REALISTIC PACKET REORDERING - PowerPoint Presentation

Slide1

REALISTIC PACKET REORDERINGFOR NETWORK EMULATION AND SIMULATION

Aisha Syed, Robert RicciUniversity of Utah

1Slide2

IntroductionPacket reordering common in real networks

Retransmissions due to loss, multipath forwarding, load balancing within routers, etc.Performance affected by reordering*

Streaming media, VoIP, IPTV,

etc

2

*

Mashtizadeh

‘14

,

Narasiodeyar

‘13,

Lelarge

’

08,

Piratla

’

08,

Jaiswal

‘07,

Laor

‘02, Bennett ‘99 Slide3

Need reordering for realistic simulation/emulationEmulating cause

won’t result in precise or repeatable resultsGoal is precise, repeatable, controlled reorderingUsers may want to test their apps/protocols in high reordering networks

3

IntroductionSlide4

Reorder Density (RD) Metric

Reorder Density (RD) [Piratla ’05]C

aptures reordering

by measuring displacements

of packets

from

original positions

RD calculation

algorithm

Packet trace

 RDRD sequence regeneration algorithmRD  Packet trace with reordering appliedWill be used while simulating/emulating reorderingRD  Emulator  Reordered packet trace

4Slide5

ContributionsAlgorithm for sequence regeneration from the RD reordering metricDummynet emulator extension to support reordering

5Slide6

RD calculation example6

Send:

RD Histogram

Receive:

2

1

-1

2Slide7

Sequence Regeneration Algorithm

7

Input

RD HistogramSlide8

Use max flow –like approach w

ith additional components for constraintsCreate graph to represent permutations of displacements in input histogram

Use greedy search with backtracking

Find paths that represents output permutation

8

Sequence Regeneration AlgorithmSlide9

s

uper-sink

s

uper-source

s

ub-sinks (N)

sub-sources

(#

of displacements

)

bipartite graphs

N = 4 packets

4

3

2

1

Solution:

9

1

2

1

2

-1

-1

-2

-2

Space complexity

O(

numUniqueDisplacements

*

numPackets

)

numUniqueDisplacements

usually small

Time complexity

O

(numUniqueDisplacements

2

*

numPackets

)

Worst

-

case rare

in practiceSlide10

10

Evaluation

Real Internet traces from the literature

Algorithm correctness, and performance on real traces

Synthetic traces

Algorithm scalability with

respect to amount of

reordering

Algorithm scalability

with

respect

to

number of

packets

Datapath evaluation

Evaluated our Dummynet extension to see if it was causing any unnecessary overheadSlide11

11

145 hours of TCP traffic consisting of long-lived connections from Colorado to 6 destinations around

the world

1. Real traces

Algorithm

worked correctly

Got EXACTLY the same RDSlide12

Effect of amount of reordering on algorithm runtime

Number of packets kept

constant (1K)

12

Real

t

races

2. Synthetic tracesSlide13

13

C

ontributions

RD sequence regeneration algorithm

Reordering support added in Dummynet

Evaluated algorithm correctness and scalability, and Dummynet extension for any unnecessary overhead

Works correctly and fast enough for realistic traces

Conclusion

Thank YouSlide14

Thank YouSlide15

15

Effect

of number of packets on

algorithm runtime

Amount of reordering kept

constant to RD seen on real traceSlide16

16

3 2 1 6 5 4 …

Reordering

extension

Delay/bandwidth/loss

emulation

Dummynet

D

Other

optional

c

onfig.

S

1 2 3 4 5 6 …

Sequence Regen.

Algorithm

Reorder

config

.

Input

RD(s)

Experimenter WorkflowSlide17

[1]  Dummynet references from Citeseer. http://citeseerx.ist.psu.edu/viewdoc/

summary?doi=10.1.1.57.2969, 1:401–414, 2013. [2]  Packet reordering trace. http://www.cnrl.colostate.edu/Projects/PacketReordering/ Trace/packet reordering trace.htm, pages 401–414, 2013.

[3]  T. Banka. Metrics for degree of reordering in packet sequences. Proc. 27th IEEE Conference on Local Computer Networks, 1:333–342, November 2002.

[4]  J. C. R. Bennett. Packet reordering in not pathological network behavior. IEEE/ACM Trans.

Netw

., 7:789–798, 1999. [5]  P. E. Black. Fisher-Yates shuffle. Dictionary of Algorithms and Data Structures [online], US National Institute of Standards and Technology, 2005.

