1
00:00:00,440 --> 00:00:06,200
behind me right now is the biggest

2
00:00:03,640 --> 00:00:11,160
supercomputer in the country it'll be serving researchers across Canada

3
00:00:09,240 --> 00:00:16,440
studying the human genome in bioinformatics particle physics

4
00:00:13,320 --> 00:00:19,039
materials research even humani research

5
00:00:16,440 --> 00:00:23,480
it's called Cedar it cost the federal government through the Canadian

6
00:00:20,439 --> 00:00:26,560
foundation for Innovation over $16

7
00:00:23,480 --> 00:00:28,960
million and we get to be the first to

8
00:00:26,560 --> 00:00:31,960
unbox this Beast

9
00:00:40,320 --> 00:00:49,399
Savage jerky is created without the use of nitrates or preservatives use offer

10
00:00:46,000 --> 00:00:50,600
code LTT to save 10% at the link in the

11
00:00:49,399 --> 00:00:59,199
video description so Cedar is a big data

12
00:00:54,760 --> 00:01:01,640
machine it takes up a quar of the 5,000

13
00:00:59,199 --> 00:01:07,360
ft Data Center it occupies meaning actually that there's room for it to

14
00:01:03,480 --> 00:01:11,080
grow but right now it has

15
00:01:07,360 --> 00:01:18,080
27,000 Intel Zeon processing course

16
00:01:11,080 --> 00:01:24,000
190 terab of RAM 64 petabytes of storage

17
00:01:18,080 --> 00:01:26,159
584 gpus and a total power draw of

18
00:01:24,000 --> 00:01:32,399
560,000 wats though with that said it's

19
00:01:29,360 --> 00:01:37,200
efficiency is a shocking

20
00:01:32,399 --> 00:01:39,280
1.07 on the Pue scale where one would be

21
00:01:37,200 --> 00:01:44,520
perfect and a typical data center would be 1 and 1/2 to two we'll get into how

22
00:01:42,280 --> 00:01:49,439
they did that a little bit later though so our tour starts right here behind me

23
00:01:47,079 --> 00:01:55,920
are what they call the high availability racks so everything back there has dual

24
00:01:53,520 --> 00:02:02,079
power supplies for redundancy with a battery backup for that and a diesel

25
00:01:58,600 --> 00:02:04,680
generator back up that everything back

26
00:02:02,079 --> 00:02:11,480
here is Mission critical things like networking login servers and management

27
00:02:07,880 --> 00:02:14,680
servers are all here and this is also

28
00:02:11,480 --> 00:02:17,160
where you'll find the bulk of Cedar's

29
00:02:14,680 --> 00:02:22,959
storage let's get in for a closer look at Cedar's connection to the outside

30
00:02:19,440 --> 00:02:27,680
world this networking Appliance from

31
00:02:22,959 --> 00:02:32,599
Huawei has a street price of around a

32
00:02:27,680 --> 00:02:35,480
million doll W and right here this is

33
00:02:32,599 --> 00:02:41,800
where it gets really bananas these guys are seedar

34
00:02:37,400 --> 00:02:44,599
dual 100 gabit connections through

35
00:02:41,800 --> 00:02:52,080
Vancouver and then as if that wasn't enough these orange ones here are dual

36
00:02:48,680 --> 00:02:54,680
40 gbit connections through nearby sui

37
00:02:52,080 --> 00:02:58,400
just in case somebody puts a back ho through one of these other fiber lines

38
00:02:56,920 --> 00:03:00,879
and they would have otherwise lost their internet connectivity I mean that's

39
00:03:00,080 --> 00:03:08,000
their Backup backup but Ethernet is not really the

40
00:03:05,920 --> 00:03:17,799
way you want to connect high performance Computing nodes this this right here is

41
00:03:12,080 --> 00:03:21,560
the true networking heart of Cedar these

42
00:03:17,799 --> 00:03:23,319
are 48 Port omn path switches and

43
00:03:21,560 --> 00:03:32,519
they're configured in what's called an island topology so the island is in

