[peoplefromibiza]

WhatsApp did it for a billion daily active users (Twitter has/had more or less 200 million active users/day) with 50 engineers.

Using Erlang.

Tiwtter uses Scala that should be much more scalabale, considering the staggering amount of money spent on the JVM compared to the BEAM.

[Wherecombinator]

Not the same. A single tweet may need to be distributed to 50 million followers. I’m not a distributed database guy, but that to to me looks like a very different problem than what WhatsApp faces.

[peoplefromibiza]

> Not the same. A single tweet may need to be distributed to 50 million followers

Of course not the same, but the question is, how much harder?

10 times harder?

100 times harder?

Given 1/5 of the users and probably less than 1/200 of the traffic (*) it should be feasible with less than 7,500 employees.

I don't believe WhatsApp employed the best 50 engineers in the World, I bet many of them worked at Twitter too.

(*) 100 billion messages sent via WhatsApp every day VS 500 million tweets per day

[kaba0]

Let’s not compare all employees to devs. Twitter has (had) absolutely business critical other teams, they are a very human-facing company. Does whatsapp have to make decisions whether to ban/unban a former US president? Also, I think Twitter gets a much more uneven usage - their daily usage is lower, but peaks can be similar (~1 billion user).

Also, graph algorithm’s likely don’t scale linearly, 10 times more edges may have much much more resource usage.

[peoplefromibiza]

> absolutely business critical other teams

So you're saying WhatsApp had no critical other teams, but Facebook bought them anyway for a record breaking sum of money at the time, because they looked nice on picture?

[kaba0]

I don’t know, but I’m fairly sure that the user base is the most valuable part of a social network/messenger platform, not the tech stack, or the staff.

[prirun]

If Twitter's Grand Pubah sends one tweet, it goes to 118M people. I can't see how many times he tweets per day, but it's way over 10, and just that would generate 1.2B messages.

When Elon tweets, I'm guessing Twitter doesn't go do 118M write transactions to all his followers. But what it would have to do is flag all 118M accounts as "something as changed". Then the next time one of his followers is on Twitter, which could be a month from now, that flag means "uh oh, better go see what changed", and to do that, all of the follower's "following" accounts have to be accessed to find the change, if there is only 1 change flag per Twitter account.

If Twitter keeps a change flag per "Following" user, that makes finding changes easier, but also means lots more flags to keep updated.

Or, you could go "stateless" and not use flags, but then it seems like you'd be doing a lot of work for users who may not see it for days or months.

Way different problem than 1-on-1 messaging.

[peoplefromibiza]

> If Twitter's Grand Pubah sends one tweet, it goes to 118M people

if my mom sends a message to the family group it goes to 50 people.

How many billion WhatsApp groups are there?

with the same constraints and problems Twitter has to face (offline users, notifications, keep users messages until they are online again, etc.)

[kaba0]

How many people do you talk with on WhatsApp? Also, how many of those live in very very different geographical areas? I feel twitter’s graph has much more edges, and the nodes (users) can’t be partitioned as well.

Also, Twitter does real-time data analysis, which is another of their money sources, and takes a huge amount of computing.