[6]  M. Carbone. Dummynet revisited. SIGCOMM

Comput

.

Commun

.

2010. [7]  B. Chun. PlanetLab: an overlay testbed for broad-coverage services. SIGCOMM Comput. Commun. Rev., 33(3):3–12, 2003. [8]  S. Jaiswal. Measurement and classification of out-of-sequence packets in a tier-1 IP backbone. IEEE/ACM Transactions on Networking (ToN), 2007. [9]  A. P. Jayasumana. Improved packet reordering metrics. RFC 5236, 1:401–414, June 2008. [10]  M. Laor. The effect of packet reordering in a backbone link on application through- put. IEEE Network, 16(5):28–36, 2002. 17

ReferencesSlide18

[11]  M. Lelarge. Packet reordering in networks with heavy-tailed delays. Mathematical Methods of Operations Research, 67(2):341–371, 2008. [12]  A. Morton. Packet reordering metrics. RFC 4737, 1:401–414, November 2006. [13]  A. Morton. Packet reordering metrics. IETF internet-standard: RFC4737, 2006.

[14]  V. Paxson. End-to-End Internet packet dynamics. Proc. ACM SIGCOMM Con- ference on Applications, Technologies, Architectures, and Protocols for Computer Communication, 1:401–414, 1997. [15]  N. M. Piratla

. On reorder density and its application to characterization of packet reordering. Proc. 30th IEEE Local Computer Networks (LCN) Conference, Sydney, Australia, 1:401–414, November 2005.

[16]  N. M.

Piratla

. Rd: A formal, comprehensive metric for packet reordering. Proc. IFIP Networking Conference, Ontario, Canada, LNCS 3462:78–79, May 2005.

[17]  N. M.

Piratla

. Reordering of packets due to multipath forwarding – An analysis. Proc. IEEE International Conference on Communications, 1:401–414, June 2006.

[18]  N. M.

Piratla

. Metrics for packet reordering – A comparative analysis. International Journal of Communication Systems, 21:99–113, 2008. [19]  J. Sommers. Improving accuracy in end-to-end packet loss measurement. ACM SIGCOMM Computer Communication Review, 35:157–168, August 2005. [20]  B. White. An integrated experimental environment for distributed systems and networks. Proc. of the Fifth Symposium on Operating Systems Design and Imple- mentation, Boston, MA, 1:255–270, December 2002. 18

ReferencesSlide19

Sequence Regeneration AlgorithmSophisticated algorithm needed because have to solve a c

onstraint problemNaïve approach wouldn’t workNeed a specific permutation that meets constraints19Slide20

20

RD generated from

Internet packet trace

145

hours of packet data

from the host in USA to one in India

Source

:

Colorado State UniversitySlide21

21

Evaluation

Plan followed

Take traces

software generated

and from

real

datasets

Calculate reordering metrics

Feed those metrics into my implementation

Measure metrics on the resulting stream, and show they are very close to the ones calculated in Step 2Slide22

22

Reordering scheduler

Delay/bandwidth/loss

emulation

Dummynet

Destination

Source

Sequence Regen.

Algorithm

O

ptional

config.

f

ile

f

or delay,

loss,

etc

Reordering

config. file

Input file

containing

RD for

emulation

Reordered packet stream

2

1

3

4

6 5 7 8 …Slide23

23

5 4 3 2 1 …

Reordering

scheduler

Delay/bandwidth/loss

emulation

Dummynet

D

Other

optional

c

onfig.

S

1 2 3 4 5 …

Sequence Regen.

Algorithm

Reorder

config

.

Input

RDSlide24

24

Workflow: Take packet trace -> calculate RD -> sequence regeneration algorithm -> feed it to dummynet -> emulationSlide25

25

Destination host in Cape Town

N = ~ 130K

Runtime = ~1sSlide26

SummaryReordering

Prevalent network phenomenonIncreasingly becoming important to pay attention toSophisticated metrics neededHenceImportant to include as a feature in emulatorsImplement support for RDCurrently most

sophisticated metric

available

Incomplete

without sequence regen algorithm

26Slide27

27

Input packets



Finite queue representing router buffer

Scheduler

Pipe representing a communications link, has an associated delay and bandwidth

Output packets

Slide28

Reorder Density (RD)Measures displacements of packets from their original positions in a sequence

Considers both early and late packet arrivalRelatively very comprehensive metric28