44
00:03:26,959 --> 00:03:34,760
almost all cases 32 compute nodes each

45
00:03:32,519 --> 00:03:42,640
of those compute nodes is connected to 32 ports on one of these switches in its

46
00:03:38,680 --> 00:03:46,720
rack then the remaining 16 ports come

47
00:03:42,640 --> 00:03:51,239
back to here that means that every

48
00:03:46,720 --> 00:03:54,519
island gets a dedicated line to each of

49
00:03:51,239 --> 00:03:57,159
the core switches giving you failover

50
00:03:54,519 --> 00:04:04,280
and massive bandwidth each one of these fiber links

51
00:04:00,640 --> 00:04:07,640
right here is capable of 100 gbit per

52
00:04:04,280 --> 00:04:11,920
second so even though between islands we

53
00:04:07,640 --> 00:04:13,560
are let's say bottlenecked by our 16

54
00:04:11,920 --> 00:04:19,720
connections so that's only half the total theoretical speed within an island

55
00:04:15,959 --> 00:04:22,600
we're still talking over 100 gabyt per

56
00:04:19,720 --> 00:04:29,280
second so it's not really an issue okay now let's move on to SFU and

57
00:04:26,360 --> 00:04:35,160
compute Canada's version of petabyte project spoiler alert theirs is better

58
00:04:32,240 --> 00:04:43,080
in every conceivable way so in the five cabinets behind me we've got Cedar 50

59
00:04:39,199 --> 00:04:46,240
petabyte IBM tape Library System they

60
00:04:43,080 --> 00:04:50,560
have a 40 gabit link to the rest of the

61
00:04:46,240 --> 00:04:53,320
supercomputer and each of the 5,000 10

62
00:04:50,560 --> 00:05:00,039
tbte magnetic tapes inside can be grabbed out of storage moved with like a

63
00:04:57,039 --> 00:05:01,960
robotic ARM into a reader and the data

64
00:05:00,039 --> 00:05:05,680
can then be accessed when needed and this is done

65
00:05:03,639 --> 00:05:12,000
automatically cool right okay yeah but due to the slowness

66
00:05:08,840 --> 00:05:15,199
of that swapping process this is still

67
00:05:12,000 --> 00:05:16,919
what we would consider to be cold or

68
00:05:15,199 --> 00:05:24,039
archival storage next up here is general purpose

69
00:05:20,440 --> 00:05:27,080
storage land where any data that's being

70
00:05:24,039 --> 00:05:30,720
used for any current research project

71
00:05:27,080 --> 00:05:35,360
would be housed so here here they're

72
00:05:30,720 --> 00:05:37,080
using offthe shelf five U racks Each of

73
00:05:35,360 --> 00:05:43,039
which contains let's see if we can crack One open here a total of two kind of

74
00:05:40,400 --> 00:05:53,120
trays here and 84 8 terab uh let's have a look here

75
00:05:48,680 --> 00:05:55,639
Enterprise capacity SAS drives from

76
00:05:53,120 --> 00:06:02,120
Seagate but there's actually more to this system than meets the eye every

77
00:05:58,720 --> 00:06:05,639
four of of these storage nodes requires

78
00:06:02,120 --> 00:06:08,039
two nodes of what they're calling object

79
00:06:05,639 --> 00:06:15,280
storage servers these act as a high-speed cache with their SAS 10,000

80
00:06:11,080 --> 00:06:18,400
RPM drives as well as as kind of like a

81
00:06:15,280 --> 00:06:21,919
a a traffic cough for everything behind

82
00:06:18,400 --> 00:06:24,840
it so every single read or write to

83
00:06:21,919 --> 00:06:30,759
these hard drives actually goes through these nodes so right now General Storage

84
00:06:28,360 --> 00:06:38,720
Land is 10 ped byes but in the near to Mid future it will be expanding to 20

