How to approach System Design? — Transcript

Full transcript

1. 0:00so system design is tricky but it does
2. 0:01not have to be difficult in this video
3. 0:03i'll walk you through on the two
4. 0:04approaches that i have been using
5. 0:06to build massively scalable systems over
6. 0:09last 10 years of my career right we'll
7. 0:11start with what is system design then
8. 0:12we'll take a look at the two approaches
9. 0:14and then i'll share the three key
10. 0:15pointers that you should remember while
11. 0:17you're designing any system out there
12. 0:19but before we move forward i'd want to
13. 0:20talk to you about a code based course of
14. 0:22system design that i have been running
15. 0:23since march 2021 and if you're looking
16. 0:26to learn system design from the first
17. 0:27principles this course is for you yeah
18. 0:30because this is a cohort based course it
19. 0:31will not just be me rambling a
20. 0:34semi-optimized solution thinking it's
21. 0:35the most amazing solution out there
22. 0:37instead it will be a collaborative
23. 0:39environment where every single person
24. 0:40who is part of the cohort
25. 0:42will can pitch in his or her ideas and
26. 0:45we will evolve our system around that
27. 0:47right every single problem statement
28. 0:49comes with a brainstorming session where
29. 0:51we all together brainstorm and evolve
30. 0:53our system that's why everyone
31. 0:55understands the kind of trade-offs we
32. 0:57made while making that decision instead
33. 0:59of just saying hey we'll use a
34. 1:00particular queue we'll have the
35. 1:02justification why we use only that queue
36. 1:04why we use that particular database why
37. 1:06sequel why not no sql right how are we
38. 1:08leveraging throughput how are we
39. 1:10ensuring that our system scales that's
40. 1:11the highlight of this course this course
41. 1:13is taken by more than 500 engineers to
42. 1:16date
43. 1:18spanning nine countries and seven
44. 1:19cohorts right people from all top
45. 1:21companies have taken this course
46. 1:23and the outline is very intriguing it's
47. 1:26very exciting so we start with week one
48. 1:28around we start with the core foundation
49. 1:30of the course where we design online
50. 1:32offline indicator then we try to design
51. 1:33our own medium then we go into database
52. 1:36where we go in depth of database logging
53. 1:38and take and see few very amazing
54. 1:39examples of data log or database logging
55. 1:42in in action and how do we ensure that
56. 1:45our system scales through that
57. 1:48then the third week is all about going
58. 1:49distributed where we design our load
59. 1:52balancer i'll walk you through the
60. 1:53actual code of
61. 1:54a toy load balancer and understand how
62. 1:56tcp connections are managed and how
63. 1:57simple it is to build load balancer then
64. 1:59week four is about all about social
65. 2:01networks week five is all about building
66. 2:03your own storage indians like we'll
67. 2:04build that intuition on if you were to
68. 2:06ever design your storage agent how would
69. 2:08you do that right then week six is about
70. 2:10building high throughput system seven is
71. 2:12about building
72. 2:13uh ir systems basically information
73. 2:15retrieval systems and add-on designs
74. 2:16where we design our own message brokers
75. 2:18like sqs where we design distributed
76. 2:20task scheduler and we conclude the
77. 2:22course with week 8 where we talk about
78. 2:24the super clever algorithms that has
79. 2:26powered or that has made those systems
80. 2:28possible right i have also attached a
81. 2:31video verbatim as is from my first code
82. 2:34where we designed and scaled instagram
83. 2:35notifications i will highly encourage
84. 2:36you to check this video out right and
85. 2:39now back to the video
86. 2:41so what is system design system design
87. 2:43to be put it in a very simple terms it's
88. 2:46when your customer has some requirement
89. 2:48when hey let's say your customer says i
90. 2:50need a live class solution hey i need
91. 2:52real time interaction hey i need to
92. 2:55upload my data into a particular
93. 2:57location right so there is some customer
94. 2:59need which is then translated into some
95. 3:01product and business requirement then
96. 3:03your product manager or your business uh
97. 3:05team sends a requirement to the product
98. 3:06even product team designs that part and
99. 3:08then it comes to you right and that's
100. 3:10what translates into that system so
101. 3:12everything
102. 3:13boils down to the business or the
103. 3:15product problem statement and then you
104. 3:16have to solve it right a system what
105. 3:18exactly system so system is a very
