RTT (sec)

0.00 0.05 0.10 0.15 0.20 0.25 0.30

0 1 2 3 4 5 6 7 8

149,930 149,960 149,990

6.8

3.79

2.63

2.0100

1.63 6.92

3.61

2.45

1.8500

1.49

0 1 2 3 4 5 6 7 8

0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.90

T h ro u g h p u t (p k ts /s e c )

RTT (sec)

Compression No Compression

(a)

0 5 10 15 20 25 30 35 40 45 50

-20%

-15%

-10%

-5%

0%

5%

10%

1 2 3 4 5

T h ro u g h p u t E ff ic ie n c y ( th ro u g h p u t/ s e c )

P e rc e n ta g e o f Im p ro v e m e n t

Increment of RTT (times)

No Compression Compression

(b)

Figure4.13: Case study of throughput versus

RT T

^{with dierent}

τ

Fig. 4.11 shows throughput and

RT T

^{urves versus}

λ

^{for the three dierent}

om-pressiontimes(

t c

^{)above. As mentionedearlier,throughputwith orwithoutompression}

willalways derease as

λ

^inreases,

RT T

^{withorwithoutompression gets worse aswell.}

RT T

^{with ompression will always be linearly shifted higher from without ompression}

aording to the

t c

^{value and always be worse in ase of} ompression. With extremely swift ompression, throughput is only better when inoming rate is already is very low.

This soon hanges when the input rate inreases even slightly; the advantage of

om-pression immediatelydisappears. In ase of more realistiompression delay even in the

future, both throughput and

RT T

^{are alwaysway worse thanwithout} ompression.

Fig. 4.12 shows the satter graph of throughputversus

RT T

^{to visualize the}

orrela-tion between the two dierent outputfor the dierent ompression times (

t c

^{) mentioned}

above. Eah urve shows 5 onstant input rates senario

λ

^{(149930, 149945, ..., 149990)}

pkts/se. In ase of 149,930 pkts/se (the red marked points), ompression time with 5

mshas 9.82%betterthroughput thanwithoutompression,although

RT T

^is3.0%more.

Whilethe ompression time with 25 ms, throughput is almost the same with or without

ompressionand

RT T

^{iseven longerforthe same}

λ

^{senario. The throughputurvewith}

5 ms ontinues to have higher throughput and longer

RT T

^{than without ompression}

until

λ

^{beomes 149,975 pkts/se, where the throughput is almostequal to the}

through-put without ompression urve. After that, the throughput of ompression goes below

throughput without ompression. The realisti ase of ompression time with 50 ms is

far worse with respet to both

RT T

^and throughput. In summary, if the ompression time an ever reah 5 ms or less while still maintainingvery high

CR

^{, there is a hane}

forompression toimprovethroughput,but onlywhenboth

B

^and

λ

^{are quitesmalland}

annot beinreased.

Fig. 4.13a shows the satter graph of throughput versus

RT T

^{to visualize the}

orre-lation between the two dierent output for the dierent propagation delay,

τ

^{from 150}

ms

∼

^{750 ms, with 25 msof onstant ompression time,}

t c

ⁱⁿ

RT T

^{. The}

RT T

^{results of}

ompressionare always way worse thanwithoutompression,but throughputare getting

better when

τ

ⁱⁿ

RT T

^{is inreased. To learly determine the eet of}

τ

ⁱⁿ

RT T

^{to the}

throughput,throughput eieny with unit of throughputper seondis omputed[117℄.

Fig. 4.13b shows when the inrement of

RT T

^{reahes 3 times of the starting}

τ

⁽⁴⁵⁰

ms), throughput eieny of ompression outperforms the throughput eieny of no

ompression.

4.4.6 Summary

All the urves show that throughput dereases with any of the following fators sorted

aordingtoimportane;

λ

^inrease,

B

^{derease or}

t c

^{inrease. Throughput}improvement is mostly aeted by

λ

^then

B

^{more than it is aeted by}

CR

^and

t c

^{. Assuming that}

the input rate (

λ

^{) is}unontrollable,then

CR

^and

t c

^{an be usedto improve the eet of}

B

^{onthroughput when more memory isnot possible.} Compressionwith more aggressive shemesevenif itisslow,an reallyenhane throughputwhenaddedontop ofinreasing

B

^{. For}

RT T

^{, it an only be dereased by nding extremely fast ompression whih}

approahes zero time, while stillprovidinggoodompression toimprovethroughput.

Eient Congestion Management

(ECM) Framework

5.1 Introdution

The purpose of ECM framework is to eiently implement dierent existing ongestion

avoidaneand ongestion ontrolapproahes inone adaptiveframeworkto minimizethe

impatofongestednetworkwhilebetterutilizingnetworkresoures. Therearenumerous

existingongestionavoidaneandongestionontrolapproahes,manyofwhihhavebeen

already disussed in the Chapter 2. Most of those approahes were designed with one or

more partiular purpose in mind, whih makes this framework useful as multipurpose

used inall network onditions. The ECMframeworkoers adaptiveseletionto manage

the dierent ongestion solving approahes aording to adaptive learning from history.

ECM is mostly a ongestion management framework, sending ontrolmessages to other

ongestion ontrolmehanismsto handlethe dierent network situations.

Figure5.1shows theblokdiagramof proposedECM frameworkinthe dierentsteps

of tra or operation ow. The framework is an overall network manager oordinator

between the dierent layers of the network stak, ranging from appliation layer (e.g.,

Moleular Sequene Redution) till the physial layer (e.g., network oding). The

addi-tional ongestion lassier introduedin the framework together with the aompanying

ontrolsignallingand tra forwarding links are used to ahieve the oordination of the

framework.

The ongestion detetor in the lassier is atually split into two detetion levels.

First, the initialdetetor orrst level,roughly and swiftly heks ongestion ourrene.

Aurayissariedintherst levelinfavourofbothspeedandenergy. Whendisabling

therestoftheunusedframeworkmodulesuntilongestionissuspeted,energythatwould

haveonsumedbythoseextramodulesouldbesaved. Additionally,takingthosemodules

of the ritial path of the framework operation helps speed up the framework operation

in normaltra status.

Local control signals

L o w e r L a y e rs

Network Coding

Database

Congestion Prediction Congestion

Detection Congestion Avoidance Congestion

Mending N e tw o rk C o n g e s tio n M e c h a n is m s M o le c u la r S e q u e n c e R e d u c tio n (C o m p re s s io n )

Pattern Discovery

Dictionary Synchronization

Compression Dictionary Updating Local

Monitoring

Interface To Lower Layers

TCP Header Management Segmentation

Physical Network Status

Decoupling Coupling Traffic Switching

and Tunneling

MPLS TE

ドキュメント内 JAIST Repository https://dspace.jaist.ac.jp/ (ページ 85-89)