yi-neng ’ s weekly report

39
Yi-Neng’s Weekly Re port

Upload: marlo

Post on 13-Jan-2016

44 views

Category:

Documents


3 download

DESCRIPTION

Yi-Neng ’ s Weekly Report. Half-year Plan (2004/01~2004/06). *See the footnote section for further notes. Weekly Report (1/5/ ’ 04~1/9/ ’ 04). Work done last week (12/29/03~12/31/03) Lab Modify the code according to the refined transition Status: not finished. L7 - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Yi-Neng ’ s Weekly Report

Yi-Neng’s Weekly Report

Page 2: Yi-Neng ’ s Weekly Report

Task Subtask1/2

1/16

1/30

2/13

2/27

3/12

3/26

4/9

4/23

5/7

5/21

6/4

Finish Date

Remark

NP modeling

Preliminary analytical

and simulation result

★ ★ ★ 1/30

Complete result ★ ★ ★ 3/12

Paper writing ★ ★ ★ ★ ☆ ☆ ☆ 4/9 ☆: prepare for MICRO-37

SG over NP

NetBSD porting ★ ★ ★ 1/30

System stability ★ ★ ★ ★ 2/27

System benchmark ★ ★ ★ ★ ★ 3/26 Help solve initial barrier

Paper writing ★ ★ ★ ★ ★ ★ 6/4

Contest 構想書、書面報告、 CIC 訓練

★ ★ ★ ★ ★ ★ ★ 5/15 5/15 決賽 (demo)

Intel proj Final report ★ ★ 6/4

Half-year Plan (2004/01~2004/06)

*See the footnote section for further notes

Page 3: Yi-Neng ’ s Weekly Report

Weekly Report (1/5/’04~1/9/’04) Work done last week (12/29/03~12/31/03)

Lab Modify the code according to the refined transition

Status: not finished. L7

Comparison table between NetScreen 5GT and IXP@L7 Try to modify the kernel configuration for the upgraded mass prod. bo

ard Plans in this week

Lab Please see the footnote for details

L7 Help test the functionalities of the ported WALL Prepare for the demo for Intel visitors

Status: Done today Submit Tunnel-Min. paper

Status: Done today

Page 4: Yi-Neng ’ s Weekly Report

Weekly Report (1/12/’04~1/16/’04) Work done last week (01/05/’04~12/31/’04)

Lab Finish (2,2) simulation model generation Modify the code according to the refined transition

Status: not finished.

L7 Presentation for Intel visitors Identify the cause of memory shortage.

Plans in this week Lab

Submit the Tunnel Min. paper (will discuss it in Thu.’s meeting) Help Chiuan-Hung on his slides and presentation skills Please see the footnote for other details

L7 Help test the functionalities of the ported WALL

Page 5: Yi-Neng ’ s Weekly Report

Weekly Report (02/02/’04~02/06/’04) Work done last week (01/12/’04~01/16/’04)

Lab Submit the Tunnel Min. paper (will discuss it in Thu.’s meetin

g) Help Chiuan-Hung on his slides and presentation skills

L7 Help test the functionalities of the ported WALL

Plans in this week Register the Embedded System Contest (including the

3-page 創意構想書 )– Done Discuss with Chiuan-Hung on the external benchmark

items Study the IXP425 documents for internal benchmark Please see the footnote for other details

Page 6: Yi-Neng ’ s Weekly Report

Weekly Report (02/09/’04~02/13/’04) Work done last week (02/02/’04~02/06/’04)

Register the Embedded System Contest (including the 3-page 創意構想書 )– Done

Discuss with Chiuan-Hung on the external benchmark items

Study the IXP425 documents for internal benchmark

Plans in this week Refine the 創意構想書 and finish the registration

Done Prepare for the 卓越計畫 review demo Discuss with Chiuan-Hung on the internal benchmark ite

ms Please see the footnote for other details

Page 7: Yi-Neng ’ s Weekly Report

Weekly Report (02/16/’04~02/20/’04) Plans in last week

Refine the 創意構想書 and finish the registration (done) Prepare for the 卓越計畫 review demo (done) Discuss with Chiuan-Hung on the internal benchmark ite

ms (not finished yet) Detailed outline of my paper (done) Explain the mismatch between sim. and ana. results.

