1
00:00:00,000 --> 00:00:03,200
Have you ever opened an Azure Bill and found a cost you didn't expect?

2
00:00:03,200 --> 00:00:06,320
Or found out a public app had more access than it should?

3
00:00:06,320 --> 00:00:09,560
Maybe an app stopped working late at night and nobody knew where to start.

4
00:00:09,560 --> 00:00:12,400
Those problems don't always mean you picked the wrong Azure service.

5
00:00:12,400 --> 00:00:14,860
Often the service works exactly as designed.

6
00:00:14,860 --> 00:00:16,800
The trouble comes from the decisions around it.

7
00:00:16,800 --> 00:00:17,960
How many copies do you need?

8
00:00:17,960 --> 00:00:18,880
Who can access the data?

9
00:00:18,880 --> 00:00:20,160
What happens when a part fails?

10
00:00:20,160 --> 00:00:21,080
Who checks the bill?

11
00:00:21,080 --> 00:00:23,720
Can the app cope when lots of users arrive at once?

12
00:00:23,720 --> 00:00:27,240
The Azure Well-Architected Framework helps you ask those questions

13
00:00:27,240 --> 00:00:28,920
before they become expensive problems?

14
00:00:28,920 --> 00:00:30,560
It isn't an Azure product you turn on.

15
00:00:30,560 --> 00:00:32,040
It isn't a compliance badge you earn.

16
00:00:32,040 --> 00:00:36,160
It's a decision framework for designing and running a workload that can hold up over time.

17
00:00:36,160 --> 00:00:39,560
Think of a workload as one business area inside a modern office building.

18
00:00:39,560 --> 00:00:42,000
Maybe the customer order desk or the payroll department.

19
00:00:42,000 --> 00:00:46,160
It needs rooms, people, locks, power records and a plan for when something breaks.

20
00:00:46,160 --> 00:00:48,400
Over the next few minutes, we'll look at the five pillars,

21
00:00:48,400 --> 00:00:52,000
the choices they force you to make and one simple way to review a workload.

22
00:00:52,000 --> 00:00:57,040
First, we need to separate the buildings shared foundation from the workload working inside it.

23
00:00:57,040 --> 00:00:59,200
Waff versus Caff, the runway and the plane.

24
00:00:59,200 --> 00:01:02,000
When people first meet the Azure Well-Architected Framework,

25
00:01:02,000 --> 00:01:05,280
they often mix it up with the Cloud Adoption Framework, also called Caff.

26
00:01:05,280 --> 00:01:07,400
They work together, but they answer different questions.

27
00:01:07,400 --> 00:01:08,920
Start with the word workload.

28
00:01:08,920 --> 00:01:11,440
A workload isn't just an app or a virtual machine.

29
00:01:11,440 --> 00:01:14,080
It's everything needed to produce a business result.

30
00:01:14,080 --> 00:01:19,320
For a customer portal that might include the website code, customer data, user identities,

31
00:01:19,320 --> 00:01:22,960
Azure services, monitoring, the team that supports it,

32
00:01:22,960 --> 00:01:25,320
and the daily work that keeps it running.

33
00:01:25,320 --> 00:01:27,480
The portal might look like one thing to a customer.

34
00:01:27,480 --> 00:01:29,440
Behind the scenes, it has many moving parts.

35
00:01:29,440 --> 00:01:31,920
The Cloud Adoption Framework looks at the company-wide foundation.

36
00:01:31,920 --> 00:01:34,680
It helps a company prepare a Azure for many workloads

37
00:01:34,680 --> 00:01:38,960
with shared rules for governance, security, networking, subscriptions and landing zones.

38
00:01:38,960 --> 00:01:42,800
A landing zone is simply a prepared place in Azure where workloads can live.

39
00:01:42,800 --> 00:01:46,760
Think of it as the shared building foundation, the front gate, the electrical system,

40
00:01:46,760 --> 00:01:49,320
the building rules and the shared security desk.

41
00:01:49,320 --> 00:01:52,840
The Well-Architected Framework focuses on one workload at a time.

42
00:01:52,840 --> 00:01:57,520
It asks whether this customer portal, this order system or this reporting app can do its job well.

43
00:01:57,520 --> 00:02:01,560
It doesn't inspect your whole Azure subscription as one giant pass or fail test.

44
00:02:01,560 --> 00:02:03,920
A simple way to remember the difference is an airport.

45
00:02:03,920 --> 00:02:05,400
SAF prepares the airport.

46
00:02:05,400 --> 00:02:09,480
It puts the runway, tower, security rules and shared services in place.

47
00:02:09,480 --> 00:02:11,240
WAF helps one plane fly safely.

48
00:02:11,240 --> 00:02:12,920
The plane still depends on the airport,

49
00:02:12,920 --> 00:02:17,400
but the flight team needs to think about its own route, fuel, passengers, safety checks,

50
00:02:17,400 --> 00:02:18,920
and what happens if the weather changes.

51
00:02:18,920 --> 00:02:21,960
Before you use WAF, write down what the workload must do.

