Thanks much Ben for the inputs and references.
-Original Message-
From: Benoit Ganne (bganne)
Sent: Friday, April 17, 2020 2:39 PM
To: Prashanth Candade ; vpp-dev@lists.fd.io
Subject: RE: [E] RE: VPP and Core allocation
[EXTERNAL EMAIL] DO NOT CLICK links or attachments unless you
You can reference reassembly code, which does something similar.
e.g. ip4_sv_reass.c
HTH,
Klement
> On 17 Apr 2020, at 18:03, Satya Murthy wrote:
>
> Hi ,
>
> We are having a scenario to support as below and we would like to know what
> we are doing here is correct or not.
>
> 1. Our graph
Hi ,
We are having a scenario to support as below and we would like to know what we
are doing here is correct or not.
1. Our graph node receives a frame with TWO buffers
2. Graph node decides to process FIRST buffer and enqueues the packet to a
next-node.
3. Graph node decides to store the
Hello all,
It's time to begin the preparations for the 20.05 VPP release.
According to the agreed release plan [1], the API freeze will happen
on the May 6, 2020, at 18:00 UTC,
so this is a first gentle reminder.
We are now just slightly more than three weeks before that milestone.
[1]
Running on 2 sibling threads on haswell/bradwell usually gives around 10% more
performance, and on skylake/cascadelake number is 20% or more.
Caveat is that you need to rely on RSS or some other mechanism do equally
distribute traffic between 2 cores….
—
Damjan
> On 17 Apr 2020, at 09:12,
> We are using Skylake CPUs.
> A Followup question is whether
> a) we can run main-core on one core and run any other application on
> sibling core
> b) we can run main-core on one core and run worker thread on sibling core.
We usually only run worker threads on sibling cores - workers should be
Hi Prashanth,
It depends of your CPU. We usually recommend to disable hyperthreading for
Haswell, but to enable it for Skylake and Cascade Lake. When enabled, you
should run 2 workers per physical core, 1 per hyper-thread. You can find more
details here:
