HSTCP-LP: A Protocol for Low-Priority Bulk Data Transfer in High-Speed High-RTT Networks
description
Transcript of HSTCP-LP: A Protocol for Low-Priority Bulk Data Transfer in High-Speed High-RTT Networks
Rice Networks Grouphttp://www.ece.rice.edu/networks
Aleksandar Kuzmanovic Edward W. Knightly
Rice UniversityR. Les Cottrell
SLAC/SCS-Network Monitoring
HSTCP-LP: A Protocol for Low-Priority Bulk Data Transfer in
High-Speed High-RTT Networks
Kuzmanovic, Knightly, and Cottrell
Motivation
Traditional view of service differentiation:– High priority: real-time service– Best-effort: everything else
What’s missing?– Low-priority (receiving only excess bandwidth)– Lower than best-effort!
Non-interactive applications, bulk download»It will last long anyway...
Speeds up best-effort service (e.g., web)
Kuzmanovic, Knightly, and Cottrell
Applications for Low Priority Service
LP vs. fair-share:– Bulk downloads
SLAC->CERN Improve my other
applications
LP vs. rate-limiting:– P2P file sharing
Often rate limited Isolation vs. sharing
T im e
R a te lim it
C a pa c ityA va ila bleba ndw idth
G a in
AB
Kuzmanovic, Knightly, and Cottrell
Problem Formulation & Design Objectives
Low-priority service objectives– Utilize the “excess/available” capacity
What no other flows are using
– TCP-transparency (non-intrusiveness)– Inter-LP flow fairness (fair-share of the available
bandwidth)
Design end- host- based transm ission protocol that em ulates the low- priority service
Kuzmanovic, Knightly, and Cottrell
Origins of the Available Bandwidth
Why is excess bandwidth available when TCP is greedy?
– Time-of-day effects Night- vs. day-time traffic
– Short-lived flows Majority traffic is web browsing
– TCP is imperfect Low-aggregation regimes
Kuzmanovic, Knightly, and Cottrell
TCP fairness
In presence of TCP
cross-traffic:– TCP achieves fairness
A B
T C P 1
T C P 2 T im e
F ull C ap ac ityT C P 1 d e m a n d
T im e
F ull C ap ac ityT C P 2 d e m a n d
T im e
B a n d w id th s h a re
In ter-T C P -fairn es s
Kuzmanovic, Knightly, and Cottrell
Illustration of TCP Transparency
LP flow utilizes only excess bandwidth
– Does not reduce the throughput of TCP flows
A B
T C P
LP T im e
F ull C ap ac ity
T C P d e m a n d
T im e
F ull C ap ac ity
L P d e m a n d
T im e
F ull C ap ac ity
B a n d w id th s h a re
L P flo w u tilize sa v a ila b le b a n d w id th
T C P -tra n s p a re n c y
Kuzmanovic, Knightly, and Cottrell
Fairness Among LP Flows
Inter-LP-fairness is
essential for
simultaneous
file transfers
A B
T C P
LP 1LP 2 T im e
F ull C ap ac ityT C P d e m a n d
T im e
F ull C ap ac ityL P 1 a n d L P 2 d e m a n d s
T im e
B a n d w id th s h a re
In ter-L P -fairn es s
Kuzmanovic, Knightly, and Cottrell
HSTCP-LPA Congestion Control Protocol
Problem– TCP-LP [Infocom03]
AIMD control not
suitable for high speeds
Approach– Add HSTCP [Floyd03] high-speed mechanisms
Challenge– Develop large windows and achieve low-
prioritization at the same time
Lowspeed
Highspeed
Excess bandw idth utilization
Transparency
Kuzmanovic, Knightly, and Cottrell
Early Congestion Indication
For transparency, HSTCP-LP must know of congestion before non-LP flows
Idealized objective: buffer threshold indication
Endpoint inference: one-way delay threshold
)( minmaxmin dddsdi
b uffer thres ho ld
Kuzmanovic, Knightly, and Cottrell
Reverse Cross-Traffic
HSTCP-LP uses one-way packet delays (RFC1323) for congestion indication – Source-destination time stamping– Synchronized clocks not needed
One-way delay: – Eliminates the impact of the reverse cross-traffic
HSTCP-LP can loose Gb/s of available bandwidth if RTT is used
A
B
D a ta pa thA ck pa th
Kuzmanovic, Knightly, and Cottrell
HSTCP-LP Timeline IllustrationC
on
ge
stio
n W
ind
ow
T im e
packet loss • Reach high throughput ...quickly in slow-start• Estimate one-way ...delay and compute ...delay threshold
Kuzmanovic, Knightly, and Cottrell
HSTCP-LP Timeline IllustrationC
on
ge
stio
n W
ind
ow
T im e
packet loss
sm oothedone-w ay delayindications
Apply HSTCP response function with early congestion indications
Kuzmanovic, Knightly, and Cottrell
HSTCP-LP Timeline IllustrationC
on
ge
stio
n W
ind
ow
T im e
packet loss
sm oothedone-w ay delayindications
• HSTCP response ...function above ...Low_Window• TCP-LP response ...function below ...Low_Window• Send 1 packet/RTT
Kuzmanovic, Knightly, and Cottrell
HSTCP-LP Timeline IllustrationC
on
ge
stio
n W
ind
ow
T im e
packet loss
sm oothedone-w ay delayindications
HSTCP increase function with early congestion indications
Kuzmanovic, Knightly, and Cottrell
Implementation and Experiments
Linux-2.4.22-web100 kernel: – HSTCP HSTCP-LP
Time stamping, modified congestion control...
Paths:– Stanford, CA Ann Arbor, MI– Stanford, CA Gainesville, FL– Available bitrate ~ 450 Mb/s
Kuzmanovic, Knightly, and Cottrell
Low-Priority Background Data Transfer
HSTCP-LP is highly non-intrusive to other advanced TCP stacks
– Strict prioritization for larger bottleneck queue lengths – Lighter prioritization for (max_delay < 50 ms)
Kuzmanovic, Knightly, and Cottrell
Utilizing Excess Bandwidth
Extensive experiments on a number of high bandwidth-delay production networks– no cross-traffic– periodic UDP cross-traffic– reverse cross-traffic– http://www.slac.stanford.edu/~hadrien
HSTCP-LP utilizes 80-127% of the available bandwidth when compared to other high-speed stacks– 127% when transmission buffer (tqueuelen) is
large
Kuzmanovic, Knightly, and Cottrell
Conclusions
HSTCP-LP: a new protocol for high bandwidth-delay production networks– General low priority service (compared to “best-effort”)
Attractive for low-priority bulk downloads: ftp, web updates, P2P
HSTCP-LP is incrementally deployable– Sender side modification of HSTCP without changes to
routers
Implementation and evaluation in the Internet
http://www.ece.rice.edu/networks/TCP-LP
Kuzmanovic, Knightly, and Cottrell
Questions
http://www.ece.rice.edu/networks/TCP-LP