85
00:06:35,080 --> 00:06:41,479
20 now that DIY approach to storage is

86
00:06:38,720 --> 00:06:45,680
great for scaling up at a low cost but when it comes to Performance they went

87
00:06:43,440 --> 00:06:54,400
for this data direct network storage Appliance because it has got the real

88
00:06:50,000 --> 00:06:58,759
Goods now in the rack next to this brain

89
00:06:54,400 --> 00:07:01,479
you'll find a mere 4 pedabytes of actual

90
00:06:58,759 --> 00:07:06,000
storage due to its higher cost but thanks to its proprietary Hardware

91
00:07:03,400 --> 00:07:13,919
custom software and solid state burst buffers this thing can handle up to 40

92
00:07:10,319 --> 00:07:16,960
gabt per second of sustained throughput

93
00:07:13,919 --> 00:07:19,759
making it perfect for data intensive

94
00:07:16,960 --> 00:07:26,960
applications that rely on humongous data sets now let's get into

95
00:07:23,879 --> 00:07:30,199
compute there are about half a dozen

96
00:07:26,960 --> 00:07:32,800
different types of compute nodes all

97
00:07:30,199 --> 00:07:38,080
connected to the same high-speed Omni paath Network backbone that are

98
00:07:34,879 --> 00:07:41,720
optimized for different types of

99
00:07:38,080 --> 00:07:45,240
research we'll begin with the base

100
00:07:41,720 --> 00:07:49,479
compute note there are a whopping

101
00:07:45,240 --> 00:07:52,159
576 of these each of these is a computer

102
00:07:49,479 --> 00:07:59,840
so there's actually four in a single toou shell Each of which contains two

103
00:07:55,039 --> 00:08:02,800
Zeon E5 2683 16 core processors 128 gigs

104
00:07:59,840 --> 00:08:10,800
of RAM and about a terabyte of raid zero SSD storage for Scratch so each rack

105
00:08:06,879 --> 00:08:14,599
here contains two islands so that's a

106
00:08:10,800 --> 00:08:15,840
total of 64 compute nodes giving us a

107
00:08:14,599 --> 00:08:23,840
whopping 2,48 compute units per rack so these

108
00:08:20,400 --> 00:08:26,520
nodes are the basic Workhorse of Cedar

109
00:08:23,840 --> 00:08:32,000
handling everything from Monte Carlo simulations for Material Science to sim

110
00:08:29,639 --> 00:08:38,719
ating Dynamic processes in nature with a high degree of Randomness like snowfall

111
00:08:34,880 --> 00:08:42,000
or rainfall they would also be used in

112
00:08:38,719 --> 00:08:44,480
any highly parallelized workload because

113
00:08:42,000 --> 00:08:49,480
if you need you know 10,000 CPU cores for one job there

114
00:08:47,399 --> 00:08:55,800
aren't enough cores in any other class of server to handle that kind of

115
00:08:52,839 --> 00:09:01,880
load moving right on up we've got the big memory nodes there are 48 of these

116
00:08:59,079 --> 00:09:09,600
and half of them are just like the basic nodes except with 512 gigs of RAM while

117
00:09:06,240 --> 00:09:15,200
the other half of them these puppies

118
00:09:09,600 --> 00:09:18,399
have 1 and half terabytes of system

119
00:09:15,200 --> 00:09:21,399
memory these ones take up twice as much

120
00:09:18,399 --> 00:09:24,680
rack space though each of these one use

121
00:09:21,399 --> 00:09:27,079
is a single dual socket system because

122
00:09:24,680 --> 00:09:35,320
you know what there just wasn't enough go darn room for old 24 64 gig memory

