1 Course Matters. NUS.SOC.CS5248 OOI WEI TSANG 2 Paper Review October 2003 MonTueWedThuFriSatSun 171819 20212223242526 Paper Reviews Due 20% off for each.
Post on 29-Jan-2016
213 Views
Preview:
Transcript
1
Course Matters
NUS.SOC.CS5248OOI WEI TSANG
2
Paper Review
October 2003
Mon Tue Wed Thu Fri Sat Sun
17 18 19
20 21 22 23 24 25 26
Paper Reviews Due20% off for each late day
NUS.SOC.CS5248OOI WEI TSANG
3
Diwali/Deepavali
October 2003
Mon Tue Wed Thu Fri Sat Sun
17 18 19
20 21 22 23 24 25 26
No Class
NUS.SOC.CS5248OOI WEI TSANG
4
Guest Lecture
October 2003
Mon Tue Wed Thu Fri Sat Sun
17 18 19
20 21 22 23 24 25 26
CS6212 Guest Lecture“A/V Synchronization”SR1 4-6pm
NUS.SOC.CS5248OOI WEI TSANG
5
Books
NUS.SOC.CS5248OOI WEI TSANG
6
Project Grading
Proposal5%
Paper20%
Poster20%
Contribution20%
Methodology 35%
7
Previously, on CS5248
NUS.SOC.CS5248OOI WEI TSANG
8
Cello
OS disk schedulerSupport multiple classesProtect classes from each otherWork conservingUse two-level scheduling
NUS.SOC.CS5248OOI WEI TSANG
9
CPU vs Disk SchedulingTask computation time
unpredictable
Task can be preemptive
NUS.SOC.CS5248OOI WEI TSANG
10
Problem
How can multimedia applications co-exists with normal applications?
NUS.SOC.CS5248OOI WEI TSANG
11
Same Old Idea
NUS.SOC.CS5248OOI WEI TSANG
12
Same Gang
P. Goyal, X. Guo, and H. M. Vin.
“A hierarchical CPU scheduler formultimedia operating systems.”
OSDI’96
NUS.SOC.CS5248OOI WEI TSANG
13
So Far
Best-Effort Real-Time
Scheduling
NUS.SOC.CS5248OOI WEI TSANG
14
How to Guarantee Services?
15
Resource Reservation: CPU
NUS.SOC.CS5248OOI WEI TSANG
16
Resource Reservations
How to reserve?Overbook?Overuse?
NUS.SOC.CS5248OOI WEI TSANG
17
Example: Memory
How to reserve?malloc
Overbook? return NULL
Overuse? crash, throw exception
NUS.SOC.CS5248OOI WEI TSANG
18
Example: Memory
How to reserve?“I will need 128KB”
Overbook? “Sorry!”
Overuse? malloc returns NULL
NUS.SOC.CS5248OOI WEI TSANG
19
Example: CPU
How to reserve?Overbook?Overuse?
Memory is discrete. CPU time is continuous.
20
How to Reserve?
NUS.SOC.CS5248OOI WEI TSANG
21
CPU Reservation
“I need C units of time, out of every T units.”