Plans for this week Finish the methodology for internal benchmark Please see the footnote for other details

Page 8: Yi-Neng ’ s Weekly Report

Weekly Report (02/23/’04~02/27/’04) Plans in last week

Finish the methodology for internal benchmark Discuss with Chiuan-Hung Refine the detailed outline of my paper

Plans for this week Identify the target of Chiuan-Hung’s thesis

Done. It will be HPCA-11 which dues in mid-July Please see the footnote for other details

Page 9: Yi-Neng ’ s Weekly Report

Weekly Report (03/01/’04~03/05/’04) Plans in last week

Verify again the target of my paper Discuss with Chiuan-Hung Complete chapter 1 Expand (I,J)

Plans for this week Discuss with Chiuan-Hung (Wed.) Please see the footnote for other details

Page 10: Yi-Neng ’ s Weekly Report

Weekly Report (03/08/’04~03/12/’04) Plans in last week

Discuss with Chiuan-Hung on the benchmark results Verify in detail and logically the possible figures to have Adjust the tunnel_min paper for Globecom ‘04

Plans for this week Discuss with Chiuan-Hung on internal benchmark items Please see the footnote for other details

Page 11: Yi-Neng ’ s Weekly Report

Weekly Report (03/15/’04~03/19/’04) Plans in last week

Discuss with Chiuan-Hung on the internal benchmark items

Compute the total cycle counts of IP Forwarding and DiffServ

Extend the simulation model from (1,1) to (1,4) Implement some other statistic scripts

Plans for this week Chiuan-Hung: detailed outline and review his introduction Interview with Intel crew Please see the footnote for other details

Page 12: Yi-Neng ’ s Weekly Report

Weekly Report (03/22/’04~03/26/’04) Plans in last week

Prepare for the Intel’s interview Refine the simulation model Chiuan-Hung: detailed outline and review his introduction

Plans for this week Hacking of the porting: come up with the overview Review chapter 2 of Chiuan-Hung’s thesis Please see the footnote for other details

Page 13: Yi-Neng ’ s Weekly Report

Weekly Report (03/29/’04~04/02/’04) Plans in last week

Section 2 of my paper, almost done Solve the un-dividable problem (# of mem acc vs. # of th

d) Porting hacking

Discussed with Chiuan-Hung. Have come up with an overview.

Plans for this week Please refer to the footnote for details

Page 14: Yi-Neng ’ s Weekly Report

Weekly Report (04/05/’04~04/09/’04) Plans in last week

Solve the “un-dividable” problem Implement memory disk Extend (2,J)’s Regular discussion with Chiuan-Hung

Plans for this week Revise Chiuan-Hung’s paper: he has not finished it yet Finalize related figures with Chiuan-Hung Slides for presentation at Intel Please refer to the footnote for details

Page 15: Yi-Neng ’ s Weekly Report

Weekly Report (04/12/’04~04/16/’04) Plans in last week

Discussions with Chiuan-Hung on the performance figures Reformat, reorganize Chiuan-Hung’s slides for Intel presentation; redra

w some figures Prepare for the midterm of data mining

Plans for this week Revise Chiuan-Hung’s chapter 2 Please see the footnotes

for detailsPerformance improvement by multithreading

(SF: 235 inst, 12 mem DS: 555 inst 30 mem)

0

10

20

30

40

50

60

(1,2) (1,3) (1,4) (1,5)

(I,J) combinations

Impr

ovem

ent (

%)

Simple Forwarding

DiffServ

Page 16: Yi-Neng ’ s Weekly Report

Weekly Report (04/19/’04~04/23/’04) Plans in last week

Revise Chiuan-Hung’s chpt 2; Prepare for the presentation in group meeting

Plans for this week Finish revising Chiuan-Hung’s chpt2 Discuss Chiuan-Hung’s progress of chpt3 Review Globecom paper Please refer to the footnote for details

Page 17: Yi-Neng ’ s Weekly Report

Weekly Report (05/03/’04~05/07/’04) Plans in last week

Contest report Simulation based on parameters of IXP2400 Discussion with Chiuan-Hung on the porting details Course presentation ( 生物資訊導論 )

Plans for this week 5/15 presentation & demo at 雲科大 for contest Please refer to the footnote for details

Page 18: Yi-Neng ’ s Weekly Report

Weekly Report (05/17/’04~05/21/’04) Plans in last week

Presentation in group meeting Contest presentation and demo Review the analytical & simulation models of blocking

case

Plans for this week Data Mining final (Monday, done)

Page 19: Yi-Neng ’ s Weekly Report

Weekly Report (06/07/’04~06/11/’04) Plans in last week

Submit the paper to MICRO37 Revise Chiuan-Hung’s thesis

Plans for this week Data Mining term project 生物資訊 term paper

Page 20: Yi-Neng ’ s Weekly Report

Weekly Report (07/26/’04~07/30/’04) Plans in last week

Submit Chiuan-Hung’s paper to IEEE Comm. Mag. Revise I-Wei’s paper. NP-Complete proof State-space reduction

Plans for this week Re-submit I-Wei’s paper to TDSC Transition rates for the reduced state space