123
00:09:32,120 --> 00:09:37,560
modules that are required for that much

124
00:09:35,320 --> 00:09:43,959
RAM first world problem

125
00:09:39,760 --> 00:09:47,480
yes these guys are really special these

126
00:09:43,959 --> 00:09:50,320
are the aptly named 3 tbte nodes there

127
00:09:47,480 --> 00:09:58,959
are only a handful of them but these are quad socket machines with Zeon

128
00:09:53,600 --> 00:10:01,640
4809 v4s four of them but wait a tick

129
00:09:58,959 --> 00:10:06,920
those are only eight core processors these don't even have more processing

130
00:10:04,560 --> 00:10:13,640
cores than those little tiny ones that take up half of you what's the deal here

131
00:10:10,600 --> 00:10:16,160
well it turns out that some

132
00:10:13,640 --> 00:10:22,200
bioinformatics workloads like genome sequencing don't actually scale very

133
00:10:19,000 --> 00:10:24,720
well with more processors they just need

134
00:10:22,200 --> 00:10:30,519
massive amounts of memory to hold the data sets that they need to work on so

135
00:10:28,360 --> 00:10:37,000
while the team here probably isn't super stoked on using up for use just so they

136
00:10:34,399 --> 00:10:41,720
can stuff more memory into the system until until optane reaches a higher

137
00:10:39,639 --> 00:10:47,440
level of maturity this is the only choice they have now finally we're

138
00:10:44,720 --> 00:10:56,800
getting to my favorite nodes the most expensive nodes these are the GPU nodes

139
00:10:53,720 --> 00:10:59,320
and while they're actually quite similar

140
00:10:56,800 --> 00:11:04,560
to the base nodes with respect to their CPU and RAM configurations what's got

141
00:11:02,959 --> 00:11:09,560
the researchers in the fields of molecular Dynamics Ai and machine

142
00:11:07,360 --> 00:11:16,519
learning all amped up about these are the quad NVIDIA Tesla p100 graphics

143
00:11:13,399 --> 00:11:19,120
cards that they have cram into each one

144
00:11:16,519 --> 00:11:24,240
I mean seriously with 1,500 watts of power being consumed by

145
00:11:22,480 --> 00:11:29,920
each one of these is it it is an engineering Marvel that they've crammed

146
00:11:27,040 --> 00:11:34,680
enough power and cool to make this whole thing work so

147
00:11:32,680 --> 00:11:40,680
actually now that you think about it how exactly did they do that so the Keen ey

148
00:11:38,360 --> 00:11:46,440
among you might have already caught a couple of hints earlier in this video

149
00:11:42,680 --> 00:11:49,959
but the secret lies in the rear doors on

150
00:11:46,440 --> 00:11:54,680
the server racks look how thick this is

151
00:11:49,959 --> 00:11:58,200
yes my friends this entire door is a

152
00:11:54,680 --> 00:12:00,800
gigantic heat exchanger so their servers

153
00:11:58,200 --> 00:12:04,880
don't actually have have water blocks that would be more expensive what

154
00:12:03,000 --> 00:12:08,920
they're doing is they've just got the fronts of the racks all sealed up so

155
00:12:06,839 --> 00:12:14,519
there's no Backdraft pressure and they've got normal air cooled servers

156
00:12:12,200 --> 00:12:19,199
that pass the air from the front where they just draw in room temperature air

157
00:12:16,360 --> 00:12:24,560
in here and it comes out hot like 30 plus de and push it through this heat

158
00:12:21,800 --> 00:12:29,959
exchanger where it is actually cool to my skin that's how efficient these are

159
00:12:28,279 --> 00:12:35,440
and that cooling system system is massively expandable too you can

160
00:12:32,240 --> 00:12:37,920
actually see above me I am standing

161
00:12:35,440 --> 00:12:46,120
where we've got a blue and green cooling pipe connected to a whole bunch of quick

162
00:12:42,040 --> 00:12:49,079
release fittings ready to add more racks

163
00:12:46,120 --> 00:12:52,360
right here but to see what they actually do with the heat we're actually going to

