1
00:00:03,240 --> 00:00:07,521
And I just wanted to say quick note about 
interconnection networks. 

2
00:00:07,521 --> 00:00:12,036
if you guys are, get interested in 
interconnection networks as we go along, 

3
00:00:12,036 --> 00:00:16,024
I highly recommend this book. 
I have not assigned anything from this 

4
00:00:16,024 --> 00:00:19,778
book for this class. 
But this is Bill Dalley and Brian Towles' 

5
00:00:19,778 --> 00:00:22,006
interconnections book. 
It's quite good. 

6
00:00:22,006 --> 00:00:25,290
It's kind of the definitive guide on the 
subject matter. 

7
00:00:25,290 --> 00:00:26,080
Okay. 
So, 

8
00:00:26,080 --> 00:00:29,260
interconnection networks. 
We, 

9
00:00:29,260 --> 00:00:36,038
we talked about buses and we talked about 
memory protocol about buses. 

10
00:00:36,038 --> 00:00:45,060
That's only one way to share information. 
Now, people may argue that's a intuitive 

11
00:00:45,060 --> 00:00:49,822
way to share information because we had 
the ability to do load and storage from 

12
00:00:49,822 --> 00:00:53,739
our processor. 
But there are other ways to share 

13
00:00:53,739 --> 00:00:57,276
information. 
So, in today's lecture, we are going to 

14
00:00:57,276 --> 00:01:02,319
talk about two main pieces of two main 
topics. One, how do you share information 

15
00:01:02,319 --> 00:01:08,039
in, in a different way which commingles 
the movement of data with a 

16
00:01:08,039 --> 00:01:13,340
synchronization primitive. 
We are going to call that messaging, 

17
00:01:13,340 --> 00:01:20,755
[COUGH] which is in contrast to memory or 
communicating via memory addresses. 

18
00:01:20,755 --> 00:01:25,974
We're also going to talk about different 
ways to connect together processors, 

19
00:01:25,974 --> 00:01:30,786
which can either have better performance 
or better scalability, i.e., 

20
00:01:30,786 --> 00:01:37,565
have more nodes in the system. 
And, okay, so let's, let's compare and 

21
00:01:37,565 --> 00:01:44,767
contrast buses to other forms of network 
and see why we might want to change. 

22
00:01:44,767 --> 00:01:48,736
So, this, this is going back to what we 
had just talked about. 

23
00:01:48,736 --> 00:01:53,697
Let's say, you have two cores on a bus. 
And let's forget about how you want to 

24
00:01:53,697 --> 00:01:57,137
communicate. 
It could either be via shared memory or 

25
00:01:57,137 --> 00:02:01,437
it could be via messaging or it could be 
via some other protocol, 

26
00:02:01,437 --> 00:02:04,480
Ethernet, 
whatever, whatever you want put here, 

27
00:02:04,480 --> 00:02:09,208
which is basically a form of messaging. 
We have one core and it wants to 

28
00:02:09,208 --> 00:02:13,229
communicate with another core. 
Note, I don't draw any caches here 

29
00:02:13,229 --> 00:02:16,301
because there may not be caches, there 
may be caches. 

30
00:02:16,301 --> 00:02:18,882
It's doesn't, it's kind of immaterial 
here. 

31
00:02:18,882 --> 00:02:23,676
If one person wants to talk to another 
person, they can just yell at the other 

32
00:02:23,676 --> 00:02:26,502
person. 
It's two people so there's, it's pretty 

33
00:02:26,502 --> 00:02:29,145
easy to do. 
We, now, there is some challenges, 

34
00:02:29,145 --> 00:02:33,692
we can't both talk at the same time. 
We might not be able to understand each 

35
00:02:33,692 --> 00:02:36,457
other. 
So, there's some arbitration that needs 

36
00:02:36,457 --> 00:02:39,223
to happen. 
But, in general, that arbitration is 

37
00:02:39,223 --> 00:02:42,234
pretty simple. 
Only two, two cores or entities on this 

38
00:02:42,234 --> 00:02:43,656
bus. 
Okay. 