106. 3:20generic part but what exactly is that so
107. 3:22system could be that what you are
108. 3:24supposed to solve could be an
109. 3:26application an end to end application
110. 3:28having a front end and a back end that
111. 3:31user uses day in and day out it could be
112. 3:33a micro service it could just be a pure
113. 3:36microsoft pure engineering solution
114. 3:38where your target customer is some other
115. 3:40team it is not your end user but some
116. 3:42other team which is sitting right next
117. 3:44to you you are building micro service
118. 3:45for them it's a pure engineering
119. 3:47solution that you are trying to build
120. 3:48then it could be just a library that
121. 3:51everyone else was struggling with and
122. 3:52you thought of building it for example
123. 3:54building your own web
124. 3:56basically building your own web
125. 3:57framework could be just a library that
126. 3:59you are building it to make your overall
127. 4:02http request response very smooth
128. 4:04right or it could be an actual physical
129. 4:07hardware that you are shipping where you
130. 4:09were given a product requirement and the
131. 4:11only thing that you could do is actually
132. 4:12ship that hardware that embedded system
133. 4:15to your users or to your customers
134. 4:18right now when someone says design a
135. 4:21system
136. 4:22it could be one or all of these three
137. 4:25things first a high level architecture
138. 4:27diagram
139. 4:29where you get to see a macro bird's eye
140. 4:31view of that system where your design we
141. 4:33are supposed to design a macro bird's
142. 4:35eye view of that system where hey these
143. 4:36would be the micro services that talk to
144. 4:38each other and then you are and then and
145. 4:41then this is how the interaction would
146. 4:42happen this how this basically this is
147. 4:44how the persistence would happen and all
148. 4:45on a high level architecture side right
149. 4:49then it could also be a logical design
150. 4:52in which they ask you to
151. 4:55do like you are required to do business
152. 4:57logic algorithms core data structures
153. 5:00right core storage techniques that you
154. 5:02would be using core persistent
155. 5:04algorithms that you would be using that
156. 5:06part so
157. 5:07it's basically not just the high level
158. 5:09but you are scratching the surface and
159. 5:10going a level deep into identifying hey
160. 5:13what exactly would i use over here and
161. 5:15then the third design or or basically
162. 5:17then the third aspect could be a
163. 5:18physical design where you are supposed
164. 5:20to design the storage part of it the
165. 5:22actual i o the actual hardware which
166. 5:24instance type which gpu which processor
167. 5:27that you are using what would be the
168. 5:28overall capacity of it how would the
169. 5:30backup and restore would happen how
170. 5:32would you structure the data pipelines
171. 5:34and all right so these are the three
172. 5:36things like now depending on what you
173. 5:38are discussing or what you are designing
174. 5:40where you are designing what is the
175. 5:42state there you would be required to
176. 5:44design one or all three
177. 5:46be it your work your then you are
178. 5:48definitely supposed to design all three
179. 5:50if you are in the interview then i would
180. 5:51highly encourage you to limit the scope
181. 5:53of your discussion and pick one and then
182. 5:55solve it really nicely right now talking
183. 5:57about how to approach system design the
184. 6:00two approaches that i've used in my
185. 6:02career the first one is what i call a
186. 6:04spiral approach the spiral approach is
187. 6:06where you decide the core of your system
188. 6:09design and then you start building
189. 6:10around it it so then your system evolves
190. 6:13as a spiral
191. 6:15right now for example you start with
192. 6:17storage hey that you decided hey i want
193. 6:18to use this particular database and then
194. 6:20you move to the next part you have now
195. 6:22that i have database i want i'd want a
196. 6:24bunch of api servers to to basically
197. 6:26support it right and then
198. 6:28you evolve to the next part where hey
199. 6:30you said now i have this service now i
200. 6:32need another service right for example
201. 6:34this is the first step that you started
202. 6:35with a simple database then you added a
203. 6:38a a very simple way to interact with
204. 6:40that this forms the core of your then
205. 6:42you evolve they do say that hey i need a
206. 6:44couple of services to interact with each
207. 6:46service having its own database let's
208. 6:48say payment and a payment gateway the