164
00:12:50,920 --> 00:13:01,880
have to go upstairs where we'll find the final and

165
00:12:57,279 --> 00:13:05,639
perhaps the uh coolest stop in our tour

166
00:13:01,880 --> 00:13:10,160
here this is the mechanical room where

167
00:13:05,639 --> 00:13:13,480
the pumps and these freaking pipes take

168
00:13:10,160 --> 00:13:17,120
all the water from downstairs and dump

169
00:13:13,480 --> 00:13:20,120
it into three cooling towers outside the

170
00:13:17,120 --> 00:13:21,880
building now right now the weather is

171
00:13:20,120 --> 00:13:28,880
favorable to cooling the ambient temperature is quite low so it's just

172
00:13:25,040 --> 00:13:31,480
operating as gigantic radiators but get

173
00:13:28,880 --> 00:13:37,440
this when the conditions become less favorable in the summer they kick things

174
00:13:34,399 --> 00:13:41,399
into high gear with an automated system

175
00:13:37,440 --> 00:13:44,320
that sprays water onto the fins of the

176
00:13:41,399 --> 00:13:48,760
radiators in the cooling towers and if you watched our bong cooling video which

177
00:13:47,079 --> 00:13:53,920
you can check out right here you'll be familiar with this concept already but

178
00:13:51,079 --> 00:13:59,639
this is called evaporative cooling and by these means even in ambient

179
00:13:55,959 --> 00:14:04,279
temperatures up to 30° C

180
00:13:59,639 --> 00:14:07,560
they can achieve the 17° coolant levels

181
00:14:04,279 --> 00:14:10,880
that they need to without employing the

182
00:14:07,560 --> 00:14:15,759
massive Chiller unit that they have over

183
00:14:10,880 --> 00:14:15,759
on the other side of the

184
00:14:17,199 --> 00:14:23,839
room Squarespace is the way to build a

185
00:14:21,519 --> 00:14:29,399
website whether it's for your small business or for your you know uh local

186
00:14:27,759 --> 00:14:35,000
freaking book club it does doesn't matter if you want a web presence

187
00:14:31,800 --> 00:14:36,880
affordably and quickly Squarespace gets

188
00:14:35,000 --> 00:14:42,120
it done for you it starts at just 12 bucks a month and you get a free domain

189
00:14:39,440 --> 00:14:47,680
if you buy Squarespace for the year you just pick one of their templates and

190
00:14:44,800 --> 00:14:53,920
boom you upload some pictures you fill in some text it's all cloud-based and

191
00:14:50,680 --> 00:14:57,440
your website will simple as that look

192
00:14:53,920 --> 00:14:59,959
great on any device every website comes

193
00:14:57,440 --> 00:15:05,279
with a free online store and their cover pages feature allows you to set up a

194
00:15:02,320 --> 00:15:10,759
beautiful one-page online presence in just minutes so start a trial with no

195
00:15:08,720 --> 00:15:16,560
credit card required and start building your website today then when you decide

196
00:15:14,120 --> 00:15:23,000
to sign up for Squarespace don't forget head over to squarespace.com

197
00:15:19,160 --> 00:15:24,399
LTT and use offer code LTT to get 10%

198
00:15:23,000 --> 00:15:30,759
off your first purchase so a massive thank you to SFU

199
00:15:27,839 --> 00:15:35,079
and compute Canada for allowing us to uh run a muck in their data center thanks

200
00:15:33,120 --> 00:15:38,160
to you guys for watching if you dislike this video you know what to do but if

201
00:15:36,399 --> 00:15:41,839
you liked it hit that like button get subscribed maybe consider checking out

202
00:15:39,639 --> 00:15:45,399
where to buy the stuff we featured at the link in the video description also

203
00:15:44,120 --> 00:15:50,759
down there you'll find a link to our merch store which has cool shirts like this one and our community Forum which

204
00:15:48,720 --> 00:15:53,759
you should totally join
