I often see these kind of comments and find them extremely myopic.
Sure you can create a twitter clone all by yourself, but good luck scaling it to ~150k tweets per second writes[1]. All the while ensuring they're fanned out to all the respective followers some of whose tweets may need to be fanned out to tens of millions of users in a matter of seconds.
You'll need 30 engineers just to build and maintain the datastores, caching mechanisms, and service deployment infrastructure. You then have the actual services, real time streaming platforms, streaming applications, batch infra (hadoop), analytics pipelines, notifications, security, anti-spam, recommendation/machine learning pipelines, ads infrastructure etc to name a few.
Can't find it now, but there should be a circa 2011 architecture of twitter in one of the blogs somewhere. I recommend you have a look at it.
If you're comparing it with WhatsApp, note that WhatsApp messages are mostly 1->1 and it's easier to shard the servers by recipient. Also, not having to bother persisting all the messages forever simplifies the system by a lot.
I think you're making a few assertions that would be more honestly served if you conceded as assumptions.
Twitter validated it's own employee count with an absurd overhead.
It has concerns of scale, performance and user experience. So does GitHub, Shopify, Pandora, Spotify, Snapchat, LinkedIn, MailChimp, WordPress.com, etc, and yet, manage will much less resources and overhead. Oh yea, and fucking Netflix - which - as far as I can tell - has much more and difficult problems to solve at a much faster pace. They employ 4,700.
As of this comment, Twitter employs 3,583 people.
What they needed to do was hire 100 talented people, allow third-party clients, try and occasionally acknowledge the developers who implement their platform, and stopped with the asinine "we're trying so hard guys" blog posts. We get it. You're having a hard time and you want everyone off of your grass. That's fine, but don't come asking for sugar at my house 'cause you ain't Beyonce.
Netflix has less "hard problems" in terms of operations than Twitter.
With Twitter everyone's data is interconnected. With Netflix, you could theoretically have each server just be completed isolated from each other (modulo account management).
Netflix has operational problems to solve in terms of bandwidth, though. There's definitely major difficulties there, but they're of a different variety to the high coupling Twitter deals with
I'll say what I've said in another thread on this matter, which is, no one outside of this bubble buys this shit. There are global supply-chain networks that move millions of products, and they don't employ as much as a micro-blogging platform. It's hubris to think we're so special.
100% agreed that running a network on the scale of Twitter, with all the strong interconnections, is extremely hard. Probably the hardest, moreso than even things like Facebook.
I do wonder what the thousands of engineers are doing. I imagine there's a lot of fire-fighting, but is it only that?
It could be the case that twitter is constantly falling over, still.
That's the thing, any time we raise eyebrows at the size of Twitter's dev team we're met with "well YOU try to operate a business at the scale of Twitter then!". I am, like you, genuinely curious what kind of work these people are doing all day. If they're prototyping new products, optimising poorly performing DB queries and API calls, building out frameworks (like Bootstrap), maintaining some internal admin UI etc.
I've read bits and pieces, heard rumours and it'd just be nice to see what's going on. Unless I see something like this I'm just gonna think "BOY that's a lot of developers" whenever I remember Twitter's headcount ...
I got an email from a recruiter last year asking if I wanted to come build the future of live video at Twitter. I think it's plausible many of them are working on new products that end up abandoned due to constant direction changes.
They definitely attribute a lot of their early success to this choice. It seems reasonable to me that choosing a tool developed to create modern real time communications software is a pretty good idea ;)
You'll need 30 engineers just to build and maintain the datastores, caching mechanisms, and service deployment infrastructure. You then have the actual services, real time streaming platforms, streaming applications, batch infra (hadoop), analytics pipelines, notifications, security, anti-spam, recommendation/machine learning pipelines, ads infrastructure etc to name a few. Can't find it now, but there should be a circa 2011 architecture of twitter in one of the blogs somewhere. I recommend you have a look at it.
If you're comparing it with WhatsApp, note that WhatsApp messages are mostly 1->1 and it's easier to shard the servers by recipient. Also, not having to bother persisting all the messages forever simplifies the system by a lot.
[1] https://blog.twitter.com/engineering/en_us/a/2013/new-tweets...