1
00:00:00,080 --> 00:00:05,200
on may 29 2020 haddis law filed a class

2
00:00:03,200 --> 00:00:10,559
action lawsuit against western digital accusing the company of secretly selling

3
00:00:07,600 --> 00:00:15,679
hard drives built with smr technology and no the problem wasn't that

4
00:00:13,599 --> 00:00:19,840
these hard drives were whispering neck tingling sweet nothings into their users

5
00:00:18,000 --> 00:00:24,960
ears it's that smr not only makes the drive

6
00:00:22,240 --> 00:00:29,679
slower but in some cases completely unusable for their marketed purpose

7
00:00:27,519 --> 00:00:32,559
here's the thing even though judge judy is

8
00:00:30,560 --> 00:00:37,280
everything i aspire to be normally we don't delve into companies getting sued

9
00:00:35,360 --> 00:00:42,000
on this channel but for this lawsuit i'm making

10
00:00:39,600 --> 00:00:46,800
exception for two reasons one because patrick from serve the home asked me to

11
00:00:43,680 --> 00:00:48,960
and he's just a super nice guy and two

12
00:00:46,800 --> 00:00:53,840
because the reasons that this whole thing is happening and what it means for

13
00:00:51,120 --> 00:00:58,640
the industry and for you are super interesting and super techy

14
00:00:56,879 --> 00:01:02,800
and today's video was brought to you by glasswire keep track of the weird stuff

15
00:01:00,719 --> 00:01:06,479
that's connecting to your pc even when you aren't using it with glasswire you

16
00:01:05,040 --> 00:01:13,840
can see if a strange device joins your wi-fi and block it instantly get 25 off

17
00:01:09,200 --> 00:01:13,840
today using codelinus at the link below

18
00:01:20,479 --> 00:01:26,479
smr stands for shingled magnetic recording and it's a relatively new

19
00:01:24,479 --> 00:01:32,000
technology in the world of hard drives having just entered the market in 2014.

20
00:01:29,600 --> 00:01:37,840
now as you may know inside a hard drive are magnetic actuator arms that read and

21
00:01:34,720 --> 00:01:40,640
write data to a set of spinning platters

22
00:01:37,840 --> 00:01:46,799
now in perpendicular magnetic recording or pmr which is also known as

23
00:01:43,280 --> 00:01:49,200
conventional magnetic recording or cmr

24
00:01:46,799 --> 00:01:54,880
those platters have single tracks laid out side by side this makes each track

25
00:01:52,560 --> 00:01:59,920
easy to access by the read and right arms but capacity is limited to how many

26
00:01:57,920 --> 00:02:05,439
tracks you can squeeze next to each other on a single platter in shingled

27
00:02:02,479 --> 00:02:12,400
magnetic recording or smr platters are made up of chunks of overlapping tracks

28
00:02:09,360 --> 00:02:14,959
kind of like the shingles on a roof this

29
00:02:12,400 --> 00:02:21,360
technique allows more tracks to fit onto a single platter increasing capacity

30
00:02:17,760 --> 00:02:24,480
but it comes at a cost you see each time

31
00:02:21,360 --> 00:02:27,120
a new track is written it overlaps part

32
00:02:24,480 --> 00:02:31,760
of a previously written track which must then also be rewritten

33
00:02:29,440 --> 00:02:36,319
i mean it sounds kind of insane doesn't it i mean imagine a word processor where

34
00:02:34,560 --> 00:02:40,720
every time you wanted to change a word in a paragraph you had to go back and

35
00:02:38,720 --> 00:02:44,879
rewrite the rest of the paragraph along with your changes terrible

36
00:02:42,879 --> 00:02:49,760
i i mean that experience not ltteststore.com ltdstore.com is great

37
00:02:47,760 --> 00:02:55,519
great products but anyway for hard drives this approach is actually very

38
00:02:52,560 --> 00:03:00,480
sane you see ssds kind of have the whole speed thing on lock these days so most

39
00:02:58,720 --> 00:03:05,120
hard drive innovation over the last few years has been pretty focused on

40
00:03:02,480 --> 00:03:11,760
capacity and shingle magnetic recording allows drive manufacturers to exchange a

41
00:03:08,000 --> 00:03:14,560
bit of performance for as much as 25 to

42
00:03:11,760 --> 00:03:20,239
50 percent higher capacity per platter that means either greater capacities or