39
00:02:43,656 --> 00:02:49,006
Now we, now we go to more cores. 
So, we have four people in a room trying 

40
00:02:49,006 --> 00:02:53,676
to shout to each other. 
Well, or four people on a bus trying to 

41
00:02:53,676 --> 00:02:57,254
shout to each other. 
And as, as we just talked about, the 

42
00:02:57,254 --> 00:03:01,872
bandwidth can be a challenge here. 
The arbitration for the bus can be a 

43
00:03:01,872 --> 00:03:04,603
challenge. 
And because we're talking about 

44
00:03:04,603 --> 00:03:09,232
interconnection networks, 
the wire delay and capacitance of the 

45
00:03:09,232 --> 00:03:13,263
network can be worse or it can be a 
challenge here. 

46
00:03:13,263 --> 00:03:18,478
So, if we got one core, it needs to drive 
the shared multidrop bus. 

47
00:03:18,478 --> 00:03:24,563
There's a lot of capacitance on this bus, 
much more so than this case because all 

48
00:03:24,563 --> 00:03:30,173
of a sudden, we've, we've, doubled the 
length of the bus, so the wires are 

49
00:03:30,173 --> 00:03:33,966
longer and we've also put more loads on 
the bus. 

50
00:03:33,966 --> 00:03:37,760
So, there's actually more capacitance on 
this bus. 

51
00:03:40,040 --> 00:03:45,557
Okay. Now, we start to think about trying 
to build a bus that has a lot more cores. 

52
00:03:45,557 --> 00:03:50,114
In this case, twelve. 
And through this core, if you go shot to 

53
00:03:50,114 --> 00:03:54,395
that core, there's no pipelines around on 
this bus or anything. 

54
00:03:54,395 --> 00:03:59,137
You go to shout and has to propagate all 
the way down here and, you know, we're, 

55
00:03:59,137 --> 00:04:02,298
we're talking about high rates of 
communication. 

56
00:04:02,298 --> 00:04:07,764
You actually have to wait for the time of 
flight of light from here to get down to 

57
00:04:07,764 --> 00:04:10,680
there. 
And because we're, if we're using 

58
00:04:10,680 --> 00:04:14,861
something, let's say, like a snoopy 
protocol or a broadcast protocol, because 

59
00:04:14,861 --> 00:04:20,543
that's all we have here, 
we have to wait for and node here to 

60
00:04:20,543 --> 00:04:23,863
communicate with every other node. 
So, we have to wait for the worst case 

61
00:04:23,863 --> 00:04:26,860
time for this node to communicate to that 
node, every clock cycle. 

62
00:04:28,280 --> 00:04:31,717
Hm, okay. 
and as I said there is capacitance, so 

63
00:04:31,717 --> 00:04:36,609
it's not quite a, just a transmission 
line, so it's not just a transmission 

64
00:04:36,609 --> 00:04:40,823
line problem here. 
We also have to worry about [COUGH] the 

65
00:04:40,823 --> 00:04:45,214
capacitance in trying to drive all of 
these different receivers. 

66
00:04:45,214 --> 00:04:50,841
And it's a multidirectional bus so we 
have to have effectively tri-states and 

67
00:04:50,841 --> 00:04:56,396
the ability to drive or just receive. 
Well, all of a sudden, we have twelve 

68
00:04:56,396 --> 00:04:59,782
people and actually, we have twelve 
people in this room. 

69
00:04:59,782 --> 00:05:04,215
So, let's all try to pick a number 
between one and ten and shout it real 

70
00:05:04,215 --> 00:05:06,616
fast on the count of three. 
One, two, three, 

71
00:05:06,616 --> 00:05:07,232
five. 
Okay. 

72
00:05:07,232 --> 00:05:11,357
I could, I, I do, I shouted five, I don't 
know what everyone else said. 

73
00:05:11,357 --> 00:05:14,866
So, [LAUGH] that's does anyone, could 
everyone hear everyone else's? 

74
00:05:14,866 --> 00:05:19,484
Does everyone know exactly what all other 
ten people said at the same time, 

75
00:05:19,484 --> 00:05:23,855
or twelve people said at the same time? 
You heard your nearest neighbor. 

76
00:05:23,855 --> 00:05:27,180
Okay, but did you know, do you know what 
Yankey said? 

77
00:05:28,860 --> 00:05:31,480
Yeah. Okay. 
So, 

78
00:05:31,480 --> 00:05:35,568
this is, this is the challenge. 
And if we need to guarantee that only one 

79
00:05:35,568 --> 00:05:39,089
person can yell on the bus at a time, we 
need some arbitration. 