52
00:02:21,960 --> 00:02:24,120
Who uses it, does it hold sensitive data?

53
00:02:24,120 --> 00:02:25,840
What does downtime cost the business?

54
00:02:25,840 --> 00:02:27,680
How much data could you afford to lose?

55
00:02:27,680 --> 00:02:29,080
What budget can support it?

56
00:02:29,080 --> 00:02:31,760
And what demand do you expect next month or next year?

57
00:02:31,760 --> 00:02:33,120
Those answers shape the design.

58
00:02:33,120 --> 00:02:37,720
Once the workload has a clear purpose, the five pillars become five practical questions.

59
00:02:37,720 --> 00:02:38,840
Reliability.

60
00:02:38,840 --> 00:02:40,760
Can your workload keep its promise?

61
00:02:40,760 --> 00:02:43,200
Reliability starts with a simple promise.

62
00:02:43,200 --> 00:02:45,600
Can people use the workload when they need it?

63
00:02:45,600 --> 00:02:48,920
Users won't remember that your Azure design looked tidy on a diagram.

64
00:02:48,920 --> 00:02:52,600
They remember the moment an order wouldn't go through, a file wouldn't open, or a payment

65
00:02:52,600 --> 00:02:53,960
set spinning on the screen.

66
00:02:53,960 --> 00:02:58,680
A reliable workload stays available when it can, recovers when something fails and protects

67
00:02:58,680 --> 00:03:01,000
data while that recovery happens.

68
00:03:01,000 --> 00:03:02,760
Think about the office building again.

69
00:03:02,760 --> 00:03:05,520
Reliable buildings have backup power when the main supply fails.

70
00:03:05,520 --> 00:03:07,760
They have fire exits when the usual route is blocked.

71
00:03:07,760 --> 00:03:11,680
They may have more than one elevator, so one broken elevator doesn't stop everyone from

72
00:03:11,680 --> 00:03:12,680
reaching their floor.

73
00:03:12,680 --> 00:03:14,680
They also have a plan that people have practiced.

74
00:03:14,680 --> 00:03:16,600
Your workload needs the same kind of thinking.

75
00:03:16,600 --> 00:03:21,040
Every failure can be prevented because networks fail, services become slow, software has bugs

76
00:03:21,040 --> 00:03:22,720
and people make mistakes.

77
00:03:22,720 --> 00:03:26,400
Reliability means preparing for those moments instead of hoping they won't happen.

78
00:03:26,400 --> 00:03:28,920
Start with the business promise, not the Azure service.

79
00:03:28,920 --> 00:03:32,640
How long can this workload be unavailable before the business feels real pain?

80
00:03:32,640 --> 00:03:34,120
That is your acceptable downtime.

81
00:03:34,120 --> 00:03:36,320
How much recent data can you lose if something breaks?

82
00:03:36,320 --> 00:03:38,640
Maybe losing five minutes of orders is manageable?

83
00:03:38,640 --> 00:03:40,600
Maybe losing even one payment record isn't?

84
00:03:40,600 --> 00:03:42,640
Then look at the user journeys that matter most.

85
00:03:42,640 --> 00:03:46,320
For an online shop, browsing products is useful, but placing an order and taking payment

86
00:03:46,320 --> 00:03:47,320
may matter more.

87
00:03:47,320 --> 00:03:50,320
Your reliability work should protect those moments first.

88
00:03:50,320 --> 00:03:52,320
From there you choose the building blocks.

89
00:03:52,320 --> 00:03:55,920
Availability zones can place parts of a workload in separate physical locations within

90
00:03:55,920 --> 00:03:56,920
an Azure region.

91
00:03:56,920 --> 00:03:59,920
If one location has a problem, the others can keep running.

92
00:03:59,920 --> 00:04:03,920
Backups protect copies of data, but a backup only helps if you can restore it when you need

93
00:04:03,920 --> 00:04:04,920
it.

94
00:04:04,920 --> 00:04:08,720
Health monitoring watches the workload and tells you when a service becomes unhealthy.

95
00:04:08,720 --> 00:04:11,320
Retrieves let an app try a short-lived failed request again.

96
00:04:11,320 --> 00:04:15,560
Circuit breakers stop an app from repeatedly calling a dependency that is already failing,

97
00:04:15,560 --> 00:04:19,040
which can prevent one problem from spreading through the rest of the workload.

98
00:04:19,040 --> 00:04:21,440
Picture a busy sales day for an online retailer.

99
00:04:21,440 --> 00:04:24,640
The order system needs a shipping service to show delivery choices.

100
00:04:24,640 --> 00:04:27,360
That shipping service starts failing under heavy demand.

101
00:04:27,360 --> 00:04:31,360
Without a plan, every checkout request waits for it, times out, and the whole order process

102
00:04:31,360 --> 00:04:32,360
slows down.

103
00:04:32,360 --> 00:04:36,000
With a circuit breaker and a fallback choice, the site can temporarily show a standard delivery

104
00:04:36,000 --> 00:04:37,400
option instead.

105
00:04:37,400 --> 00:04:38,760
Customers can still pay.