43
00:03:17,920 --> 00:03:23,519
fewer platters for a given capacity lowering the total cost

44
00:03:22,480 --> 00:03:28,000
so then that's problem number one really hard

45
00:03:25,840 --> 00:03:31,840
drive manufacturers selling a cheaper product without labeling it as such

46
00:03:30,640 --> 00:03:37,200
but why are people so upset

47
00:03:34,640 --> 00:03:42,000
well you see it's not just that smr drives have slower performance it's that

48
00:03:39,680 --> 00:03:47,680
to work properly they need special firmware or software that can navigate

49
00:03:44,720 --> 00:03:53,599
through all of these overlapping tracks so there are three different varieties

50
00:03:50,080 --> 00:03:54,720
of smr in host managed smr the host

51
00:03:53,599 --> 00:03:59,599
computer must give special commands to the drive

52
00:03:58,080 --> 00:04:04,480
if the drive doesn't receive these commands it simply will not work these

53
00:04:02,400 --> 00:04:09,920
kinds of drives are most commonly used in cloud servers owned by you know

54
00:04:06,720 --> 00:04:12,080
amazon or google where vast quantities

55
00:04:09,920 --> 00:04:18,079
of drives must be controlled the second type host aware smr allows the drive to

56
00:04:15,599 --> 00:04:23,520
respond to these special commands or to regular commands so you don't

57
00:04:20,639 --> 00:04:29,680
necessarily need special software to use that second type but it certainly helps

58
00:04:27,199 --> 00:04:34,400
the third type of smr is where things get messy

59
00:04:31,199 --> 00:04:36,720
in device managed smr the drive doesn't

60
00:04:34,400 --> 00:04:41,440
require special commands and in fact it shows up in your pc as a completely

61
00:04:39,520 --> 00:04:46,320
normal drive most of the time this isn't catastrophic

62
00:04:43,680 --> 00:04:52,240
in fact this type of smr drive usually has a cmr cache which can be written

63
00:04:49,520 --> 00:04:56,240
more quickly for a short period masking that slower write performance this

64
00:04:54,240 --> 00:05:02,800
approach is actually very similar to how many slower qlc ssds use a high speed

65
00:04:59,759 --> 00:05:05,120
slc cache as we discussed in this video

66
00:05:02,800 --> 00:05:08,880
right here so the idea is that when you write some data to the drive it goes

67
00:05:07,039 --> 00:05:12,960
into the higher speed cache and your file transfer is complete your data is

68
00:05:10,880 --> 00:05:17,520
safe but then 20 minutes 30 minutes later you might

69
00:05:15,199 --> 00:05:22,639
actually still hear the drive thrashing away as it dumps the data out of the

70
00:05:19,600 --> 00:05:24,560
cache onto the slower shingled portion

71
00:05:22,639 --> 00:05:28,880
for most home users honestly that's probably fine and they

72
00:05:27,039 --> 00:05:32,080
might even be happy with the trade-off for more capacity on the cheap

73
00:05:31,280 --> 00:05:37,759
but using a device-managed smr drive when

74
00:05:34,800 --> 00:05:43,840
you don't know that it's an smr drive can be catastrophic as our good buddy

75
00:05:40,400 --> 00:05:48,080
patrick found out he and his team tested

76
00:05:43,840 --> 00:05:50,880
cmr drives from hgst and seagate against

77
00:05:48,080 --> 00:05:55,440
two drives from wd's red lineup which is marketed specifically for network

78
00:05:53,199 --> 00:06:00,960
attached storage systems only one small problem without clear markings one of

79
00:05:58,320 --> 00:06:06,240
them was built with cmr and the other was built with smr as expected they

80
00:06:04,000 --> 00:06:12,160
found that the smr drive was slower in almost every typical storage benchmark

81
00:06:08,880 --> 00:06:15,360
which was bad then it got worse the

82
00:06:12,160 --> 00:06:17,520
kicker was the freenas raid z rebuild

83
00:06:15,360 --> 00:06:22,319
now raid is designed to protect your data through redundant copies in the

84
00:06:19,440 --> 00:06:27,120
event of a physical drive failure so the test involves wiping one of the drives

