1
00:00:00,000 --> 00:00:05,240
It's three in the morning and you can't sleep. So the natural response is of course to bust out your phone

2
00:00:05,240 --> 00:00:08,720
and waste some time on Reddit, get to swiping on Tinder

3
00:00:08,720 --> 00:00:14,080
or blow a few more dollars on that freemium game that's quickly becoming a real problem.

4
00:00:14,080 --> 00:00:17,280
And luckily for you, all of these services work

5
00:00:17,280 --> 00:00:24,120
without a hitch because they're all being provided to you from huge data centers that are expected to be reliable

6
00:00:24,120 --> 00:00:28,600
no matter how many people are connected or what time of day it is.

7
00:00:28,680 --> 00:00:33,440
But how do they stay running with almost no downtime?

8
00:00:33,440 --> 00:00:37,960
Well, our friends at IBM sponsored this video to tell you guys all about it.

9
00:00:37,960 --> 00:00:43,360
And it starts of course with having facilities and computers that are purpose built

10
00:00:43,360 --> 00:00:46,720
to handle lots of incoming connections.

11
00:00:46,720 --> 00:00:50,160
Now, although some services have their own data centers,

12
00:00:50,160 --> 00:00:55,080
many others who either don't need or can't afford a dedicated facility

13
00:00:55,080 --> 00:01:00,520
or need additional capacity actually rent server space and processing time

14
00:01:00,520 --> 00:01:04,920
from a larger companies, a strategy called co-location.

15
00:01:04,920 --> 00:01:08,240
Data centers are typically made up of lots of servers

16
00:01:08,240 --> 00:01:11,280
sitting in racks to maximize how many machines

17
00:01:11,280 --> 00:01:16,000
can fit into one bit of floor space. But of course, setting up a data center

18
00:01:16,000 --> 00:01:19,080
is a lot more complicated than just throwing a bunch

19
00:01:19,080 --> 00:01:22,080
of servers into a warehouse and calling it a day.

20
00:01:22,080 --> 00:01:27,680
Larger data centers hold so much equipment that they are actually built to be sturdier

21
00:01:27,680 --> 00:01:31,360
than your average building to accommodate all the extra weight

22
00:01:31,360 --> 00:01:36,920
from these large racks of servers. Some of them are so large that the workers inside

23
00:01:36,920 --> 00:01:40,520
are even given small vehicles like scooters or bicycles

24
00:01:40,520 --> 00:01:43,760
so they can get around and troubleshoot issues more quickly.

25
00:01:43,760 --> 00:01:47,600
Also, those servers generate a ton of heat

26
00:01:47,600 --> 00:01:52,760
so elaborate cooling systems including water chilling are often employed.

27
00:01:52,760 --> 00:01:56,560
Additionally, data centers are often laid out to be more efficient.

28
00:01:56,560 --> 00:01:59,800
For example, servers will usually either face

29
00:01:59,800 --> 00:02:03,520
toward each other or directly away from each other

30
00:02:03,520 --> 00:02:07,480
to create what are called corridors of hot air

31
00:02:07,480 --> 00:02:10,520
that can be pushed out more easily.

32
00:02:10,520 --> 00:02:15,760
But when there's an environmental hazard that isn't considered, huge problems can result.

33
00:02:15,760 --> 00:02:21,400
For example, a few years ago, Facebook actually had weather in the form of clouds

34
00:02:21,440 --> 00:02:26,280
inside one of its data centers which caused some of the equipment to short out.

35
00:02:26,280 --> 00:02:29,560
So humidity control can also be very important

36
00:02:29,560 --> 00:02:33,720
for larger operations. Some servers are even designed to withstand

37
00:02:33,720 --> 00:02:39,160
even more serious hazards like earthquakes using braces and extra floor mounting.

38
00:02:39,160 --> 00:02:44,000
So of course, data centers are protected with advanced physical and human security.

39
00:02:44,000 --> 00:02:49,160
But what about the more technical challenges? Well, aside from encrypting it,

40
00:02:49,160 --> 00:02:52,920
data is often kept safe by spreading the processing

41
00:02:52,920 --> 00:02:56,360
and storage across multiple locations

42
00:02:56,360 --> 00:03:00,320
rather than having them just on one single machine in one place.

43
00:03:00,320 --> 00:03:04,000
To make this simpler, servers are very often virtualized

44
00:03:04,000 --> 00:03:08,880
meaning that one physical server can be seen as several different systems.

45
00:03:08,880 --> 00:03:12,520
This is incredibly useful because it allows a much greater number of tasks

46
00:03:12,520 --> 00:03:17,760
to be performed by one server. And that's really important for co-location.

47
00:03:17,760 --> 00:03:22,280
So as long as each virtual machine is separated well enough from the others,

48
00:03:22,280 --> 00:03:26,480
this can even bolster security. Load balancing is another technique

49
00:03:26,480 --> 00:03:29,480
that ensures that servers are being used efficiently.

50
00:03:29,480 --> 00:03:35,000
I mean, you don't want a situation where some of the servers are getting slammed with requests

51
00:03:35,000 --> 00:03:39,040
and extra processing while others are sitting idle

52
00:03:39,040 --> 00:03:43,160
like that kid who contributed nothing to your group projects back in school.

53
00:03:43,160 --> 00:03:48,800
So instead, servers are often configured to have their workload and their data balanced

54
00:03:48,800 --> 00:03:53,520
more evenly between them. This prevents overloading of certain machines

55
00:03:53,520 --> 00:03:57,280
and bottlenecks. This is also often done automatically

56
00:03:57,280 --> 00:04:02,360
in cases where one server might need to be taken down for maintenance so that whatever it was working on

57
00:04:02,360 --> 00:04:09,080
can just be picked up by other servers. In a similar vein, redundancy is a critical feature

58
00:04:09,080 --> 00:04:14,600
of any modern data center. Copies of data are usually kept on multiple servers

59
00:04:14,600 --> 00:04:17,680
or even across multiple data centers.

60
00:04:17,680 --> 00:04:21,000
And these facilities often have multiple pipelines

61
00:04:21,000 --> 00:04:25,960
leading out to the public internet in case one internet service provider has a problem.

62
00:04:25,960 --> 00:04:33,800
So all this means that enterprise-grade data centers are designed for at least 99.999% uptime,

63
00:04:34,240 --> 00:04:37,680
which works out to about five minutes

64
00:04:37,680 --> 00:04:42,680
of time offline per year. Some of them are actually even more reliable than that.

65
00:04:42,680 --> 00:04:49,120
And while websites and online services do obviously go down sometimes, it really is amazing

66
00:04:49,120 --> 00:04:56,040
how much has gone into making sure that we can access nearly anything nearly any time.

67
00:04:56,040 --> 00:04:59,240
Like if only everything in your life was that dependable, right?

68
00:04:59,240 --> 00:05:03,600
So thanks for watching, guys. Like, dislike, check out our other videos. Leave a comment if you have this suggestion

69
00:05:03,600 --> 00:05:08,240
for a future fast as possible and don't forget to subscribe because even the most reliable data center

70
00:05:08,240 --> 00:05:11,240
can't give you a notification if you're not subscribed, can it?