106
00:04:38,760 --> 00:04:39,960
Order still enter the system.

107
00:04:39,960 --> 00:04:43,680
The team can fix the shipping link without turning one failure into a full outage.

108
00:04:43,680 --> 00:04:44,920
That is graceful fallback.

109
00:04:44,920 --> 00:04:48,840
The workload gives up a less important feature so it can protect the main business action,

110
00:04:48,840 --> 00:04:50,560
but people often make two mistakes here.

111
00:04:50,560 --> 00:04:53,160
First they create a backup and assume the job is done.

112
00:04:53,160 --> 00:04:56,960
Then, during an incident, they discover nobody has tested the restore process, the backup

113
00:04:56,960 --> 00:05:00,320
is incomplete, or it takes longer than the business can accept.

114
00:05:00,320 --> 00:05:04,080
Second, they add extra copies of everything because redundancy sounds safe.

115
00:05:04,080 --> 00:05:05,960
Without agreeing on a recovery target first.

116
00:05:05,960 --> 00:05:09,640
More copies, more availability zones, and more regions can improve recovery.

117
00:05:09,640 --> 00:05:14,240
They also raise the Azure Bill and give the team more systems to monitor, test, and maintain.

118
00:05:14,240 --> 00:05:17,200
So the question isn't how much reliability can we buy.

119
00:05:17,200 --> 00:05:22,520
It is what level of reliability does this workload need and what are we willing to support?

120
00:05:22,520 --> 00:05:25,160
Keeping a service available is only part of the promise.

121
00:05:25,160 --> 00:05:29,480
You also need to decide who should be allowed through each door and into each room.

122
00:05:29,480 --> 00:05:32,200
Security, who gets through the reception desk?

123
00:05:32,200 --> 00:05:34,920
A workload can stay online all day and still fail the business.

124
00:05:34,920 --> 00:05:39,000
That happens when the wrong person can read customer records, change a payment setting,

125
00:05:39,000 --> 00:05:42,360
download private files, or take control of an admin account.

126
00:05:42,360 --> 00:05:46,280
Security means protecting identities, data, and systems from misuse and attack.

127
00:05:46,280 --> 00:05:49,440
Think of enter ID as the reception desk for your digital building.

128
00:05:49,440 --> 00:05:51,520
Before someone enters, it checks who they are.

129
00:05:51,520 --> 00:05:55,560
Then even after they enter, that person should only reach the rooms needed for their job.

130
00:05:55,560 --> 00:05:57,440
A customer might view their own order.

131
00:05:57,440 --> 00:06:01,600
A support worker might update delivery details and administrator might manage the service.

132
00:06:01,600 --> 00:06:03,920
Those are different jobs so they need different permissions.

133
00:06:03,920 --> 00:06:08,040
Giving everyone a master key is easier at first, but it creates a large problem later.

134
00:06:08,040 --> 00:06:10,040
This is the thinking behind zero trust.

135
00:06:10,040 --> 00:06:12,760
Don't trust a request just because it came from inside your network.

136
00:06:12,760 --> 00:06:13,760
Check the identity.

137
00:06:13,760 --> 00:06:15,680
Check what that identity is trying to do.

138
00:06:15,680 --> 00:06:19,520
Give the smallest amount of access needed and expect that one account or one device could

139
00:06:19,520 --> 00:06:21,040
eventually be compromised.

140
00:06:21,040 --> 00:06:22,920
For a beginner, start with identity.

141
00:06:22,920 --> 00:06:27,440
Use strong sign-in protection through Microsoft Enter ID, including multi-factor authentication

142
00:06:27,440 --> 00:06:28,440
where it fits.

143
00:06:28,440 --> 00:06:30,040
Give people only the roles they need.

144
00:06:30,040 --> 00:06:31,320
That is called least privilege.

145
00:06:31,320 --> 00:06:33,320
Your applications need identities too.

146
00:06:33,320 --> 00:06:38,000
Instead of placing a password or long-lived access key inside code, use managed identities

147
00:06:38,000 --> 00:06:39,160
where you can.

148
00:06:39,160 --> 00:06:42,600
Azure can then give an app a controlled identity so the app can reach the resource it needs

149
00:06:42,600 --> 00:06:44,400
without a secret sitting in a file.

150
00:06:44,400 --> 00:06:46,000
Data needs protection as well.

151
00:06:46,000 --> 00:06:47,920
First know what data the workload holds.

152
00:06:47,920 --> 00:06:51,560
Public product details need different treatment from payroll records, health data or customer

153
00:06:51,560 --> 00:06:52,840
payment information.

154
00:06:52,840 --> 00:06:57,840
That is data classification, naming the sensitivity of data so you can apply the right controls.

155
00:06:57,840 --> 00:07:00,920
Encrypt sensitive data when it moves and when it is stored.

156
00:07:00,920 --> 00:07:04,720
Separate network areas so a public website cannot freely reach a private database.

157
00:07:04,720 --> 00:07:08,280
If an attacker gets through one door, that separation can limit how far they move.