Page 21: Yi-Neng ’ s Weekly Report

Weekly Report (08/02/’04~08/06/’04) Plans in last week

Revise I-Wei’s paper. NP-Complete proof State-space reduction

Status: came up with a new model Try out WorkBench

Plans for this week Re-submit I-Wei’s paper to TDSC

Have asked TDSC for another weeks of extension State-space reduction (ex: merging) Discussion with Yao-Chung; survey of related works

Page 22: Yi-Neng ’ s Weekly Report

Weekly Report (08/09/’04~08/13/’04) Plans in last week

State-space reduction Status: came up with a new model

Try out Teja software Plans for this week

Paper review State-space reduction Discussion with Yao-Chung: Inside the Snort Re-submit I-Wei’s paper to TDSC

Page 23: Yi-Neng ’ s Weekly Report

Weekly Report (08/23/’04~08/27/’04) Plans in last week

State-space reduction Status: came up with a new model

Try out Teja software Plans for this week

Camera-ready version for Globecom State-space reduction Discussion with Yao-Chung: Thread/ME allocation

Page 24: Yi-Neng ’ s Weekly Report

Weekly Report (08/30/’04~09/03/’04) Plans in last week

Think/Discussion over the review comments Globecom: camera-ready version Discussion/Review on Yao-Chung’s proposal

Plans for this week Discussion with Yao-Chung Study the Jackson Networks and Mean Value Analysis Please see the footnote for details

Page 25: Yi-Neng ’ s Weekly Report

Weekly Report (10/04/’04~10/08/’04) Plans in last week

Submission to ISPASS and RTAS State & transition diagram for the new model Discussion with Yao-Chung over the implementation me

thod Plans for this week

Matrix generation, preliminary result for the new model Discussion with Yao-Chung: new data structure for impl

ementing AC algorithm Please see the footnote for details

Page 26: Yi-Neng ’ s Weekly Report

Weekly Report (10/11/’04~10/15/’04) Plans last week

Model determination, matrix generation Done

Discussion with Yao-Chung: new data structure for implementing AC algorithm

Please see the footnote for details Plans for this week

Preliminary result for the new model Please see next three slides for the details of the new m

odel

Page 27: Yi-Neng ’ s Weekly Report

Simplified Model (040924 )

MMPP0: switched out

, where , a

Arrival rate =

)atencymean_mem_l/(1J

M

J

Pu

0)1( rr a

State def: , where

q

),,( sa mmq

smam

Modeling granularity: single thread

: num of pkts in the queue of the thread

: arrival state; 0 or 1

: service state; 0 or 1B memory banks

…Mem subsystem

m

(M/M/c)

MMPP 0: switched out

Jr 1

Page 28: Yi-Neng ’ s Weekly Report

Two Kinds of Transitions

(0,0)q => q+1: rate = 0

q => q -1: rate = 0

(0,1)q => q+1: rate = 0

q => q -1: rate =

a(1,0)q => q+1: rate =

q => q -1: rate = 0

else: rate = 0

* (1,1) is illegal !

(0,0)

(1,0)(0,1)

0 or

N

N

N )1(

N

N

N )1(

0or N

N

only changes q does not change

(if q<>0)

(if q=0)

N : mean # of non-empty idle threads

q

smam

),,( sa mmqState def:

To obtain:

1. => get s-s JN )0,,0P( amq 2. , get JmqN a )0,,0P( )0,,0P( amq 3. Repeat until is stableN

Page 29: Yi-Neng ’ s Weekly Report

Let be the steady-state prob of being in state , we can have:

Results to Obtain),,(̂ sa mmqh

1. Queue size distribution :qx

),,( sa mmq

a

sam sm

mmqq hx ),,(̂

2. Mean thread service time = , system time =

, and therefore mean queue size q

qxq q

/q /qJ

3. Prob that a thread is

busy: )1,0P(P sbusy mq

active: ,u

JPbusyactive 1

/PP

4. Processor utilization: JactiveP

)atencymean_mem_l/(1J

M

J

Pu

, processor queue size = Jq

Page 30: Yi-Neng ’ s Weekly Report

Weekly Report (10/25/’04~10/29/’04) Plans last week

Determine the transition rate of the new model Discuss with Yao-Chung on the description of the dispat

ching algorithm No finished yet since there are still some unresolved design iss

ues

Plans for this week Preliminary analytical results for the new model Please see the next three slides for details

Page 31: Yi-Neng ’ s Weekly Report

Simplified Model (041015 )

State def: , where

q

),,,( accmmq sa

smam

Modeling granularity: single thread

: num of pkts in the queue of the thread

: arrival state; {0, 1}

: service state; 0 to (round it to an integer)J

M

Transition of : (assume )),( accms 2J

M

(2,0) (2,1) (2,2) (1,0) (1,1) (0,0)inst rdymem inst mem

(2,0)

