[c-nsp] Leaked Video or Not (Linux and Cisco for internal Sales folks)

James Bensley jwbensley at gmail.com
Fri Jun 22 10:29:50 EDT 2018

Hi Saku,

On 22 June 2018 at 10:13, Saku Ytti <saku at ytti.fi> wrote:
> Cool stuff, thanks. Didn't look closely, why do they say research and
> class-room?

This is mostly driven by academia, a bunch of universities
collaborating together. Minimal operator input which is a shame.

>> [1] Right now FGPA NICs + low end CPU are still mega expensive versus
>> a standard Ethernet NIC + DPDK + high end CPU. However I do think it
>> is worth getting on the front foot with these technologies, Juniper
>> have started to support P4 which again, I convince anyone at $dayjob
>> to look into.
> XEON and BRCM aren't really significantly different in BOM, and BRCM
> has orders of magnitude more pps. I'm not saying there is no use-case
> for XEON, what I am saying, if you need significant PPS XEON is very
> expensive.

Yes that was the point I was trying to make too.

> Use-cases I see for XEON forwarding-plane is when you need to do
> something that BRCM and friends simply cannot do, some b it more
> exotic use-cases, then it quickly becomes extremely viable solution.
> And of course if you already have compute which you cannot use for
> anything revenue generating or when your business unit can budget
> compute easily, but not networking. Or when CAPEX simply doesn't
> matter (mostly it doesn't, if you're looking whole company bottom
> line) and from OPEX POV it's just simpler for you to work with XEON
> boxes.

So an "easy" use case might be a BGP free P node for example; a low
touch config box with minimal features and less to go wrong. VPP is
pushing the 1Tbps mark on x86 hardware now. But 10x100Gbps Ethernet
NICs and 64 cores ain't cheaper than a whitebox device right now.


More information about the cisco-nsp mailing list