Live data from Hacker News

Meta’s Hyperscale Infrastructure: Overview and Insights

cacm.acm.org

21–30 of 182 posts

Re: Meta’s Hyperscale Infrastructure: Overview and Insights

#21
post #13

Not a word about Thrift, perhaps it was too low level for an infra overview, but I would have expected it to make some technical impact from a global perspective.

I wouldn't be surprised if they have switched to GRPC for improved performance. They mentioned that RPC libraries are centrally maintained in their monorepo; a migration from Thrift to GRPC might have taken less than 6 months.

gRPC and Thrift are comparable in performance and there is actually an opposite trend of switching from gRPC to Thrift in the few places where the former is still used.

Re: Meta’s Hyperscale Infrastructure: Overview and Insights

#22
post #3

Very interesting, in particular the explicit comparisons with hyperscalers. I almost wonder if this is preparation for them launching their own public cloud. Anyone from Meta care to comment?

Offering a Heroku level of deployment abstraction to one organization's own software engineers, while maintaining performance, is an amazing achievement. Developing a cloud product and all the packaged services with account separation, autoscaling, and multiple regions is another massive endeavor.

Think of it as the difference between OCI and AWS.

Meta would be unlikely to launch a public cloud uncompetitive with Amazon's feature set.

Re: Meta’s Hyperscale Infrastructure: Overview and Insights

#23
post #20

Earlier quoted context omitted.

What? It has more than 300M users.

How many of those are bots?

How many users on X (Twitter) or Bluesky are bots? It's reasonable to assume the percentages are the same, given the lack of public information for the major text-based broadcast social networks. X is estimated to have 250M daily active users. Mark Zuckerberg recently stated that Threads' DAUs were 100M. Threads achieving a bit under half the size in such a short time is impressive, especially since Threads still lacks many features that X has had for years.

Re: Meta’s Hyperscale Infrastructure: Overview and Insights

#24

Earlier quoted context omitted.

What? It has more than 300M users.

Today, maybe? I haven't kept up. I'm talking specifically about its launch.

They had 100m account creations within the first 5 days.

Re: Meta’s Hyperscale Infrastructure: Overview and Insights

#25

Earlier quoted context omitted.

What? It has more than 300M users.

Today, maybe? I haven't kept up. I'm talking specifically about its launch.

Threads' launch was intentionally rushed in order to capitalize on the user discontent at the time. Without a large alternative, enthusiasm to migrate to another social network would have waned. Note that Bluesky was still invite-only when Threads launched.

Re: Meta’s Hyperscale Infrastructure: Overview and Insights

#26
post #21

Earlier quoted context omitted.

I wouldn't be surprised if they have switched to GRPC for improved performance. They mentioned that RPC libraries are centrally maintained in their monorepo; a migration from Thrift to GRPC might have taken less than 6 months.

gRPC and Thrift are comparable in performance and there is actually an opposite trend of switching from gRPC to Thrift in the few places where the former is still used.

I participated in a painfully slow migration from Thrift to gRPC. I did not record the performance metrics, but it was internally advertised to be significantly more performant. There are still some Thrift services running at the organization, but most were migrated to gRPC and certainly not migrated back.

Re: Meta’s Hyperscale Infrastructure: Overview and Insights

#27
post #20

Earlier quoted context omitted.

What? It has more than 300M users.

How many of those are bots?

And how many are Instagram users who created a Threads account after being prompted but never actually use it?

Re: Meta’s Hyperscale Infrastructure: Overview and Insights

#28
post #4

> Moreover, once it [Threads] was developed, the infrastructure teams were given only two day’s notice to prepare for its production launch. Most large organizations would take longer than two days just to draft a project plan involving dozens of interdependent teams, let alone execute it. At Meta, however, we quickly established war rooms across distributed sites, bringing together both infrastructure and product te…

They might have launched it quickly, but nobody (relatively speaking) gave a tinkers damn about the end product.

All form and no function.

Re: Meta’s Hyperscale Infrastructure: Overview and Insights

#29
post #4

> Moreover, once it [Threads] was developed, the infrastructure teams were given only two day’s notice to prepare for its production launch. Most large organizations would take longer than two days just to draft a project plan involving dozens of interdependent teams, let alone execute it. At Meta, however, we quickly established war rooms across distributed sites, bringing together both infrastructure and product te…

I’m jealous

Re: Meta’s Hyperscale Infrastructure: Overview and Insights

#30
post #20

Earlier quoted context omitted.

How many of those are bots?

And how many are Instagram users who created a Threads account after being prompted but never actually use it?

Daily Active Users (DAUs) and Monthly Active User counts represent the amount of users that perform activity on the application daily and monthly, respectively.
Post reply on HN