGCOMM’ 05, Philadelphia, PA 08 / 23 / 2005](https://reader036.vdocument.in/reader036/viewer/2022062511/551beea4550346af588b6577/html5/thumbnails/26.jpg)
VCP
VCP keeps small bottleneck queue
26
queue length in % buffer size
TCP VCPXCP
bandwidth (Mbps)
queue length in % buffer size
TCPVCP XCP
round-trip delay (ms)
per-flow b/w-delay-product < 1 packet
![Page 27: One More Bit Is Enough Yong Xia, RPI Lakshmi Subramanian, UCB Ion Stoica, UCB Shiv Kalyanaraman, RPI SIGCOMM’ 05, Philadelphia, PA 08 / 23 / 2005](https://reader036.vdocument.in/reader036/viewer/2022062511/551beea4550346af588b6577/html5/thumbnails/27.jpg)
VCP
Vary the number of flows
27
number of flows
VCP
TCP
bottleneck utilization
queue length in % buffer size
XCP
VCP
TCP
XCP
![Page 28: One More Bit Is Enough Yong Xia, RPI Lakshmi Subramanian, UCB Ion Stoica, UCB Shiv Kalyanaraman, RPI SIGCOMM’ 05, Philadelphia, PA 08 / 23 / 2005](https://reader036.vdocument.in/reader036/viewer/2022062511/551beea4550346af588b6577/html5/thumbnails/28.jpg)
VCP
Influence of RTT on fairness
To some extent, VCP distributes reasonably fairly
28
![Page 29: One More Bit Is Enough Yong Xia, RPI Lakshmi Subramanian, UCB Ion Stoica, UCB Shiv Kalyanaraman, RPI SIGCOMM’ 05, Philadelphia, PA 08 / 23 / 2005](https://reader036.vdocument.in/reader036/viewer/2022062511/551beea4550346af588b6577/html5/thumbnails/29.jpg)
VCP
Influence of RTT on fairness (cont’d)
29
![Page 30: One More Bit Is Enough Yong Xia, RPI Lakshmi Subramanian, UCB Ion Stoica, UCB Shiv Kalyanaraman, RPI SIGCOMM’ 05, Philadelphia, PA 08 / 23 / 2005](https://reader036.vdocument.in/reader036/viewer/2022062511/551beea4550346af588b6577/html5/thumbnails/30.jpg)
VCP
VCP converges onto fairness
30
![Page 31: One More Bit Is Enough Yong Xia, RPI Lakshmi Subramanian, UCB Ion Stoica, UCB Shiv Kalyanaraman, RPI SIGCOMM’ 05, Philadelphia, PA 08 / 23 / 2005](https://reader036.vdocument.in/reader036/viewer/2022062511/551beea4550346af588b6577/html5/thumbnails/31.jpg)
VCP
VCP converges onto fairness faster with more bits
31
![Page 32: One More Bit Is Enough Yong Xia, RPI Lakshmi Subramanian, UCB Ion Stoica, UCB Shiv Kalyanaraman, RPI SIGCOMM’ 05, Philadelphia, PA 08 / 23 / 2005](https://reader036.vdocument.in/reader036/viewer/2022062511/551beea4550346af588b6577/html5/thumbnails/32.jpg)
VCP
Responsiveness
50 flows + 150 flows – 150 flows 32