158
00:07:08,280 --> 00:07:10,720
That security doesn't begin after the app is built.

159
00:07:10,720 --> 00:07:12,360
Before building, ask what could go wrong.

160
00:07:12,360 --> 00:07:14,840
Could a user reach another user's record?

161
00:07:14,840 --> 00:07:16,880
Could a stolen account change a critical setting?

162
00:07:16,880 --> 00:07:19,440
Could an outside service send harmful data into the workload?

163
00:07:19,440 --> 00:07:23,800
This is threat modeling and it turns vague worries into things the team can design against.

164
00:07:23,800 --> 00:07:28,720
While building, protect the code, the deployment process and the settings that go into Azure.

165
00:07:28,720 --> 00:07:32,800
After release, watch for unusual activity, keep alerts useful and know who responds when

166
00:07:32,800 --> 00:07:34,000
something looks wrong.

167
00:07:34,000 --> 00:07:37,120
A locked building still needs someone watching the alarm panel.

168
00:07:37,120 --> 00:07:39,840
Many teams treat security like a firewall they add near the end.

169
00:07:39,840 --> 00:07:44,200
A firewall can help, but it can't fix a workload where every developer has broad admin

170
00:07:44,200 --> 00:07:49,400
rights or where static keys are copied into scripts, chat messages and source code.

171
00:07:49,400 --> 00:07:52,920
Security works best when it is part of every choice from sign-in to data storage to daily

172
00:07:52,920 --> 00:07:53,920
operations.

173
00:07:53,920 --> 00:07:55,080
There can be a trade-off.

174
00:07:55,080 --> 00:07:57,120
Extra checks may add another sign-in step.

175
00:07:57,120 --> 00:07:59,080
Encryption and inspection can add a small delay.

176
00:07:59,080 --> 00:08:02,640
Those are reasonable costs when the alternative is exposing data the business cannot afford to

177
00:08:02,640 --> 00:08:03,640
lose.

178
00:08:03,640 --> 00:08:08,440
Not balance comes from the risk around the workload, not from making every action as fast as possible,

179
00:08:08,440 --> 00:08:11,320
and a secure workload can still waste money quietly.

180
00:08:11,320 --> 00:08:13,480
A service may run overnight with no users.

181
00:08:13,480 --> 00:08:15,120
Old storage may keep growing.

182
00:08:15,120 --> 00:08:19,480
Large resources may sit mostly idle, while nobody knows which team owns the bill.

183
00:08:19,480 --> 00:08:21,640
The next pillar asks a direct question.

184
00:08:21,640 --> 00:08:24,720
Does every Azure pound or dollar have a clear job?

185
00:08:24,720 --> 00:08:26,040
Cost optimization?

186
00:08:26,040 --> 00:08:27,040
Spend on purpose.

187
00:08:27,040 --> 00:08:29,720
Cloud costs rarely jump for one dramatic reason.

188
00:08:29,720 --> 00:08:33,160
They grow because the test system stays on after everyone goes home, a service gets

189
00:08:33,160 --> 00:08:35,280
sized for a rush that never comes.

190
00:08:35,280 --> 00:08:39,400
Old files remain in expensive storage, and nobody checks whether the workload still needs

191
00:08:39,400 --> 00:08:41,800
what it is paying for.

192
00:08:41,800 --> 00:08:45,640
Cost optimization means getting the business results you need without paying for waste.

193
00:08:45,640 --> 00:08:48,440
It doesn't mean choosing the cheapest setting every time.

194
00:08:48,440 --> 00:08:50,600
Think about renting space in an office building.

195
00:08:50,600 --> 00:08:54,320
You need enough desks for the people working there, enough power for the equipment, storage

196
00:08:54,320 --> 00:08:57,400
for records, and perhaps extra space during a busy season.

197
00:08:57,400 --> 00:09:00,440
But you wouldn't keep paying for three empty floors all year because you might need them

198
00:09:00,440 --> 00:09:01,760
for one afternoon.

199
00:09:01,760 --> 00:09:05,760
As your works in a similar way, you pay for compute, storage, network traffic, backups,

200
00:09:05,760 --> 00:09:07,720
logs, and other services.

201
00:09:07,720 --> 00:09:10,200
Each item should have a clear reason for being there.

202
00:09:10,200 --> 00:09:11,840
Start by making ownership visible.

203
00:09:11,840 --> 00:09:14,520
Taxes are labels you attach to Azure resources.

204
00:09:14,520 --> 00:09:18,240
Attack can tell you which workload uses a resource, which environment it belongs to, and

205
00:09:18,240 --> 00:09:19,480
who owns it.

206
00:09:19,480 --> 00:09:22,760
When a cost appears, the team should not need to guess who can explain it.

207
00:09:22,760 --> 00:09:26,920
Give each workload an owner, set a budget, then set alerts before the budget is reached.

208
00:09:26,920 --> 00:09:31,120
An alert doesn't stop spending by itself, but it gives the team time to investigate.

209
00:09:31,120 --> 00:09:35,080
A regular cost review turns that alert into a habit instead of a surprise at the end of

210
00:09:35,080 --> 00:09:36,080
the month.

