really high quality best-in-class system design  answer so that you can use it to prepare for  

your interviews and use it to know exactly what  standard you should be aiming for with us they  

give us the answers the candidate today I have  a former engineering manager at Google it's Mark  

Mark how are you doing. I'm doing great Tom thanks  for having me I'm looking forward to doing this  

oh thanks for thanks for coming on it's  great to have you before we get started  

just want to quickly give people an idea of  your background as a as an engineering manager  

yeah I was an engineering manager at Google for  13 years and worked on several large-scale systems  

with lots of great engineers and so I I hope  I've learned a little bit from them and it'll  

be interesting to see me on the other side of the  fence of uh of uh doing the being the interview  

candidate here so I'm looking forward to it I'm  sure you do a great job yeah and Mark also is a  

coach on our platform so if you need some expert  feedback ahead of your interviewer Google Facebook  

anywhere any tech company then do check him out  on our platform okay uh if you're ready Mark  

let's crack straight into it the question I want  to ask you today is how would you design Spotify

okay

uh so yeah so uh Spotify the probably one  of the most popular music streaming services  

out there I guess uh so let's let me let me  think about that just for a moment [Music]

um because there's lots there's a lot of aspects  of Spotify here let me see if I can actually maybe  

I'll just take some notes here so that I can kind  of I think I think through this if that's okay and  

and can I get some uh make some assumptions so  I have my notes I'll just put this right in here  

and then if it's okay I'll just use this for  notes as well as sort of the design diagrams  

and things like that later on yeah sounds good  so yeah so let's yeah so let's talk about just  

some different sort of Spotify okay so I for  Spotify I can I can think of things I mean  

obviously there's songs music that's the main  thing uh there's you know you have playlists  

I mean there's users there's uh artists uh I  guess Spotify even has things like podcasts  

that you know you can listen to and things like  that so there's a lot to it and so in terms of the  

system design for today uh is there is there like  can we can we constrain this a little bit just  

so that we can actually get through a design what  would you what would we would be good to focus on  

here yeah let's constrain it let's limit ourselves  to let's look at finding and playing music  

okay okay so let me uh so use cases finding uh  and playing music I'm just gonna write it down  

like verbatim here great yeah I think enough  to talk about I think so yeah okay good so uh  

if I if I think about that let me let me try  and ask a little bit or think a little bit about  

the use cases a little bit more so my typical use  case then as a user if I'm signed up to Spotify  

and I have an app on a phone uh I am going to  browse some music maybe I'll find a particular  

search for a particular song but but then the  most core thing is that I play a song and it's  

you know coming back through my phone maybe my  car or whatever wherever I might be headphones uh  

to listen to it with big broad questions like this  you need to constrain the problem make it solvable  

in the space of an hour and Mark did this well  establishing the use cases we wanted to focus on  

if you're asked to design an already existing  system in this case Spotify share what you  

know about it with the interviewer they can  correct you or fill you in if you have any  

knowledge gaps so if I think about other  if I think about metrics here let's think  

about size let's let's do some quick quick  drilling down into into some numbers here  

and pardon me while I get this going here so  in numbers so tell me about how many users  

you're thinking about here for for this  design yeah let's save a billion users

okay one billion users okay what about number  of songs yeah I think I saw I think I saw it  

somewhere that Spotify has about 80 million  songs but let's say we need to have capacity  

for 100 million on there okay all right so 100  million songs a billion users and you know we're  

we're going to focus on the song piece of it so  but let me do a little bit of uh of math here  

or a little bit of thinking and you can I can  I'll just you know double check with you to  

make sure that this makes sense but for from what  I know think about a typical mp3 audio is going  

to be about for a typical song is is about five  megabytes and I mean I think that depends on oh  

the encoding of the song how long the song is and  things like that but uh if that if that if that's  

okay we can just kind of make that assumption  is that fair yeah that sounds good to me

[Music]

yeah I mean I think you can encode  these things and you know really low  

quality like 96 kilobits per second all  the way up to 320 kilobits but right in  

the middle 12860 is about maybe about  that so all right so let's assume that  

and so then let's do let me just extrapolate from  this so if you've got five megabytes per song  

you've got 100 million songs so I think uh that  translates to I'm trying to do my math here so  

100 million would be uh going to a thousand times  that would be a 100 billion and then 100 trillion  

would be a million times times five would be  500 trillion so we're talking about uh total  

audio is something like 500 terabytes which is I  guess it's the same thing is half a petabyte of  

data and so that's I'm making that assumption and  so then depending on how you replicate this data  

because you typically want to have these songs in  multiple you know have multiple copies of these  

things so they don't get lost Etc and if some  replica is down so you know maybe you do let's  

say oops sorry my keyboard here 3x replication so  that would mean like one and a half petabytes of  

of raw audio data and then and then each song  in terms of the metadata meaning the you know  

the the song title and things like that and  and artists and all these things the metadata  

is probably not very big it might be I don't know  you know 100 bytes per song or something like that  

per song of metadata and so you might wind up with  okay so let me do the you have it and that's uh uh  

billion 10 billion so that's only 10 gigabytes  of of song metadata it's not really very much  

and even if you say it's it's a kilobyte then  that's 100 gigabytes so it's not very much and  

let me just double check my math here 100  million times 100 really 100 bytes yeah I  

think that would be 10 gigabytes so even if we  said if we round it up dramatically and said  