85
00:06:24,720 --> 00:06:31,280
in a raid array intentionally and then tasking the system with rebuilding all

86
00:06:29,360 --> 00:06:37,039
the data that was supposed to be on it the cmr drives all took over 16 hours

87
00:06:34,560 --> 00:06:40,560
which is already a long time to be worried that your data is in this

88
00:06:38,639 --> 00:06:44,639
vulnerable state where if another drive dies it could all be lost

89
00:06:42,479 --> 00:06:49,440
but it is normal for this process to take a while as for the smr drive it

90
00:06:47,360 --> 00:06:52,400
took nine and a half days

91
00:06:51,440 --> 00:06:57,520
what now patrick and co naturally thought

92
00:06:55,520 --> 00:07:03,120
something must be wrong with their test setup so they ran the test again

93
00:07:00,560 --> 00:07:09,440
and got a similar result ours technica also ran some tests finding that in a

94
00:07:05,440 --> 00:07:12,400
random write latency test the wd red smr

95
00:07:09,440 --> 00:07:19,280
drive could take as long as 1.3 seconds to save a meg of data compared to 0.1

96
00:07:15,599 --> 00:07:21,199
seconds on the seagate cmr drive now we

97
00:07:19,280 --> 00:07:25,919
need to mention that servtohome's test server ran on freenas a popular open

98
00:07:24,080 --> 00:07:30,000
source operating system for network attached storage and while smr drives

99
00:07:28,240 --> 00:07:33,599
have clearly been operating without disastrous consequences on other types

100
00:07:32,400 --> 00:07:38,319
of servers freenas uses the zfs file system which

101
00:07:36,639 --> 00:07:44,880
can also be found in a ton of other software from ubuntu and proxmox to nas

102
00:07:41,440 --> 00:07:47,520
systems from qnap and apparently zfs

103
00:07:44,880 --> 00:07:51,599
according to patrick's tests does not play nicely with shingle magnetic

104
00:07:49,680 --> 00:07:56,240
recording now there's some indication that smr might cause problems for other

105
00:07:53,919 --> 00:08:01,759
file systems too since synology has listed wd's smr drives as incompatible

106
00:07:59,520 --> 00:08:05,120
with a huge swath of their nas systems but we don't have the specifics on those

107
00:08:03,759 --> 00:08:10,720
ones anyway all of this would have been fine

108
00:08:08,240 --> 00:08:16,000
if customers were educated properly so they didn't accidentally end up with smr

109
00:08:13,039 --> 00:08:21,759
drives except as we said before wd didn't label their smr drives the only

110
00:08:19,440 --> 00:08:25,840
indication that a given drive was smr or cmr was a single letter in the model

111
00:08:24,160 --> 00:08:29,759
number and because these drives are device managed the end user wouldn't

112
00:08:27,840 --> 00:08:33,760
even realize what they had purchased when they plugged it into their system

113
00:08:31,440 --> 00:08:38,320
now to be fair to wd seagate and toshiba also sold unlabeled

114
00:08:36,880 --> 00:08:43,279
smr drives but just not in their nas product lines

115
00:08:40,959 --> 00:08:47,200
so it was far more unlikely that those drives would lead to the type of

116
00:08:44,959 --> 00:08:50,080
unusable performance seen in patrick's tests

117
00:08:48,240 --> 00:08:53,920
and the good news is that all three vendors have now committed to labeling

118
00:08:52,080 --> 00:08:58,160
the recording technology used in their products and to wd's credit they seem to

119
00:08:56,480 --> 00:09:02,560
be responding well to customers requesting cmr replacements for their

120
00:09:00,240 --> 00:09:07,279
smr drives meaning that the only lingering question after all this is

121
00:09:05,440 --> 00:09:13,120
why how exactly did western digital a titan

122
00:09:10,480 --> 00:09:18,640
of the storage industry for 50 years miss the fact that smr drives are

123
00:09:15,519 --> 00:09:21,440
completely inappropriate for zfs a

124
00:09:18,640 --> 00:09:26,320
perfectly common file system for nasa's and then market these drives for nas use

125
00:09:24,800 --> 00:09:33,120
a question that is made all the more confusing by this video from 2015