NUS.SOC.CS5248OOI WEI TSANG
22
Two Problems
How to determine C?take measurement and estimate
What if C is not constant?take the maximum
NUS.SOC.CS5248OOI WEI TSANG
23
Effects of T and C
0%10%20%30%40%50%60%70%80%90%
100%
0 10 20 30 40 50 60Reservation Period (ms)
% F
ram
es
Re
ndere
d
20% 15% 10%
24
CPU Scheduling Algorithm
25
Classic Real-Time Scheduler
NUS.SOC.CS5248OOI WEI TSANG
26
Rate Scheduling AlgorithmsRate MonotonicEarliest Deadline First
NUS.SOC.CS5248OOI WEI TSANG
27
Rate Monotonic
NUS.SOC.CS5248OOI WEI TSANG
28
EDF
NUS.SOC.CS5248OOI WEI TSANG
29
Classic Schedulability TheoremsRate Monotonic Scheduler
Earliest Deadline First2ln
i
i
T
C
2lni
i
T
C
1i
i
T
C
NUS.SOC.CS5248OOI WEI TSANG
30
Assumptions
programs are periodiccomputation time is constantzero context switch timeprograms are independent
NUS.SOC.CS5248OOI WEI TSANG
31
Why do we care about RM?Theoretically, RM has lower CPU
utilization (70%) butRM is more efficient than EDFCan support up to 80-90% CPU
utilization in practiceDon’t want to commit 100% CPU
anyway (for emergency)
32
Rialto Scheduler
From MS Research
NUS.SOC.CS5248OOI WEI TSANG
33
Overview
Given a set of task, pre-compute a scheduling graph
NUS.SOC.CS5248OOI WEI TSANG
34
Task A B C D E F
Amount 4 3 2 1 1 5
Period 20 10 40 20 10 40
EB CA
EB FD
EB A
EB D
NUS.SOC.CS5248OOI WEI TSANG
35
Task A B C D E F
Amount 4 3 2 1 1 5
Period 20 10 40 20 10 40
C
F
A
D
EB
NUS.SOC.CS5248OOI WEI TSANG
36
Advantages
Only need to make new scheduling decisions when new task is added
NUS.SOC.CS5248OOI WEI TSANG
37
Computing Scheduling GraphInput: A set of tasks with their
reservationsOutput: A scheduling graph if
feasible
NUS.SOC.CS5248OOI WEI TSANG
38
Many Possible Solutions
C
F
A
D
EBF
C
A
D
EB
NUS.SOC.CS5248OOI WEI TSANG
39
Task A B C D E F
Amount 4 3 2 1 1 5
Period 20 10 40 20 10 40
C
F
F
A
D
EB
NUS.SOC.CS5248OOI WEI TSANG
40
Computing Scheduling GraphInput: A set of tasks with their
reservationsOutput: A scheduling graph if
feasibleGoals: (a) Minimize context
switches (b) Distribute free time evenly
NUS.SOC.CS5248OOI WEI TSANG
41
Life is Tough..
NP-Hard
NUS.SOC.CS5248OOI WEI TSANG
42
Heuristic
10
10
10
10
10
10
10
NUS.SOC.CS5248OOI WEI TSANG
43
10
5
10
10
5
10
5
F
Task A B C D E F
Amount 4 3 2 1 1 5
Period 20 10 40 20 10 40
NUS.SOC.CS5248OOI WEI TSANG
44
6
1
6
6
1
6
1
F
B+E
Task A B C D E F
Amount 4 3 2 1 1 5
Period 20 10 40 20 10 40
NUS.SOC.CS5248OOI WEI TSANG
45
4
1
1
1
1
1
1
F
B+E
A+D
C
Task A B C D E F
Amount 4 3 2 1 1 5
Period 20 10 40 20 10 40
NUS.SOC.CS5248OOI WEI TSANG
46
Task A B C D E F
Amount 4 3 2 1 1 5
Period 20 10 40 20 10 40
EB C
EB DA
EB F
EB A D
47
Time Constraints
NUS.SOC.CS5248OOI WEI TSANG
48
Time Constraint
“I need to run the following code, that takes 30ms, 100ms from now, and the deadline is (now+180ms).”