0 if q0 if q

acc : thread state; {0, 1, 2}

0 if q

Page 32: Yi-Neng ’ s Weekly Report

Two Types of Transitions (1/2)

Only changesq

1. Arrival: 1qq

1am : rate = , wherea 0)1( rr a Jr 1

2. Departure: 1qq

),( accms : (1,1) => (0,0)

rate = , where PM

MMPP0: switched out

, where , a

Arrival rate 0)1( rr a

MMPP

Jr 1

)0,0( accms)1,0( accms)2,0( accms

: inst processing

: memory access

: ready

)0,0( accms : idle

Page 33: Yi-Neng ’ s Weekly Report

Two Types of Transitions (2/2)

(0,0,0)

(1,0,0)

0 or

N

N

N )1(

N

N

N )1(

0or N

N

q does not change

(if q<>0)

(if q=0)

N : mean # of non-empty idle threads

ninteractio ),,( accmm sa

To obtain:

1. => get s-s JN )0,,0P( amq 2. , get JmqN a )0,,0P( )0,,0P( amq 3. Repeat until is stableN

1. )1,()0,( ss mm rate = , where P

M

)0,0,0( sm

),( accms changes

2. )1,( sm

rate = latencymem _1

0 and 1 if ,2),( qmm ss

0 and 1 if ,0),0( qmsrate = latencymem _1

Page 34: Yi-Neng ’ s Weekly Report

Weekly Report (11/08/’04~11/12/’04) Plans last week

Obtain the analytical result Discuss with Yao-Chung

Plans for this week Debug/Refine the model/code Please see the next five slides for details

Page 35: Yi-Neng ’ s Weekly Report

Simplified Model (041103 )

State def: , where

n

),,,( sman

ma

Modeling granularity: single thread

: num of pkts in the thread system

: arrival state; {0, 1}

: num of mem accesses for the thread; 0 toJ

MR s : thread state; {0, 1, 2}

(round it off)

MMPP0: switched out

, where , a

Arrival rate 0)1( rr a

MMPP

Jr 1

)0,0( sm)1,0( sm)2,0( sm

: inst processing

: memory access

: ready

)0,0( sm : idle

a

n

aT

sT

(I,J)=1x2

Page 36: Yi-Neng ’ s Weekly Report

Model Details

Illegal states:

)0,0,,( san)0,0,1,( mn)2,1,,( an

: two threads are working concurrently: no such states

Example transition for : (assume )2J

M

inst rdymem

instmem

nnn 0 if

)0,2,,( an )1,2,,( an )2,2,,( an

)0,1,,( an)1,1,,( an)0,0,,( anidle

: no such states

sT

1 nn

)0,0,,0( sma : no such states

Page 37: Yi-Neng ’ s Weekly Report

Types of Transitions (1/2)

1. Arrival:

),,1,1(),,1,( :ex smnsmn (fixed) 1a

rate = , wherea 0)1( rr a Jr 12. Departure: 1nn

1nn

(fixed) 0a

3. Memory access:

fixed are and both an

)0,0,0,1()1,1,0,( :ex nn

)2,,,()1,,,( :ex manman rate = latencymem _/1

rate = latencymem _/1

Page 38: Yi-Neng ’ s Weekly Report

Types of Transitions (2/2)4. Context switch:

N

uas mnmnTT rate , )1,0,1,()0,0,0,(:

uN

Ns mnmnT )1(rate , )1,0,0,()0,0,0,(:other

N

usa mnmnTT rate , )0,1,0,()2,0,1,( :

N

uRnn rate , )0,,0,1()0,0,1,(

uN

Na mnmnT )1(rate , })2,1{,0,0,(})2,1{,0,1,(:other

N

ua smnsmnT rate , ),,1,(),,0,(:other

N

us mnmnT rate , )0,1,0,()2,,0,( :other

N

uRnn rate , )0,,0,1()0,0,0,(

N : mean # of ready threads

To obtain: use first

1. Get s-s , )2,0,,P( man2. )1(,0,0),0P()2,0,,P( JanmanN

3. Repeat 1, 2 until is stableN

JN

)0,0,,0P( an

PM

Page 39: Yi-Neng ’ s Weekly Report

Let be the steady-state prob of being in state , we can have:

Results to Obtain),,,(̂ smanh

1. Queue size distribution :nx

),,,( sman

a m

smans

n hx ),,,(̂

2. Mean thread service time = , system time =

, and therefore mean queue size n

nxn n

/n /nJ

3. Prob that a thread is

busy: )0,0P(P smbusy

active: )0,0P(P smactive

4. Processor utilization: JactiveP

, processor queue size = Jn