Anyone able to help with a quick benchmark of 50k messages per second in pulsar? Would 256gb memory and 24 core cpu x3 (1 server per pulsar component) and nvme drives be able to handle this theoretically? For scaling horizontally scenario, this only really help with increased topic count?

You’d be fine with way less cpu/mem than that 😄

I should be more specific with the mem/cpu allocation per pulsar resource, let me also get that information.

We recently did 12Gb/s ingestion benchmark per node (broker, bk, zk all colocated) with Azure

Standard_L8s_v2

, with (E=3, Qw=2, Qa=2) on 1KB messages. Pulsar is very performant.

More than 1 GB per second on single node ?!

yes., network was the bottleneck for us for

L8s_v2

. NVME didn't even reach its half-IOPS