211
00:09:36,080 --> 00:09:37,840
Then look at how the workload actually behaves.

212
00:09:37,840 --> 00:09:39,560
A service may need a smaller size.

213
00:09:39,560 --> 00:09:41,840
It may need to scale down when demand drops.

214
00:09:41,840 --> 00:09:44,000
A resource that nobody uses can be removed.

215
00:09:44,000 --> 00:09:48,520
Old files can move to a lower cost storage tier through storage lifecycle rules, or be deleted

216
00:09:48,520 --> 00:09:50,480
when the business no longer needs them.

217
00:09:50,480 --> 00:09:51,640
Backups need the same thought.

218
00:09:51,640 --> 00:09:55,400
Keep backups for as long as the business and any rules require, but don't keep every

219
00:09:55,400 --> 00:09:58,040
backup forever just because a default setting allowed it.

220
00:09:58,040 --> 00:10:01,480
The retention period should match the recovery promise the workload has made.

221
00:10:01,480 --> 00:10:04,520
Once you understand normal usage, you can look at rates.

222
00:10:04,520 --> 00:10:06,880
Some workloads run at a steady level every day.

223
00:10:06,880 --> 00:10:10,760
For those reservations, savings plans or fixed price choices may lower the cost, but don't

224
00:10:10,760 --> 00:10:14,160
buy a long commitment before you understand what the workload really uses.

225
00:10:14,160 --> 00:10:17,400
A discount on the wrong service is still wasted money.

226
00:10:17,400 --> 00:10:20,480
And this is where teams can cause damage while trying to save money.

227
00:10:20,480 --> 00:10:22,000
They remove a copy of important data.

228
00:10:22,000 --> 00:10:24,240
They shorten backups without checking recovery needs.

229
00:10:24,240 --> 00:10:25,960
They turn off security controls.

230
00:10:25,960 --> 00:10:29,240
The bill may look better this month, but the risk has simply moved somewhere else.

231
00:10:29,240 --> 00:10:32,740
A cheap design that cannot recover when needed is expensive when it fails, so phrase the

232
00:10:32,740 --> 00:10:33,740
goal carefully.

233
00:10:33,740 --> 00:10:35,880
Don't say, make this workload cheaper.

234
00:10:35,880 --> 00:10:39,480
Say remove waste while keeping the agreed service recovery and security levels.

235
00:10:39,480 --> 00:10:41,120
That puts cost in its proper place.

236
00:10:41,120 --> 00:10:43,120
You are not trying to spend as little as possible.

237
00:10:43,120 --> 00:10:45,220
You are trying to spend on purpose.

238
00:10:45,220 --> 00:10:47,920
Once spending has an owner, another question appears.

239
00:10:47,920 --> 00:10:52,160
Can the team run this workload every ordinary Tuesday, without depending on one person who

240
00:10:52,160 --> 00:10:54,960
happens to remember where every switch is?

241
00:10:54,960 --> 00:10:57,440
An excellent team can run it on Tuesday.

242
00:10:57,440 --> 00:10:58,800
Launched it gets attention.

243
00:10:58,800 --> 00:11:02,560
People prepare the demo, watch the first users arrive, and breathe a sigh of relief when

244
00:11:02,560 --> 00:11:03,560
the app works.

245
00:11:03,560 --> 00:11:07,120
But the harder part comes after that, when patches need to be applied, changes need to

246
00:11:07,120 --> 00:11:10,960
be released, alerts fire at night and people move to new jobs.

247
00:11:10,960 --> 00:11:15,120
Operational excellence means making daily work repeatable, making changes safe, seeing problems

248
00:11:15,120 --> 00:11:18,280
clearly and learning from them when they happen.

249
00:11:18,280 --> 00:11:19,880
Think of a well run office building.

250
00:11:19,880 --> 00:11:24,120
It has an operating manual, an alarm panel, a maintenance schedule and staff who know

251
00:11:24,120 --> 00:11:25,120
their roles.

252
00:11:25,120 --> 00:11:27,960
It doesn't depend on one person keeping every detail in their head.

253
00:11:27,960 --> 00:11:30,480
A workload needs that same discipline.

254
00:11:30,480 --> 00:11:32,080
Infrastructure as code is one building block.

255
00:11:32,080 --> 00:11:35,800
Instead of creating resources by clicking through the Azure portal and hoping someone remembers

256
00:11:35,800 --> 00:11:37,920
the settings, you describe the setup and files.

257
00:11:37,920 --> 00:11:41,640
The team can review those files, store them with the application code and create the same

258
00:11:41,640 --> 00:11:43,920
environment again, when needed.

259
00:11:43,920 --> 00:11:45,080
Deployment pipelines are another part.

260
00:11:45,080 --> 00:11:47,960
A pipeline can test code and settings before they reach production.

261
00:11:47,960 --> 00:11:50,440
It can apply the same release steps each time.

262
00:11:50,440 --> 00:11:54,000
If a release goes wrong, the team has a clearer path to stop or reverse it.