209. 6:49next part is you have a payment and
210. 6:51payment gateway then you need a retry
211. 6:52mechanism to retry the fail payments and
212. 6:54all right so this is how you would
213. 6:57evolve your system now when would you
214. 6:59use a spiral based evolution where you
215. 7:01are pretty confident on the decisions
216. 7:03you are making that's a key thing when
217. 7:06you are experienced enough or when you
218. 7:07know the problem statement is simple
219. 7:08enough that that the decision that you
220. 7:10are making is very crisp is very optimal
221. 7:13and this is exactly how it should be
222. 7:15done you are very confident on that
223. 7:16spiral way of designing systems works
224. 7:19the best right these systems are pretty
225. 7:21predictable on how they should be
226. 7:22designed and you are just evolving them
227. 7:25very quickly
228. 7:27the second approach is an incremental
229. 7:28mvp approach where you see a bunch where
230. 7:31you start small and then you add more
231. 7:33and more features to it just to handle
232. 7:35larger larger larger scale of it right
233. 7:37these is how i represented this is
234. 7:39through non-concentric circles because
235. 7:42you know
236. 7:43you started here then you added then you
237. 7:45added some part to add more skills but
238. 7:47then you added some more and then this
239. 7:49oval which is which is very obscured in
240. 7:51its stay in its shape
241. 7:54because you thought your system is
242. 7:56evolving in a certain direction but
243. 7:57because of change because of scale it
244. 7:59evolved into some other so it's the way
245. 8:01i represented its non-concentric
246. 8:04circles and ovals or non-concentric
247. 8:06ellipses
248. 8:08right so here the idea is you start with
249. 8:11a very simple extremely basic day zero
250. 8:14architecture
251. 8:16extremely simple and then you think hey
252. 8:19now
253. 8:20this is my desire architecture let's say
254. 8:23i got more users now what for example
255. 8:26you started with a single user a single
256. 8:28api server and a single database now you
257. 8:30hit the next skill what's the next skill
258. 8:31you got multiple users so if you got
259. 8:33multiple users one api server might not
260. 8:35be able to handle it so you add a load
261. 8:36balancer right then what you do then
262. 8:38what you do you think about the next
263. 8:40step hey this system was working fine
264. 8:42but now when i got even more users my
265. 8:44system started to choke up so then what
266. 8:46you do is you added a read replica to
267. 8:48your database to handle the load then
268. 8:50you thought hey some workers or some
269. 8:51things are happening synchronously which
270. 8:53could very well happen asynchronously so
271. 8:54you add a you then you add an
272. 8:56asynchronous uh uh so basically then you
273. 8:59add a message broker to it to introduce
274. 9:01an asynchronous workflow into a system
275. 9:03design so
276. 9:04tackling one scale or one level at a
277. 9:07time is how you go for an incremental
278. 9:09mvp approach where incrementally you are
279. 9:12trying to evolve your system to just be
280. 9:14able to sustain the next scale right and
281. 9:18this is how these are two
282. 9:20very similar yet yet different
283. 9:23approaches of system design that i would
284. 9:25highly encourage you to pick your
285. 9:27problem with like
286. 9:28basically given a solution pick one over
287. 9:30other in which you are very confident
288. 9:32hey i know the like for example if you
289. 9:34know the the solution very well go for a
290. 9:36spiral based approach if you don't know
291. 9:38the solution variable start with the day
292. 9:390 architecture and then scale individual
293. 9:41component out right and then break it
294. 9:43into other services and all right
295. 9:46so key pointers to think about when you
296. 9:49are approaching a system design is first
297. 9:50of all every single system out there is
298. 9:53infinitely buildable this is something
299. 9:54that you need to understand you can add
300. 9:56as many features as you want into this
301. 9:59but depending on the time you have for
302. 10:01you to deliver your project or for you
303. 10:02to crack that interview in either case
304. 10:05it's very important to fence your
305. 10:07particular system
306. 10:09you have to restrict the scope
307. 10:11right that in the stipulated time of
308. 10:14let's say while you are at work
309. 10:16you know that this is what like you have
310. 10:19two weeks to actually implement it you
311. 10:21cannot spend more than two weeks to ship
312. 10:23this particular thing out so you would
313. 10:25be restricting the scope of your design
