[j-nsp] Juniper QFX10002 failures

[Saku Ytti]

On Sat, 8 Nov 2025 at 16:29, Tom Beecher <beecher at beecher.cc> wrote:

> Yeah the actual errors observed would be more helpful.
> The QFX line has always had problems with HMC. It's a little bit of bad luck maybe ; had they been able to wait for HBM could have been a different story.

I've understood HMC was better suited for IP lookup purposes
technically. I think in PTX after they moved from HMC to HBM they also
stopped doing lookup against HBM, and only used HBM for the delay
buffer. I think also Cisco Silicon One originally intended to do
lookups on HBM after the first 1M or so FIB, but somehow by the time
they shipped it, no lookup against off-chip was materialised.

-- 
  ++ytti