263
00:11:54,000 --> 00:11:55,840
You also need to see what users experience it.

264
00:11:55,840 --> 00:11:57,480
This is called observability.

265
00:11:57,480 --> 00:11:59,720
It sounds technical but the idea is simple.

266
00:11:59,720 --> 00:12:04,440
Collect enough signals to answer is the workload healthy and if it isn't, where is the problem?

267
00:12:04,440 --> 00:12:06,000
Logs record events.

268
00:12:06,000 --> 00:12:09,280
Matrix show numbers over time such as errors or response times.

269
00:12:09,280 --> 00:12:12,880
Traces follow one request as it moves through different parts of the workload.

270
00:12:12,880 --> 00:12:16,000
Health signals tell you whether an important service can still do its job.

271
00:12:16,000 --> 00:12:20,400
Together, those signals feed alerts and dashboards so the team can see trouble before a customer

272
00:12:20,400 --> 00:12:21,400
reports it.

273
00:12:21,400 --> 00:12:23,600
Picture a small team with a customer portal.

274
00:12:23,600 --> 00:12:27,320
One experienced engineer knows a manual fix for a problem with a background process.

275
00:12:27,320 --> 00:12:30,840
They open the Azure portal, change a setting, restart a service and it works.

276
00:12:30,840 --> 00:12:33,080
Nobody writes the steps down because it feels simple.

277
00:12:33,080 --> 00:12:35,160
Months later, the issue returns late at night.

278
00:12:35,160 --> 00:12:36,640
That engineer is unavailable.

279
00:12:36,640 --> 00:12:40,720
The person on call sees a vague alert, searches through old messages and worries about changing

280
00:12:40,720 --> 00:12:41,720
the wrong setting.

281
00:12:41,720 --> 00:12:45,000
The problem lasts longer because the knowledge only lived in one person's memory.

282
00:12:45,000 --> 00:12:46,280
A runbook would have changed that.

283
00:12:46,280 --> 00:12:49,480
A runbook is a plain set of instructions for a known situation.

284
00:12:49,480 --> 00:12:53,880
It explains what to check, what actions are safe, who needs to know and when to escalate.

285
00:12:53,880 --> 00:12:57,640
Add automation where it makes sense and the team no longer has to repeat the same manual

286
00:12:57,640 --> 00:12:58,920
fix under pressure.

287
00:12:58,920 --> 00:13:02,920
The common mistake is treating the Azure portal as the normal way to run everything.

288
00:13:02,920 --> 00:13:05,840
The portal is useful for looking, learning and investigating.

289
00:13:05,840 --> 00:13:10,320
But when routine changes depend on clicks, memory and a particular person, environments

290
00:13:10,320 --> 00:13:12,920
drift apart and mistakes become harder to spot.

291
00:13:12,920 --> 00:13:14,800
Good operations create a loop.

292
00:13:14,800 --> 00:13:17,760
After an incident, don't start by looking for someone to blame.

293
00:13:17,760 --> 00:13:21,200
Check what happened, what the signals showed, what slowed the response and what the team

294
00:13:21,200 --> 00:13:22,200
can change.

295
00:13:22,200 --> 00:13:24,120
Write down the decision and its trade-offs.

296
00:13:24,120 --> 00:13:25,480
Add the improvement to the backlog.

297
00:13:25,480 --> 00:13:29,920
Then test it so the same failure does not return as an unwelcome surprise.

298
00:13:29,920 --> 00:13:34,960
When daily work becomes predictable, the workload becomes easier to change and safer to support.

299
00:13:34,960 --> 00:13:36,560
That leads to the final question.

300
00:13:36,560 --> 00:13:41,400
When demand rises, does the workload stay responsive at the moment users need it most?

301
00:13:41,400 --> 00:13:42,880
Performance efficiency?

302
00:13:42,880 --> 00:13:44,440
Fast enough when it counts.

303
00:13:44,440 --> 00:13:47,840
An app can stay online and still frustrate every person using it.

304
00:13:47,840 --> 00:13:49,400
A page takes too long to load.

305
00:13:49,400 --> 00:13:51,120
A payment request waits in a queue.

306
00:13:51,120 --> 00:13:55,480
A customer clicks twice because nothing seems to happen and then both requests fail.

307
00:13:55,480 --> 00:13:59,160
Performance efficiency means using the right Azure resources and design choices to meet

308
00:13:59,160 --> 00:14:00,840
real demand when it arrives.

309
00:14:00,840 --> 00:14:02,960
Picture a busy office at 9 in the morning.

310
00:14:02,960 --> 00:14:07,080
You need enough elevators to move people upstairs, enough open desks for people to work and

311
00:14:07,080 --> 00:14:09,440
enough staff at reception to keep the line moving.

312
00:14:09,440 --> 00:14:12,480
But you wouldn't keep every floor fully staffed through the night just because Monday morning

313
00:14:12,480 --> 00:14:13,480
gets busy.

314
00:14:13,480 --> 00:14:16,960
But with what users expect, how quickly must the main page respond?