126
00:09:29,680 --> 00:09:36,560
showing an engineer from hgst a company

127
00:09:33,120 --> 00:09:39,839
owned by western digital explaining how

128
00:09:36,560 --> 00:09:42,880
dmsmr has tremendous compatibility

129
00:09:39,839 --> 00:09:44,399
problems with cfs and talking about how

130
00:09:42,880 --> 00:09:49,120
much work it would take to make them play nicely together wd clearly should

131
00:09:46,720 --> 00:09:53,760
have listened to manfred instead of evilly throwing their customers under

132
00:09:51,360 --> 00:09:58,320
the bus unless maybe something else happened here i don't think that this is

133
00:09:56,080 --> 00:10:02,560
some kind of evil masterminded plot to hurt nast users by western digital by

134
00:10:00,480 --> 00:10:05,839
any means i think that what's more likely is that this is just a

135
00:10:03,760 --> 00:10:09,279
communication breakdown and process breakdown between the different

136
00:10:07,040 --> 00:10:12,560
functional silos within western digital there's probably an element of this

137
00:10:10,560 --> 00:10:16,079
which is that each person has their own personal and departmental goals aligned

138
00:10:14,560 --> 00:10:20,720
to their function but not necessarily across their entire company and what

139
00:10:18,160 --> 00:10:26,640
that means is that those discussions don't happen to say hey should we even

140
00:10:23,760 --> 00:10:30,720
institute dmsmr into the space if we do what are the tests that we need

141
00:10:28,640 --> 00:10:34,240
to run to make sure that it works on all of our partner systems personally i like

142
00:10:32,720 --> 00:10:38,160
to think that best of people maybe to my detriment but haven't done management

143
00:10:35,839 --> 00:10:42,880
consulting in this space for years prior to doing sth i can totally see how this

144
00:10:40,480 --> 00:10:46,720
happens yeah okay i actually made a similar argument

145
00:10:44,959 --> 00:10:51,200
during the principled technologies benchmark scandal that Intel went

146
00:10:48,480 --> 00:10:55,120
through a number of months ago and i think patrick is probably right here as

147
00:10:53,519 --> 00:10:59,839
well it's just still extremely embarrassing for them

148
00:10:57,279 --> 00:11:05,440
and the class action lawsuit is absolutely in our opinion both

149
00:11:02,320 --> 00:11:07,680
good and necessary because normal people

150
00:11:05,440 --> 00:11:12,320
who bought wd red smr drives may have faced a loss of time data and even

151
00:11:10,240 --> 00:11:19,279
livelihood that they wouldn't have if the product was properly labeled and

152
00:11:15,360 --> 00:11:21,120
honestly they deserve some compensation

153
00:11:19,279 --> 00:11:24,720
speaking of compensation massive shout out to drop.com for being

154
00:11:23,440 --> 00:11:31,200
our sponsor thanks guys the mass drop in sennheiser pc37x gaming

155
00:11:28,880 --> 00:11:36,560
headset features angled drivers and an open back design with drivers that come

156
00:11:33,279 --> 00:11:39,200
from the same family as the hd598 and hd

157
00:11:36,560 --> 00:11:43,519
600 series headphones they offer great stereo imaging and locational accuracy

158
00:11:41,600 --> 00:11:46,320
they sound fantastic they're super comfortable they come with a noise

159
00:11:44,959 --> 00:11:49,920
cancelling microphone and they've sold over 45 000 of these things so if there

160
00:11:48,399 --> 00:11:54,320
was a big problem with them like if they were completely unusable for the

161
00:11:51,600 --> 00:11:57,440
advertised purpose you probably would have heard about it

162
00:11:55,519 --> 00:12:00,560
by now they include a two-year warranty from the manufacturer and you can check

163
00:11:59,200 --> 00:12:04,000
them out at the link in the video description if you guys enjoyed this

164
00:12:02,240 --> 00:12:08,399
video and you like these kinds of deeper dives into storage technology maybe

165
00:12:06,480 --> 00:12:11,440
check out uh oh what was that cool when we did about like tiered storage a while

166
00:12:10,160 --> 00:12:15,279
back can't remember hopefully you can find it

167
00:12:13,600 --> 00:12:19,120
oh oh maybe just server storage why don't we do uh rolling out nuwanik all

168
00:12:17,360 --> 00:12:23,600
the issues with that boy what a disaster that was huh storage it's complicated