[OpenIndiana-discuss] OI on modern Intel with P-cores and E-cores?

[Bill Sommerfeld]

On 4/20/26 10:47, Tim Mooney via openindiana-discuss wrote:
> Have you run into any situations that you could identify where you
> think it would be (or would have been) useful to force certain processes
> onto certain cores?  Or has the experience been good enough that it hasn't
> been worth investigating?
I haven't found any such cases so far; the workload it's getting is 
primarily integer ops in kernel networking code.

I saw some performance anomalies which I suspected might be CPU related 
but turned out to be bugs in TCP congestion control resulting in 
bufferbloat and rx ring exhaustion -- nothing CPU related.

I see slight differences in CPU utilization but it looks like both P and 
E cores can forward line-ish rate traffic between two 10G interfaces 
while leaving some cpu idle.   I say "line-ish" because one of the 
interfaces connects to a XGS-PON fiber ONT which provides about 8Gbit/s 
of throughput due to various overheads so I can't get full 10G throughput.

(and, in case anyone is wondering, the system is a Protectli Vault Pro 
VP6650).

					- Bill