1
00:00:03,240 --> 00:00:08,635
And I just wanted to say quick note about
interconnection networks. if you guys are,

2
00:00:08,635 --> 00:00:13,444
get interested in interconnection networks
as we go along, I highly recommend this

3
00:00:13,444 --> 00:00:18,370
book. I have not assigned anything from
this book for this class. But this is Bill

4
00:00:18,370 --> 00:00:23,003
Dalley and Brian Towles' interconnections
book. It's quite good. It's kind of the

5
00:00:23,003 --> 00:00:29,550
definitive guide on the subject matter.
Okay. So, interconnection networks. We, we

6
00:00:29,550 --> 00:00:37,200
talked about buses and we talked about
memory protocol about buses. That's only

7
00:00:37,640 --> 00:00:45,875
one way to share information. Now, people
may argue that's a intuitive way to share

8
00:00:45,875 --> 00:00:51,481
information cuz we had the ability to do
load and storage from our processor. But

9
00:00:51,481 --> 00:00:56,824
there are other ways to share information.
So, in today's lecture, we are going to

10
00:00:57,276 --> 00:01:03,222
talk about two main pieces of two main
topics. One, how do you share information

11
00:01:02,319 --> 00:01:08,039
in, in a different way which commingles
the movement of data with a

12
00:01:08,039 --> 00:01:16,477
synchronization primitive. We are going to
call that messaging, which is in contrast

13
00:01:16,477 --> 00:01:22,856
to memory or communicating via memory
addresses. We're also going to talk about

14
00:01:22,856 --> 00:01:27,940
different ways to connect together
processors, which can either have better

15
00:01:27,940 --> 00:01:34,516
performance or better scalability, i.e.,
have more nodes in the system. And, okay,

16
00:01:34,516 --> 00:01:43,378
so let's, let's compare and contrast buses
to other forms of network and see why we

17
00:01:43,378 --> 00:01:48,339
might want to change. So, this, this is
going back to what we had just talked

18
00:01:48,339 --> 00:01:53,697
about. Let's say, you have two cores on a
bus. And let's forget about how you want

19
00:01:53,300 --> 00:01:58,857
to communicate. It could either be via
shared memory or it could be via messaging

20
00:01:58,857 --> 00:02:04,149
or it could be via some other protocol,
Ethernet, whatever, whatever you want put

21
00:02:04,149 --> 00:02:09,051
here, which is basically a form of
messaging. We have one core and it wants

22
00:02:09,051 --> 00:02:14,212
to communicate with another core. Note, I
don't draw any caches here cuz there may

23
00:02:14,212 --> 00:02:19,067
not be caches, there may be caches. It's
doesn't, it's kind of immaterial here. If

24
00:02:19,067 --> 00:02:23,676
one person wants to talk to another
person, they can just yell at the other

25
00:02:23,676 --> 00:02:28,469
person. It's two people so there's, it's
pretty easy to do. We, now, there is some

26
00:02:28,469 --> 00:02:32,709
challenges, we can 't both talk at the
same time. We might not be able to

27
00:02:32,709 --> 00:02:37,502
understand each other. So, there's some
arbitration that needs to happen. But, in

28
00:02:37,502 --> 00:02:41,926
general, that arbitration is pretty
simple. Only two, two cores or entities on

29
00:02:41,926 --> 00:02:48,479
this bus. Okay. Now we, now we go to more
cores. So, we have four people in a room

30
00:02:48,479 --> 00:02:54,262
trying to shout to each other. Well, or
four people on a bus trying to shout to

31
00:02:54,262 --> 00:02:59,140
each other. And as, as we just talked
about, the bandwidth can be a challenge

32
00:02:59,140 --> 00:03:04,603
here. The arbitration for the bus can be a
challenge. And because we're talking about

33
00:03:04,603 --> 00:03:10,418
interconnection networks, the wire delay
and capacitance of the network can be

34
00:03:10,418 --> 00:03:16,740
worse or it can be a challenge here. So,
if we got one core, it needs to drive the

35
00:03:16,740 --> 00:03:23,140
shared multidrop bus. There's a lot of
capacitance on this bus, much more so than

36
00:03:23,140 --> 00:03:29,067
this case cuz all of a sudden, we've,
we've, doubled the length of the bus, so

37
00:03:29,067 --> 00:03:34,836
the wires are longer and we've also put
more loads on the bus. So, there's

38
00:03:34,836 --> 00:03:42,764
actually more capacitance on this bus.
Okay. Now, we start to think about trying

39
00:03:42,764 --> 00:03:48,599
to build a bus that has a lot more cores.
In this case, twelve. And through this

40
00:03:48,599 --> 00:03:53,802
core, if you go shot to that core, there's
no pipelines around on this bus or

41
00:03:53,802 --> 00:03:58,807
anything. You go to shout and has to
propagate all the way down here and, you

42
00:03:58,807 --> 00:04:03,483
know, we're, we're talking about high
rates of communication. You actually have

43
00:04:03,483 --> 00:04:09,063
to wait for the time of flight of light
from here to get down to there. And

44
00:04:09,063 --> 00:04:13,356
because we're, if we're using something,
let's say, like a snoopy protocol or a

45
00:04:13,356 --> 00:04:20,174
broadcast protocol, because that's all we
have here, we have to wait for and node

46
00:04:20,174 --> 00:04:23,863
here to communicate with every other node.
So, we have to wait for the worst case

