Transcript Defense - Northwestern Networks Group

L Subramanian*, I Stoica*, H Balakrishnan+, R Katz*
*UC Berkeley, MIT+
USENIX NSDI’04, 2004
Presented by Alok Rakkhit, Ionut Trestian







Introduction
OverQos Architecture
Controlled-Loss Virtual Link (CLVL)
OverQoS Implementation
Two Sample Application
Evaluation
Conclusions
2


Today’s Internet still continues to provide
only a best-effort service. The main reason is
the requirement of these proposals that all
network elements implement QoS
mechanisms.
The authors propose OverQoS, an overlay
based QoS architecture for enhancing
Internet QoS.
3

Enhancements:
 Smoothing losses
▪ Reduce or even eliminate the loss bursts by
smoothing packet losses across time
 Packet prioritization
▪ Protect important packets
 Statistical Bandwidth and Loss Guarantees
4

Assumptions
 The placement of overlay nodes is pre-specified
 The end-to-end path on top of an overlay network is fixed
Using existing approaches like RON to determine
the overlay path.
 Terms

 Virtual link – The IP path between two overlay nodes
 Bundle – A stream of application data packets carried
across the virtual link
5

Overlay-based QoS challenges
 Node Placement and Cross Traffic
 Fairness
▪ Should not hurt the cross traffic
 Stability
▪ Many virtual links overlapping on congested physical
links should be able to co-exist
6

A Solution builds on two principles
 Bundle loss control
▪ Using controlled-loss virtual link (CLVL) to bound the
loss rate
 Resource management within a bundle
▪ Control the loss and bandwidth allocations
7

The CLVL provides a loss rate bound, q.
 Using a combination of FEC and ARQ
 The bandwidth overhead should be minimized

The total traffic consists of:
 The traffic of the bundle
 The redundancy traffic

The available bandwidth for the flows in the bundle
b(t): Traffic bound at time t
r(t): Fraction of redundancy
traffic
8

If the traffic arrival rate is larger than available
bandwidth c, the extra traffic is dropped at the
entry overlay node
 With priority

Statistical bandwidth guarantees

, where u represents the probability of not
meeting the bandwidth guarantee
 As long as the total allocated bandwidth is less than cmin
9

Application-OverQoS Interface



It needs to tunnel its
packets through the
overlay network using
an OverQoS proxy
The proxy is
responsible for
signaling the
application specific
requirements to
OverQoS
OverQoS proxy is
application specific
10

End-to-end Recovery vs. Overlay CLVL
 Using FEC to apply end-to-end loss control is far more
expensive than on an aggregate level
 With a better distribution of overlay nodes, they expect
the overlay links to have much smaller RTTs than endto-end RTTs
▪ ARQ recovery is better in overlay-level

Delay guarantees
 Overlay has no control in queuing delays

Over-provisioning
 Overlay are the right platform for translating intra
domain QoS to end-to-end QoS guarantees
11

Estimating b
 Based on an N-TCP pipe abstraction which provides a
bandwidth which is N times the throughput of a single
TCP connection.
▪ Use MulTCP to emulate the behavior
▪ N is equal to the number of flows in the bundle

Node Architecture
q: target loss-rate
c: available bandwidth
p: loss rate
b: maximum sending rate
12

Achieving target loss rate q
 FEC vs. ARQ trade-off
▪ Bandwidth overhead and packet recovery time
 FEC+ARQ based CLVL
▪ Restrict # of retransmissions to at most one
▪ The expected packet loss rate
 After two rounds
 Goal
r is the redundancy factor
▪ The expected bandwidth overhead
 Minimizes
▪ The optimal solution is when r1 = 0
13


Application-dependent proxy
Choosing parameters
 N as the average number of flows observed over a larger period of time
 q = 0.1%

Startup phase
 Using a slow-start phase to estimate the initial value of b

FEC implementation
 Operating on small window sizes (n < 1000)  coding is not a bottleneck

Two enhancements
 The quality can be enhanced by converting bursty losses
into smooth losses  for streaming audio
 Recovering packets preferentially can improve the quality
 for MPEG streaming

Not consume any additional bandwidth
 Retransmits an important lost packet and drops a later
lesser important packet
15

Streaming Audio

Perceptual Evaluation of Speech Quality (PESQ)
MPEG streaming (5 is ideal)
Average loss rate
Mazu-Korea – 2%
Intel-Lulea – 3%
Increase
0.15 – 0.2
Not only improves the quality in the average case
but also the minimum quality of a stream
16

Problem
 Client unable to connect to the server
 Cause skips or get disconnected

Alleviate the problem of bursty losses by
performing:
 Recover from bursty network losses by using an FEC+ARQ
based CLVL
 Smoothly drop data packets equivalent to the size of the
burst at the overlay node
 Identify control packets based on packet size and not drop
these packets
17

Sequence number plot illustrating smoothing of packet
losses using OverQoS


Smoothing losses works
well only when the
bursty loss-periods are
relatively short by
compensating
Unable to achieve the
target loss-rate due to
congestion periods with
very high loss-rates
10% loss-rate
18

Methodology
 Wide-Area Evaluation Testbed
▪ RON and PlanetLab – use 19 diverse nodes
 Simulation Environment
▪ Ns-2 – a single congested link of 10 Mbps where they
vary the background traffic
▪ Long lived TCP connections
▪ Self similar traffic
▪ Web traffic
19

Simulations

Wide Area Evaluation
q = 0.1%
 Achieve target over 80 of the 83 virtual links
 The causes of the other 3 virtual links
▪ Short outages – a period of time all packets are lost (< 5s)
▪ Bi-modal loss distributions – bursty losses
20
Monitor 83 unique virtual links
u = 0.01 and u = 0.005
 Stability of cmin
Calculate cmin based on a
history of 200 seconds
The average sending rate of N-TCP is
between 120Kbps to 2Mbps
N-TCP, N = 10
The value of cmin is greater than 100Kbps
for more than 80% of the links
1) The value of cmin is very stable,
which does not deviate more
than 10% around its mean
2) Set P = 1%, the actual value is
no more than 1.3%
21

Overhead Characteristics
The burstier the background
traffic, the higher the amount
of FEC required to recover
from these losses
The difference between avg. loss & FEC+ARQ is
the amount of FEC used in the second round
22

Delay Characteristics
 Two reasons for increasing delay
▪ The recovery process
▪ Support in-sequence delivery of packets
Three different models
(a) No packet ordering
(b) End-to-end ordering
(c) Hop-by-hop ordering
1) E2E is better than Hop-by-hop
2) Adding new OverQoS nodes
increasing limited delay
23

Three OverQoS bundles (with N=2, N=4, N=8) compete
on a shared bottleneck under two different scenarios


No cross-traffic
Cross-traffic
consisting of five
long lived TCPs
1) Three OverQoS bundles
co-exist with each other
and with the background
traffic
2) The ratio of throughputs of
the three bundles is
preserved
24
OverQoS can enhance Internet QoS without any
support from the underlying IP network
 OverQoS is able to achieve the three enhancements
with little (i.e., 5%) or no extra bandwidth.
 Future work

 Combine admission control and path selection
 Determine the “optimal” placement of the OverQoS
nodes in the network
25