News for hackers. Stuff that matters.
Re:Cache made consistent: Meta’s cache invalidation solution by uvdn7 on Wednesday June 08, 2022 @03:33PM (#31672807)
Accepted. Just to be clear, it’s not a link bait as far as I understand it for two reasons 1. I am dead serious about making cache invalidation a simpler problem 2. “Cache invalidation might no longer be a hard problem in Computer Science” is the subtitle of the corresponding systems @scale talk I just gave.

I respect the guidelines and I am OK with the rename. I just want to clarify that I don’t think the old title is misleading.

Re:Cache made consistent: Meta’s cache invalidation solution by PKop on Wednesday June 08, 2022 @03:37PM (#31672864)
It's still massively click-bait until such time as the claim isn't pure speculation. Also, casually optimistic claims about complex and hard problems are the essence of click-bait. There isn't much unique about your instance of it
Re:Cache made consistent: Meta’s cache invalidation solution by uvdn7 on Wednesday June 08, 2022 @03:56PM (#31673125)
> until such time as the claim isn't pure speculation

Which I don't think it is.

> casually optimistic claims about complex and hard problems are the essence of click-bait

A group of people at Facebook worked on this for many years. TAO and Memcache serves quadrillions of queries a day. I wanted to make this claim – cache invalidation might no longer be a hard problem in Computer Science – years ago; but I didn't because I don't want to jump the gun. We baked the solution for years. There's nothing casual about this.

Re:Cache made consistent: Meta’s cache invalidation solution by pvg on Wednesday June 08, 2022 @04:06PM (#31673260)
Just about all papers and write-ups that describe big advances on important problems aren't titled 'Big advance in Problem X' or 'X now an ex-problem'.
Re:Cache made consistent: Meta’s cache invalidation solution by uvdn7 on Wednesday June 08, 2022 @04:14PM (#31673354)
This is a fair point. Lesson learned. Thank you :D
Re:Cache made consistent: Meta’s cache invalidation solution by o_____________o on Wednesday June 08, 2022 @05:24PM (#31674438)
Small comment of appreciation for your attitude and quick adjustment, nice to see on the internet!
Re:Cache made consistent: Meta’s cache invalidation solution by tomcam on Wednesday June 08, 2022 @06:35PM (#31675298)
Agreed. Gracefully handled. I happened to agree with the arguments, but treasure civil discourse.
Re:Cache made consistent: Meta’s cache invalidation solution by selcuka on Wednesday June 08, 2022 @07:07PM (#31675599)
Don't be too hard on yourself, naming submissions is still a hard problem.
Re:Cache made consistent: Meta’s cache invalidation solution by reitzensteinm on Wednesday June 08, 2022 @07:08PM (#31675605)
"Finally, we are building a high-level consistency API for distributed systems — think of C++’s std::memory_order, but for distributed systems."

Will you be writing this up as well? It's very ambitious!

Re:Cache made consistent: Meta’s cache invalidation solution by uvdn7 on Wednesday June 08, 2022 @07:27PM (#31675798)
I am glad you caught that! I am personally very very excited about this. Because if you think about it, distributed systems and cache coherence problems on a multi-core system are essentially the same thing.

If C++'s memory model can abstract away multiple cachelines, why can't we steal the idea and abstract away distributed system details? The single keyword of distributed systems is "order". And that's exactly what `std::memory_order` captures, which is not a coincidence.

You can DM me and I would love to chat. I am @uvdn7 on twitter.