315
00:14:16,960 --> 00:14:20,360
How many orders, messages or requests need processing each minute?

316
00:14:20,360 --> 00:14:21,720
When do busy periods happen?

317
00:14:21,720 --> 00:14:24,240
Which transaction matters most when demand rises?

318
00:14:24,240 --> 00:14:28,640
And if the workload grows next year, what part is likely to slow down first?

319
00:14:28,640 --> 00:14:31,040
Those questions give you something useful to test.

320
00:14:31,040 --> 00:14:34,160
Load testing creates realistic demand before customers do.

321
00:14:34,160 --> 00:14:37,560
Autoscale can add capacity when demand rises and reduce it later.

322
00:14:37,560 --> 00:14:41,280
Caching keeps frequently requested information close at hand, so the workload doesn't keep

323
00:14:41,280 --> 00:14:43,040
asking the same database question.

324
00:14:43,040 --> 00:14:46,880
Choose can hold work safely when one part of the system needs time to catch up.

325
00:14:46,880 --> 00:14:50,000
The right service tier gives you enough capacity for the job.

326
00:14:50,000 --> 00:14:53,280
Monitoring helps you find the actual bottleneck rather than guessing.

327
00:14:53,280 --> 00:14:55,960
Performance and cost often meet in the same decision.

328
00:14:55,960 --> 00:15:00,160
Autoscale may give users a faster experience during a rush, while also avoiding the cost

329
00:15:00,160 --> 00:15:02,040
of running at full capacity all day.

330
00:15:02,040 --> 00:15:06,720
But that only works when you know what demand looks like and what signal should trigger scaling.

331
00:15:06,720 --> 00:15:08,360
Guessing is where many teams go wrong.

332
00:15:08,360 --> 00:15:11,920
They buy the largest option for every part of the workload, hoping that more power will

333
00:15:11,920 --> 00:15:12,920
solve every problem.

334
00:15:12,920 --> 00:15:16,640
Or they test during a quiet afternoon, see that everything works and assume the design

335
00:15:16,640 --> 00:15:17,640
is ready.

336
00:15:17,640 --> 00:15:21,040
Imagine a team testing an order system with a steady stream of requests.

337
00:15:21,040 --> 00:15:22,600
At first everything looks fine.

338
00:15:22,600 --> 00:15:26,160
Then after 20 minutes, a queue begins to grow a little faster than it empties.

339
00:15:26,160 --> 00:15:29,240
The website still responds, but a background process cannot keep up.

340
00:15:29,240 --> 00:15:32,200
Given a few more hours, orders would begin waiting far too long.

341
00:15:32,200 --> 00:15:34,920
The test found the real bottleneck before customers did.

342
00:15:34,920 --> 00:15:35,920
That is the point.

343
00:15:35,920 --> 00:15:37,880
Performance isn't about buying the biggest machine.

344
00:15:37,880 --> 00:15:43,000
It is about measuring the work, finding the slow part and meeting the promise when it counts.

345
00:15:43,000 --> 00:15:45,040
Every change here touches more than one pillar.

346
00:15:45,040 --> 00:15:46,560
Faster capacity may cost more.

347
00:15:46,560 --> 00:15:48,760
A cache may change how data stays current.

348
00:15:48,760 --> 00:15:50,680
More scaling may add work for the team.

349
00:15:50,680 --> 00:15:53,200
So the real skill is making those choices visible.

350
00:15:53,200 --> 00:15:56,040
The five pillars are not separate rooms with locked doors.

351
00:15:56,040 --> 00:15:59,120
They share the same building budget and the same floor plan.

352
00:15:59,120 --> 00:16:01,080
Tradeoffs and the practical WAF review.

353
00:16:01,080 --> 00:16:03,400
No workload gets a perfect score in every pillar.

354
00:16:03,400 --> 00:16:05,400
A good design does something more useful.

355
00:16:05,400 --> 00:16:09,400
It records the choices the team made, why they made them and what they accepted in return.

356
00:16:09,400 --> 00:16:10,760
Take cost as an example.

357
00:16:10,760 --> 00:16:13,360
Reduce costs sound simple, but it can lead to bad decisions.

358
00:16:13,360 --> 00:16:17,840
A better goal is, reduce cost while keeping our agreed recovery and security targets.

359
00:16:17,840 --> 00:16:20,080
That sentence keeps the business promise in view.

360
00:16:20,080 --> 00:16:23,600
Multi-region copies can improve recovery after a large outage, but they cost more to run

361
00:16:23,600 --> 00:16:24,600
and maintain.

362
00:16:24,600 --> 00:16:28,720
Stronger sign-in checks can add a small delay, but they lower the chance of the wrong person

363
00:16:28,720 --> 00:16:29,720
entering.

364
00:16:29,720 --> 00:16:33,480
Rapid releases can help a business move quickly, but they need safer testing and release

365
00:16:33,480 --> 00:16:36,560
controls, so speed does not create avoidable failures.

366
00:16:36,560 --> 00:16:38,560
These aren't mistakes, they are design choices.