100 gigabytes that's just not a ton of data  now in terms of users uh user data you might  

have you know again like a maybe you have a  kilobyte oops kilobyte per user metadata I  

guess it's all metadata and so uh that might be  [Music] times a billion users so that would be a  

terabyte of data approximately so just to kind  of so this is just me doing some quick metrics  

quick calculations to get a rough back of  the envelope of how big these things are  

uh so does that make sense so far does  that seem yeah that will make sense so far  

before starting your high level design as marketed  get some metrics to help you identify any higher  

level decisions that might be influenced by the  scale of the system Mark asked the right questions  

about the number of users and number of songs  I made some rough calculations without getting  

bogged down in the numbers what Mark didn't  calculate was the traffic how many songs are  

streamed per day or per second even that would  have helped him figure out the scaling of the  

web servers how many he'd need and how much  bandwidth he'd need however since this didn't  

impact his high level design it was okay not  to go into the details of scaling at this point  

okay so let me let me maybe think about  some basic components here and I'm going  

to draw them but I I'll get into them  maybe a little bit more in detail later

okay

so some basic components that I would see so let  me let me start with something that is my my best  

attempt at a mobile phone here so you cut your  Spotify app and make this smaller so it fits into  

the thing so you've got your app which is on your  phone typically and I'm gonna assume here that  

we're talking about the phone app because that's  very common uh and so you've got your your phone  

the phone is going to be talking to uh an  application server a Spotify application  

server somewhere or web server I  guess and so let's draw some let  

me just put a like over here just say  there's a Spotify web server uh and uh  

and there's not going to just be one there's going  to be lots of them and assume that that's a lot  

and then of course just because this is a standard  thing that you do when you have applications uh  

you know in the cloud here this is I guess part of  the cloud you've got a load balancer and so let me  

draw let's assume that we've got an arrow here  that's talking so Spotify app's talking through  

the load balancer ultimately to these to these web  servers and so those are the those are some some  

sort of key components here let me think about  this a little bit further yeah I think that's  

good so far okay and then of course the most  important thing or one of the most important  

things here let me see if I can find a I know that  there's a shape here here we go that's a that's a  

good shape I'm just going to say call this uh the  database and let me make this a little bit more  

uh let's see that's what I wanted okay so there's  database and so the web servers are going to be  

talking to the database and getting stuff from  basically reading information writing information  

to and from the data reading and writing  information to and from the database  

so those are kind of the very high level  components in system design interviews good dual  

bandwidth communication is important that's to say  you need to communicate well via both drawing and  

speaking you should practice this thoroughly  before your interview Mark did this well he  

started drawing the components quite early on in  his answer and this allowed the interviewer to  

follow along easily it's okay to have simpler and  fewer components at this point in the interview  

you can then break it out later in your answer as  Mark did when he splits his database in two and  

I probably am thinking about this I think this  is obviously oversimplified and I'm not going to  

redesign Spotify uh you know the existing full  application as exists but I do think I can get  

a little bit more detailed here by I think I'm  thinking about the data again because it's it's  

the data is important so I'm thinking about the  metadata and the user data and the audio data

and they're different very different types of data  so I'm going to actually I'm going to split this  

database here into two things and I'll explain  that or I'll try to explain them just a little  

bit come on there we go make this just a little  bit more uh so I'm going to split this up and say  

that I'm gonna there's a  like a song audio database

and then there's going to be like a  metadata database so this is users uh songs  

uh what else I mean are ultimately they're  probably be artists and Etc things like that  

but we're focusing on the songs I think in the  in the music and so on this is also a database  

of course song audio database and metadata and  so I think let me make this a little bit more  

uh this Spotify web server is actually going to  talk to this to The Meta oops the metadata and  

uh and the audio database it's actually going  to be talking to two databases here yeah that's  

interesting can you can you can you go into a bit  more detail about why you'd split them into two

yeah yeah I'm kind of making that uh just assuming  that or making that yeah so okay let me let me  

to try and I wanted I want to  answer let me try and think of  

this in terms of the technologies that  I might use I might use something like  

Amazon S3 here and I might use Amazon RDS and  so now let me explain what I mean by that so  

the type of data that the actual MP3 files those  I think of as immutable data they're just Blobs of  

data they're files they're five megabyte roughly  on you know on average files and they're not going  

to change they're just being streamed they're  being you know stored and streamed essentially  

and so a blob database which is what S3 is lends  itself really well towards that and it could be  

something like Google drive or there's other  you know technologies that are sort of just  

document like blob storage uh systems but Amazon  S3 lends itself really well to that and it scales  

uh greatly you can just add more and more and more  and more and it scales uh linearly which is great  

and and so and then you can connect it up to to  be able to stream the data and things like that so  

that type of data and the access patterns for that  which is really mostly read you're just streaming  

this data you're never going back and forth and  writing to it that would be why I would want to  

put the songs in that kind of a database now S3  is great for that works really well but the data  

the metadata which is the the users and their  information and the songs and their information  

and uh maybe they get updated maybe you're  searching across them and having to do queries  

to find songs for a particular genre or artists or  things like that or you're trying to update your  

users like if I'm playing us playing a song and  I want to remember where I left off so that when  

I continue playing it it remembers that I'm going  to be modifying the data quite a bit or going back  

and forth between that database and doing these  queries and you can't really do queries over S3  

but you can over like a relational database so  MySQL would be another option uh I just chose  

RDS here really Amazon's relational database as  a as a because I'm going with the AWS family here  

but yeah so uh let me let me try and try and just  write something here a little bit more in terms of