80
00:05:39,089 --> 00:05:43,291
But the arbitration logic is slower now 
because we have lots of people 

81
00:05:43,291 --> 00:05:47,753
communicating so we have to run a wire 
from this node down to this node and 

82
00:05:47,753 --> 00:05:52,202
then, we had to come back in the 
arbitraration, logical, say, over here 

83
00:05:52,202 --> 00:05:56,717
that needs to make some decision. 
And the decision is slower because as 

84
00:05:56,717 --> 00:06:01,886
more layers of logic, more combination of 
logic, we will say, to make arbitration 

85
00:06:01,886 --> 00:06:04,504
decision. 
Hm, okay. Now, if we go to a thousand 

86
00:06:04,504 --> 00:06:07,840
processors or a a thousand cores on a 
bus, 

87
00:06:07,840 --> 00:06:11,418
you know, we, we could even have twelve 
people in the room shout at the same 

88
00:06:11,418 --> 00:06:13,357
time. 
You can have a thousand people in the 

89
00:06:13,357 --> 00:06:16,836
room shout at the same time, and 
physically be distanced to the wiring 

90
00:06:16,836 --> 00:06:20,614
between this thousand different nodes is 
going to decrease the speed of the bus 

91
00:06:20,614 --> 00:06:23,100
significantly. 
So, it's some, something to think about. 

92
00:06:25,720 --> 00:06:31,484
So, this, this motivates us to take the 
same twelve course and think about some 

93
00:06:31,484 --> 00:06:37,312
other ways to connect them. 
Now, what I'm going to show here is a, 

94
00:06:37,312 --> 00:06:45,900
what's known as a switched interconnect 
or sometimes known as a point-to-point 

95
00:06:45,900 --> 00:06:49,218
link solution. 
Now, point-to-point does not mean that 

96
00:06:49,218 --> 00:06:52,856
this core can communicate directly with 
every other core. 

97
00:06:52,856 --> 00:06:57,451
That has, that has a different name, 
we'll talk about that later today. 

98
00:06:57,451 --> 00:07:01,000
Instead, point-to-point just means each 
link, 

99
00:07:01,000 --> 00:07:04,935
only has one sender and one receiver. 
And then. 

100
00:07:04,935 --> 00:07:10,240
you use switches along the way to make 
decisions and to route. 

101
00:07:11,660 --> 00:07:15,676
So, if we look at this, we can actually 
have multiple nearest neighbor 

102
00:07:15,676 --> 00:07:19,051
communication happening. 
So, all of a sudden, by adding this 

103
00:07:19,051 --> 00:07:23,649
switching, we can both have connectivity 
between all the different nodes, but we 

104
00:07:23,649 --> 00:07:27,080
can also have sort of subconversations 
happening. 

105
00:07:27,080 --> 00:07:32,230
But this still allows for this processor 
here to go communicate with the one 

106
00:07:32,230 --> 00:07:36,644
that's at the farthest extent. 
And we need to decide how to do that. 

107
00:07:36,644 --> 00:07:41,928
Whether it communicates sort of this way 
or this way or that way or some other 

108
00:07:41,928 --> 00:07:45,139
squiggly line. 
[COUGH] We can also take the same 

109
00:07:45,139 --> 00:07:48,015
point-to-point switch interconnect 
network. 

110
00:07:48,015 --> 00:07:51,895
And like a bus, which we can increase the 
width of the bus, 

111
00:07:51,895 --> 00:07:55,900
which does not help us with the occupancy 
on the bus, 

112
00:07:55,900 --> 00:08:00,476
we can add more networks or we can 
affectively add multiple concurrent, 

113
00:08:00,476 --> 00:08:05,311
switching interconnection networks or we 
can increase the bandwidth on these 

114
00:08:05,311 --> 00:08:09,307
buses. 
So, it's similar sorts of ideas there and 

115
00:08:09,307 --> 00:08:14,399
similar sorts of bandwidth tricks you can 
do to increase bandwidth on buses. 

116
00:08:14,399 --> 00:08:17,880
You can play on there, switch 
interconnection always. 

117
00:08:17,880 --> 00:08:22,392
Okay. So, this is just a very broad 
overview. And now, we're getting into 

118
00:08:22,392 --> 00:08:24,520
some, some more specific ideas.