314. 10:28according to that and hey this is only
315. 10:30i'd be committing to ship and this is
316. 10:32something that you are predicting that
317. 10:33you will be very easily be able to ship
318. 10:35it right so restricting the scope is
319. 10:37very important and even same thing
320. 10:38during your interview is a technical
321. 10:39discussion hey restrain the scope so
322. 10:42that you design that one thing really
323. 10:44well right second key pointer is see
324. 10:46clarifications from your seniors or your
325. 10:48interviewer right because every single
326. 10:50system design problem is actually
327. 10:52ambiguous no one has an idea on how that
328. 10:54system is going to evolve so so seek
329. 10:56those clarifications on hey is this the
330. 10:58only requirement would this requirement
331. 11:00change what would happen if this if if a
332. 11:04breaks like seek seek those
333. 11:05clarifications on are we okay to take a
334. 11:08down time of five minutes are we okay to
335. 11:10do x y z like depending on the system at
336. 11:13hand seek those clarifications it would
337. 11:15help you design a very optimal system
338. 11:18third and the most important key pointer
339. 11:20is ask those critical questions
340. 11:22ask us critical questions that challenge
341. 11:25design decisions for example protocol a
342. 11:27versus protocol b
343. 11:29why why are we using http why can't we
344. 11:31go for grpc as those critical decisions
345. 11:34ask those critical questions that
346. 11:36actually challenges your design decision
347. 11:37hey
348. 11:38i think we should not be using web
349. 11:40sockets over here then suddenly your
350. 11:42entire architecture changes like why do
351. 11:45why does this need to be real time can
352. 11:46this not be your synchronous then
353. 11:48suddenly your entire design would change
354. 11:50so ask those critical questions because
355. 11:52asking them would make your senior or
356. 11:55your interviewer be very impressed on
357. 11:57that you are putting in a lot of thought
358. 11:59you are not taking problem statement as
359. 12:01is but instead you are you are valuing
360. 12:03the problem statement you are thinking
361. 12:05the problem statement of your own and
362. 12:07then devising the best possible solution
363. 12:09out there because independent thinkers
364. 12:11are the most critical assets that every
365. 12:13single org is looking for and every
366. 12:15single org is proud to retain right so
367. 12:18one thing in general that you would
368. 12:20always see working out for any problem
369. 12:22in general and especially in the world
370. 12:23of system design is divide and conquer
371. 12:26so even when you think about either
372. 12:28spiral based approach or
373. 12:31an incremental mvp approach we always
374. 12:33broke that problem into small component
375. 12:35we solved that one thing so in spiral
376. 12:36approach we saw that code and then we
377. 12:38built a new service and then we added a
378. 12:39users and then we added the next one in
379. 12:41incremental based approach what we did
380. 12:43is we took a day zero solution and then
381. 12:45we we basically operated at a very small
382. 12:47level and then we increased the scale to
383. 12:49handle it to the next level and then the
384. 12:51next level and the next level
385. 12:53so
386. 12:54this is very important whenever you are
387. 12:57approaching any system design try to
388. 12:58break it into manageable sub problems
389. 13:00and solve each of those problem very
390. 13:03optimally and once you do once you do
391. 13:05that when all of the sub problems come
392. 13:07and or rather one of all of your
393. 13:09subsystems come together this solves the
394. 13:11entire problem optimally right so divide
395. 13:14and conquer is a strategy that works
396. 13:15amazingly well in system reason and i
397. 13:17would and i would highly encourage you
398. 13:19to look for it
399. 13:21right
400. 13:22okay yeah so that's it for this video
401. 13:24this is a very short one but i hope uh
402. 13:27you
403. 13:27if in case you are confused about how to
404. 13:30approach because that's what the most
405. 13:32overwhelming part is so i've given you
406. 13:34two approaches to go for it this is
407. 13:35something that i personally used
408. 13:37throughout my career
409. 13:38and and even today when someone gives me
410. 13:41a system to design this is exactly how i
411. 13:43approach right so yeah that's it for
412. 13:45this one uh if you guys like this video
413. 13:48give this video a thumbs up if you guys
414. 13:49like the channel give this channel a sub
415. 13:51and i'll see in the next one thanks a
416. 13:53time