←back to thread

It's time to replace TCP in the datacenter (2023)

(arxiv.org)
188 points ilove_banh_mi | 4 comments | 18 Nov 24 01:42 UTC | HN request time: 0.83s | source
1. GoblinSlayer ◴[18 Nov 24 07:50 UTC] No.42170575[source]▶
> For many years, RDMA NICs could cache the state for only a few hundred connections; if the number of active connections exceeded the cache size, information had to be shuffled between host memory and the NIC, with a considerable loss in performance.

A massively parallel task? Sounds like something doable with GPGPU.

replies(2): >>42171362 #>>42171369 #
2. mafuy ◴[18 Nov 24 11:02 UTC] No.42171362[source]▶
This has nothing to do with computing, it is about memory access.
3. dboreham ◴[18 Nov 24 11:04 UTC] No.42171369[source]▶
Now the information has to be shuffled between the NIC and host memory and the GPU.
replies(1): >>42186896 #
4. GoblinSlayer ◴[19 Nov 24 18:58 UTC] No.42186896[source]▶
Just plug ethernet cable into the graphics card, then you need to shuffle memory between GPGPU and the wire.