367
00:16:38,560 --> 00:16:42,480
The well-architected framework gives you two useful tools for discussing them.

368
00:16:42,480 --> 00:16:44,080
Design principles guide your thinking.

369
00:16:44,080 --> 00:16:48,160
They help you ask the right sort of question before choosing a service or pattern.

370
00:16:48,160 --> 00:16:50,320
Checklists turn that thinking into actions.

371
00:16:50,320 --> 00:16:55,080
They help the team look for gaps such as an untested recovery process, unclear cost ownership

372
00:16:55,080 --> 00:16:57,480
or missing signals for a busy application.

373
00:16:57,480 --> 00:16:58,960
Keep the first review small.

374
00:16:58,960 --> 00:17:03,000
Don't begin with every Azure subscription, every team, and every resource your company

375
00:17:03,000 --> 00:17:04,000
has ever created.

376
00:17:04,000 --> 00:17:08,920
Pick one workload with a clear business purpose, perhaps a customer portal, an ordering system,

377
00:17:08,920 --> 00:17:11,360
or an internal app, people depend on each day.

378
00:17:11,360 --> 00:17:13,240
Then use the Azure Well-Architected review.

379
00:17:13,240 --> 00:17:17,360
It asks roughly 60 questions across the five pillars and produces recommendations based

380
00:17:17,360 --> 00:17:18,360
on your answers.

381
00:17:18,360 --> 00:17:20,280
Treat it as a discussion to a not an exam.

382
00:17:20,280 --> 00:17:22,200
You are not trying to earn a pass mark.

383
00:17:22,200 --> 00:17:25,440
You are trying to expose risks, make decisions, and find work.

384
00:17:25,440 --> 00:17:27,040
The team can actually do.

385
00:17:27,040 --> 00:17:29,560
Turn the recommendations into a prioritized backlog.

386
00:17:29,560 --> 00:17:34,240
Some actions may need attention soon because they protect important data or recovery needs.

387
00:17:34,240 --> 00:17:36,040
Others can wait until the next plan change.

388
00:17:36,040 --> 00:17:39,560
Assign an owner, record the reason, and keep the work visible.

389
00:17:39,560 --> 00:17:44,040
Azure Advisor can give you another source of improvement ideas from your Azure environment.

390
00:17:44,040 --> 00:17:48,400
Its signals can help point out places to investigate, but the workload team still decides what

391
00:17:48,400 --> 00:17:49,720
fits its needs.

392
00:17:49,720 --> 00:17:54,800
Run the review again after a major release, a large design change, or on a regular cycle.

393
00:17:54,800 --> 00:17:58,160
Compare milestones over time so you can see whether the workload is improving instead

394
00:17:58,160 --> 00:17:59,480
of relying on memory.

395
00:17:59,480 --> 00:18:03,640
Finish with one workload, one review, and a written list of decisions the team can act

396
00:18:03,640 --> 00:18:05,240
on.

397
00:18:05,240 --> 00:18:07,560
One workload, one better set of decisions.

398
00:18:07,560 --> 00:18:11,840
The Azure Well-Architected framework turns scattered azure choices into balanced decisions

399
00:18:11,840 --> 00:18:13,600
for one workload.

400
00:18:13,600 --> 00:18:16,080
Reliability keeps its promises when parts fail.

401
00:18:16,080 --> 00:18:18,600
Security controls who can enter and what they can reach.

402
00:18:18,600 --> 00:18:22,240
Cost optimization removes waste without cutting agreed protection.

403
00:18:22,240 --> 00:18:24,760
Operational excellence makes daily work repeatable.

404
00:18:24,760 --> 00:18:28,200
Performance efficiency helps the workload meet demand when users need it.

405
00:18:28,200 --> 00:18:31,440
Start with the workload that has the clearest impact on your business.

406
00:18:31,440 --> 00:18:32,760
Write down what it needs.

407
00:18:32,760 --> 00:18:36,920
Its uptime target recovery needs the data it holds, the budget it can support, and the

408
00:18:36,920 --> 00:18:38,760
demand it must handle.

409
00:18:38,760 --> 00:18:40,800
Keep the answers simple, but write them down.

410
00:18:40,800 --> 00:18:42,320
The team can improve a written promise.

411
00:18:42,320 --> 00:18:45,520
It can't improve assumptions that live in separate people's heads.

412
00:18:45,520 --> 00:18:50,240
Then run the Azure Well-Architected review, choose the first ten actions that fit the workload,

413
00:18:50,240 --> 00:18:51,640
give every action an owner.

414
00:18:51,640 --> 00:18:55,200
Record the trade-offs, especially when a decision saves money, adds protection, or changes

415
00:18:55,200 --> 00:18:56,720
recovery expectations.

416
00:18:56,720 --> 00:19:01,440
It gives your team a list they can work through, not another document that gets forgotten.

417
00:19:01,440 --> 00:19:06,320
Subscribe to m365.fm for the next knowledge nugget, where we'll keep connecting the Azure

418
00:19:06,320 --> 00:19:09,560
foundations and cloud services that make up your Microsoft platform.

