SLIDE 1
Sanidhya Kashyap* Irina Calciu Xiaohe Cheng Changwoo Min Taesoo Kim
1
2
Sanidhya Kashyap * Irina Calciu Xiaohe Cheng Changwoo Min - - PowerPoint PPT Presentation
Sanidhya Kashyap * Irina Calciu Xiaohe Cheng Changwoo Min - - PowerPoint PPT Presentation
Sanidhya Kashyap * Irina Calciu Xiaohe Cheng Changwoo Min Taesoo Kim 1 Operations / second # threads 2 Operations / second # threads 3 Operations / second # threads 4 5 1 socket > 1 socket Oversubscribed
SLIDE 2
SLIDE 3
3
# threads Operations / second
SLIDE 4
4
# threads Operations / second
SLIDE 5
5
SLIDE 6
6
Operations / second # threads
1 socket > 1 socket
Stock
Oversubscribed
SLIDE 7
7
Operations / second # threads
1 socket > 1 socket
Stock
Oversubscribed
SLIDE 8
8
Operations / second # threads
1 socket > 1 socket
Stock
Oversubscribed
SLIDE 9
9
Operations / second # threads
1 socket > 1 socket
Stock
Oversubscribed
SLIDE 10
- ○
○
- ○
○
- 10
SLIDE 11
11
Operations / second # threads
- Oversubscribed
> 1 socket 1 socket
- Stock
CST
SLIDE 12
12
Operations / second # threads
- Oversubscribed
> 1 socket 1 socket
- Stock
CST
SLIDE 13
13
Operations / second # threads
- Stock
CST
Oversubscribed > 1 socket 1 socket
SLIDE 14
14
Operations / second # threads
- Stock
CST
Oversubscribed > 1 socket 1 socket
SLIDE 15
15
- Locks’ memory footprint
140 # threads 18 Stock CST
SLIDE 16
16
- Stock
CST Locks’ memory footprint 140 # threads 18 820 Hierarchical lock
SLIDE 17
17
- Stock
CST Locks’ memory footprint 140 # threads 18 820 Hierarchical lock
SLIDE 18
18
SLIDE 19
19
SLIDE 20
20
t1 shuffler:
Socket ID tail
waiter’s qnode:
SLIDE 21
21
shuffler:
Socket ID tail
waiter’s qnode: t1 t2
SLIDE 22
22
shuffler:
Socket ID tail
waiter’s qnode: t1 t2 t3 t4
SLIDE 23
23
shuffler:
Socket ID tail
waiter’s qnode: t1 t2 t3 t4
SLIDE 24
24
- →
t1 t2 t3 t4
SLIDE 25
25
SLIDE 26
26
SLIDE 27
27
SLIDE 28
28
SLIDE 29
29
SLIDE 30
30
SLIDE 31
31
- ○
○
SLIDE 32
32
- ○
○ unlock(): lock():
SLIDE 33
33
SLIDE 34
34
- ○
○ ★
- ○
SLIDE 35
35
unlocked locked t0 shuffler:
Socket ID tail
waiter’s qnode:
SLIDE 36
36
unlocked locked t0 t0 shuffler:
Socket ID tail
waiter’s qnode:
SLIDE 37
37
unlocked locked t0 t1 t2 t3 t4 shuffler:
Socket ID tail
waiter’s qnode:
SLIDE 38
38
unlocked locked t0 t1 t2 t3 t4 shuffler:
Socket ID tail
waiter’s qnode:
SLIDE 39
39
unlocked locked t0 t1 t3 t2 t4 shuffler:
Socket ID tail
waiter’s qnode:
SLIDE 40
40
unlocked locked t0 t1 t3 t2 t4 shuffler:
Socket ID tail
waiter’s qnode:
SLIDE 41
41
unlocked locked t0 t1 t3 t2 t4 shuffler:
Socket ID tail
waiter’s qnode:
SLIDE 42
42
unlocked locked t0 t1 t3 t2 t4 shuffler:
Socket ID tail
waiter’s qnode:
SLIDE 43
43
unlocked locked t1 t3 t2 t4 shuffler:
Socket ID tail
waiter’s qnode:
SLIDE 44
44
unlocked locked t1 t3 t2 t4 shuffler:
Socket ID tail
waiter’s qnode:
SLIDE 45
45
unlocked locked t1 t1 t3 t2 t4 shuffler:
Socket ID tail
waiter’s qnode:
SLIDE 46
46
unlocked locked t1 t1 t3 t2 t4 t1 t3 t2 t4 shuffler:
Socket ID tail
waiter’s qnode:
SLIDE 47
- ○
○
- ○
- ○
○
47
https://github.com/sslab-gatech/shfllock
SLIDE 48
- ○
○
48
SLIDE 49
49
Operations / second # threads Stock CST
Oversubscribed > 1 socket 1 socket
- SHFLLOCK
SLIDE 50
50
Operations / second # threads
Oversubscribed > 1 socket 1 socket
- ○
Stock CST SHFLLOCK
SLIDE 51
51
Operations / second # threads
Oversubscribed > 1 socket 1 socket
- ○
- ○
Stock CST SHFLLOCK
SLIDE 52
52
Operations / second # threads
Oversubscribed > 1 socket 1 socket
- ○
- ○
- ○
Stock CST SHFLLOCK
SLIDE 53
53
Stock CST 140 # threads SHFLLOCK 11 18
- ➢
➢
Locks’ memory footprint
SLIDE 54
54
140 # threads 11 18
- ➢
➢
Locks’ memory footprint Stock CST SHFLLOCK
SLIDE 55
55
Stock SHFLLOCK
Improves throughput by up to 1.5x Decreases memory footprint up to 93%
Lock’s memory Messages / second # threads # threads CST
SLIDE 56
56
Stock SHFLLOCK
Improves throughput by up to 1.5x Decreases memory footprint up to 93%
Lock’s memory Messages / second # threads # threads CST
SLIDE 57
57
- ○
○
- ○
- ○
○
SLIDE 58
58
- ○
○
- ○
- ○