47
00:04:23,863 --> 00:04:29,668
time for this node to communicate to that
node, every clock cycle. Hm, okay. and as

48
00:04:29,668 --> 00:04:35,022
I said there is capacitance, so it's not
quite a, just a transmission line, so it's

49
00:04:35,022 --> 00:04:40,823
not just a transmission line problem here.
We also have to worry about the

50
00:04:40,823 --> 00:04:45,969
capacitance in trying to drive all of
these different receivers. And it's a

51
00:04:45,969 --> 00:04:51,870
multidirectional bus so we have to have
effectively tri-states and t he ability to

52
00:04:51,870 --> 00:04:57,073
drive or just receive. Well, all of a
sudden, we have twelve people and

53
00:04:57,073 --> 00:05:01,875
actually, we have twelve people in this
room. So, let's all try to pick a number

54
00:05:01,875 --> 00:05:06,616
between one and ten and shout it real fast
on the count of three. One, two, three,

55
00:05:06,616 --> 00:05:11,049
five. Okay. I could, I, I do, I shouted
five, I don't know what everyone else

56
00:05:11,049 --> 00:05:15,174
said. So, that's does anyone, could
everyone hear everyone else's? Does

57
00:05:15,174 --> 00:05:20,099
everyone know exactly what all other ten
people said at the same time, or twelve

58
00:05:20,099 --> 00:05:24,902
people said at the same time? You heard
your nearest neighbor. Okay, but did you

59
00:05:24,902 --> 00:05:33,183
know, do you know what Yankey said? Yeah.
Okay. So, this is, this is the challenge.

60
00:05:33,183 --> 00:05:37,840
And if we need to guarantee that only one
person can yell on the bus at a time, we

61
00:05:37,840 --> 00:05:42,553
need some arbitration. But the arbitration
logic is slower now because we have lots

62
00:05:42,553 --> 00:05:47,491
of people communicating so we have to run
a wire from this node down to this node

63
00:05:47,491 --> 00:05:52,202
and then, we had to come back in the
arbitraration, logical, say, over here

64
00:05:52,202 --> 00:05:57,502
that needs to make some decision. And the
decision is slower because as more layers

65
00:05:57,502 --> 00:06:02,475
of logic, more combination of logic, we
will say, to make arbitration decision.

66
00:06:02,475 --> 00:06:06,598
Hm, okay. Now, if we go to a thousand
processors or a, a thousand cores on a

67
00:06:06,598 --> 00:06:11,170
bus, you know, we, we could even have
twelve people in the room shout at the

68
00:06:11,170 --> 00:06:14,948
same time. You can have a thousand people
in the room shout at the same time, and

69
00:06:14,948 --> 00:06:18,676
physically be distanced to the wiring
between this thousand different nodes is

70
00:06:18,676 --> 00:06:22,503
going to decrease the speed of the bus
significantly. So, it's some, something to

71
00:06:22,503 --> 00:06:30,671
think about. So, this, this motivates us
to take the same twelve course and think

72
00:06:30,671 --> 00:06:37,312
about some other ways to connect them.
Now, what I'm going to show here is a,

73
00:06:37,312 --> 00:06:46,219
what's known as a switched interconnect or
sometimes known as a point-to-point link

74
00:06:46,219 --> 00:06:50,878
solution. Now, point-to-point does not
mean that this core can communicate

75
00:06:50,878 --> 00:06:55,792
directly with every other core. That has,
that has a different name, we'll talk

76
00:06:55,792 --> 00:07:01,427
about that later today. Instead,
point-to-point just means each link, only

77
00:07:01,427 --> 00:07:04,935
has one sender and one receiver. And the
n.

78
00:07:04,935 --> 00:07:12,649
You use switches along the way to make
decisions and to route. So, if we look at

79
00:07:12,649 --> 00:07:17,072
this, we can actually have multiple
nearest neighbor communication happening.

80
00:07:17,072 --> 00:07:21,379
So, all of a sudden, by adding this
switching, we can both have connectivity

81
00:07:21,379 --> 00:07:25,977
between all the different nodes, but we
can also have sort of subconversations

82
00:07:25,977 --> 00:07:31,962
happening. But this still allows for this
processor here to go communicate with the

83
00:07:31,962 --> 00:07:37,380
one that's at the farthest extent. And we
need to decide how to do that. Whether it

84
00:07:37,380 --> 00:07:42,865
communicates sort of this way or this way
or that way or some other squiggly line.

85
00:07:43,400 --> 00:07:48,751
We can also take the same point-to-point
switch interconnect network. And like a

86
00:07:48,751 --> 00:07:54,035
bus, which we can increase the width of
the bus, which does not help us with the

87
00:07:54,035 --> 00:07:59,187
occupancy on the bus, we can add more
networks or we can affectively add

88
00:07:59,187 --> 00:08:03,248
multiple concurrent, switching
interconnection networks or we can

89
00:08:03,248 --> 00:08:09,307
increase the bandwidth on these buses. So,
it's similar sorts of ideas there and

90
00:08:09,307 --> 00:08:14,915
similar sorts of bandwidth tricks you can
do to increase bandwidth on buses. You can

91
00:08:14,915 --> 00:08:19,685
play on there, switch interconnection
always. Okay. So, this is just a very

92
00:08:19,685 --> 00:08:24,520
broad overview. And now, we're getting
into some, some more specific ideas.