NUS.SOC.CS5248OOI WEI TSANG
49
Pseudocode
can = begin_constraint(start_time, deadline, estimate)
if can schedule thendo work
elseadapt
time_taken = end_constraint()
NUS.SOC.CS5248OOI WEI TSANG
50
Scheduling Constraints
C
F
F
A
D
EB
NUS.SOC.CS5248OOI WEI TSANG
51
Scheduling Constraints
C
F
F
A
D
EB
C1, C2, C5
NUS.SOC.CS5248OOI WEI TSANG
52
Advantages of Scheduling GraphOnly need to make new
scheduling decisions when new task is added
Feasibility of time constraint can be predicted accurately
53
Peaceful Coexistence
NUS.SOC.CS5248OOI WEI TSANG
54
Running Unreserved Task
C
F
F
A
D
EB
55
Media Player on Rialto/NT
NUS.SOC.CS5248OOI WEI TSANG
56
Look inside NT SchedulerScheduling unit : ThreadPriority level: 1-15, 16-31
rt 24 high 13 normal 8 idle 4
NUS.SOC.CS5248OOI WEI TSANG
57
Priority Boosting
Example:Priority +6 after awaken by
keyboard input
NUS.SOC.CS5248OOI WEI TSANG
58
Anti-Starvation
If hasn’t run for 3 secondsBoost priority to 14
Avoid “Priority Inversion”
NUS.SOC.CS5248OOI WEI TSANG
59
Mars Pathfinder
NUS.SOC.CS5248OOI WEI TSANG
60
Rialto/NT
NT Scheduler
RialtoC
F
F
A
D
EB
Boost Priorityto 30
NUS.SOC.CS5248OOI WEI TSANG
61
Win Media Player 6.4
Period Priority Thread
10 24 Kernel Mixer
45 8 GUI
100 15 Timer
100 9 MP3 Decoder
500 8 Unknown
2000 8 Disk Reader
NUS.SOC.CS5248OOI WEI TSANG
62
Experiment 1
WMP running alone
Period Priority Thread
10 24 Kernel Mixer
45 8 GUI
100 15 Timer
100 9 MP3 Decoder
500 8 Unknown
2000 8 Disk Reader
NUS.SOC.CS5248OOI WEI TSANG
63
Experiment 2
WMP running with priority 8 competitor
Period Priority Thread
10 24 Kernel Mixer
45 8 GUI
100 15 Timer
100 9 MP3 Decoder
500 8 Unknown
2000 8 Disk Reader
NUS.SOC.CS5248OOI WEI TSANG
64
Experiment 3
WMP running with priority 10 competitor
Period Priority Thread
10 24 Kernel Mixer
45 8 GUI
100 15 Timer
100 9 MP3 Decoder
500 8 Unknown
2000 8 Disk Reader
NUS.SOC.CS5248OOI WEI TSANG
65
Experiment 5
WMP running with priority 10 competitor
Period Priority Reservation Thread
10 24 Kernel Mixer
45 8 GUI
100 15 Timer
100 9 40/1024 MP3 Decoder
500 8 Unknown
2000 8 Disk Reader
NUS.SOC.CS5248OOI WEI TSANG
66
Experiment 6
WMP running with priority 10 competitor
Period Priority Reservation Thread
10 24 1/16 Kernel Mixer
45 8 GUI
100 15 Timer
100 9 40/1024 MP3 Decoder
500 8 Unknown
2000 8 Disk Reader
NUS.SOC.CS5248OOI WEI TSANG
67
Experiment 7
WMP running with priority 10 competitor
Period Priority Reservation Thread
10 24 Kernel Mixer
45 8 GUI
100 15 Timer
100 9 20/512 MP3 Decoder
500 8 Unknown
2000 8 Disk Reader
68
Summary
NUS.SOC.CS5248OOI WEI TSANG
69
CPU Scheduling
With guaranteepercentage of CPU timetask with deadline
AlgorithmsRate-monotonicEDFRialto
NUS.SOC.CS5248OOI WEI TSANG
70
CS5248 Summary
Network
Encoder
Sender
Middlebox
Receiver
Decoder
NUS.SOC.CS5248OOI WEI TSANG
71
Last Two Lectures
October/November 2003
Mon Tue Wed Thu Fri Sat Sun
17 18 19
20 21 22 23 24 25 26
27 28 29 30 31 1 2
3 4 5 6 7 8 9
NUS.SOC.CS5248OOI WEI TSANG
72
Last Two Lectures
Papers from recent conferences and journals.
top related