Discussion:
[j-nsp] [c-nsp] Leaked Video or Not (Linux and Cisco for internal Sales folks)
Tails Pipes
2018-06-28 16:29:28 UTC
Permalink
No, things changed there as well. Lookup merchant sillicon, and revise this
post every 6 months. have you heard of Barefoot networks? The days of ASICs
from Cisco are gone and we are glad, we tested the P4 DSL (cisco never got
that right with mantel) on Nexus and its wonderful.

The asics you speak of are no longer important or valuable because people
realized that in many networking planets and galaxies, the asic is reflects
the network design, they are related, and specifically for the data center,
the clos fabric design won, and that does not require fancy asics.
I guess your knowledge is out dated a bit. Cisco itself is using those
merchant sillicon ASICs happily. (lookup Chuck's comments on nexus9000,
best selling cisco switch ever)...guess it is a good switch, because bright
box pushed cisco to do that, and if any one on this list can disagree with
me here, i'm up to that challenge.

What i have discovered recently is that things happen in following way.

Your boss or his boss picks a work culture (no one gets fired for buying
IBM/Cisco), that culture (buying the shiny suits) impacts how you do work,
it makes you select vendors (the ones that sends me to vegas every year)
and not the right network design, you select cisco and you are stuck there
for life, because once they tell you how things should work (aka :
certificates), things are worse, now every time you make a new network
purchase (afraid of new CLI ), you will not be able to look the other way
because you just dont know any thing else (and loosing your certificate
value).

I wish the culture would change to, no one got fired for buying closed but
didnt get promoted either. change requires boldness.

https://toolr.io/2018/06/18/stop-abusing-the-word-open/
Tails Pipes
Sent: Friday, June 22, 2018 3:00 PM
can you easily answer this question ? why packets are not pushed in
linux ?
is it because of big switch, cumulus, pica8 ?
can you push packets in linux without writing code to do that ? who is
writing
that code ?
this is supposedly a community effort, something that older generations
dont understand.
If pure linux as NOS has some legs it'll fly regardless of cisco blessing,
don't worry no single company owns the whole industry.
Also we can argue that this is only about the OS but in reality it's also
the quality of apps running on top and the quality of the underlying HW
that plays a major role.
The quality of BGP app for instance, or the ability of the forwarding ASIC
to deliver the stated pps rate even if multiple features are enabled or
protect high priority traffic even if ASIC is overloaded.
< I am sick of having to learn all the cisco specific terms to all sorts
of different boxes and technologies
I'd recommend you read all the cisco books on networking to get yourself
educated on the topic and to get the difference between SW and HW
forwarding ( -on why packets are not routed in linux)
And while on that I suggest you read all Stanford university lectures on
how routers work too, it'll help you understand why Cisco and Juniper ASICs
are so much more expensive than white-box ASICs.
adam
netconsultings.com
_______________________________________________
juniper-nsp mailing list juniper-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/juniper-nsp
a***@netconsultings.com
2018-06-29 12:49:46 UTC
Permalink
Sent: Thursday, June 28, 2018 5:29 PM
No, things changed there as well. Lookup merchant sillicon, and revise this
post every 6 months.
Nah
have you heard of Barefoot networks?
Yes I have heard of barefoot, but have you heard of barefoot queuing architecture or pre-classifier granularity and queuing strategy or packet size based pps performance graph, well yeah neither did I.
The days of
ASICs from Cisco are gone and we are glad, we tested the P4 DSL (cisco never
got that right with mantel) on Nexus and its wonderful.
The asics you speak of are no longer important or valuable because people
realized that in many networking planets and galaxies, the asic is reflects the
network design, they are related, and specifically for the data center, the clos
fabric design won, and that does not require fancy asics.
Well there are use cases for fancy asics and use cases for simple asics.
I guess your knowledge is out dated a bit. Cisco itself is using those merchant
sillicon ASICs happily.
Yup and those are dirt cheap compared to their home grown asics , this is because for a given pps rate they have lot less smarts.
(lookup Chuck's comments on nexus9000, best selling
cisco switch ever)...guess it is a good switch, because bright box pushed cisco
to do that, and if any one on this list can disagree with me here, i'm up to that
challenge.
What i have discovered recently is that things happen in following way.
Your boss or his boss picks a work culture (no one gets fired for buying
IBM/Cisco), that culture (buying the shiny suits) impacts how you do work, it
makes you select vendors (the ones that sends me to vegas every year) and
not the right network design, you select cisco and you are stuck there for life,
because once they tell you how things should work (aka : certificates), things
are worse, now every time you make a new network purchase (afraid of new
CLI ), you will not be able to look the other way because you just dont know
any thing else (and loosing your certificate value).
Well I guess that could be a story of some of the smaller shops that can't afford investing in R&D and thus rely on well-trodden paths, but certainly not my problem.
I wish the culture would change to, no one got fired for buying closed but
didnt get promoted either. change requires boldness.
https://toolr.io/2018/06/18/stop-abusing-the-word-open/
I understand your passion for open sw and open hw architecture, but routing high pps rates on x86 will always be impractical compared to specialized asics.
Yes going x86 certainly makes sense for low pps use cases - a uCPE is a good example of such a use case.
And as far as the high pps rate use cases goes, reading other posts in this thread, it seems that the current state of the art with regards to any linux OS driving any white-box ASICs is not quite ready for primetime yet.
A viable alternative might be the FPGA NICs -but seems still not financially attractive.
Just wondering what's the latest on the GPU for packet forwarding front (or is that deemed legacy now)?


adam

netconsultings.com
::carrier-class solutions for the telecommunications industry::

_______________________________________________
juniper-nsp mailing list juniper-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/juniper-nsp
Gert Doering
2018-06-29 12:55:55 UTC
Permalink
Hi,
Post by a***@netconsultings.com
Just wondering what's the latest on the GPU for packet forwarding front (or is that deemed legacy now)?
Last I've heard is that pixel shaders do not map really nicely to the
work needed for packet forwarding - so it works, but the performance gain
is not what you'd expect to see.

gert
--
"If was one thing all people took for granted, was conviction that if you
feed honest figures into a computer, honest figures come out. Never doubted
it myself till I met a computer with a sense of humor."
Robert A. Heinlein, The Moon is a Harsh Mistress

Gert Doering - Munich, Germany ***@greenie.muc.de
James Bensley
2018-06-29 15:57:22 UTC
Permalink
Post by Gert Doering
Hi,
Post by a***@netconsultings.com
Just wondering what's the latest on the GPU for packet forwarding front (or is that deemed legacy now)?
Last I've heard is that pixel shaders do not map really nicely to the
work needed for packet forwarding - so it works, but the performance gain
is not what you'd expect to see.
Which is to be expected right? Typical GPU instruction sets and ALUs
are great for floating point operations which we don't need for packet
processing. Packets typically need low complexity tasks performed at
high rates. Various high end DC switches like Nexus boxes use GDDR5
RAM just as a graphics card would but the processing is done by an
ASIC, which makes sense to me, that is not the place for general
purpose x86 compute chips. This is a specific task.

Cheers,
James.
_______________________________________________
juniper-nsp mailing list juniper-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/juniper-nsp
Jason Healy
2018-06-30 02:07:19 UTC
Permalink
Post by a***@netconsultings.com
Just wondering what's the latest on the GPU for packet forwarding front (or is that deemed legacy now)?
Waiting for the bare-metal version of this to land (you can test it on AWS right now):

https://www.netgate.com/products/tnsr/

https://www.netgate.com/blog/the-behemoth-router-is-here.html


Jason
_______________________________________________
juniper-nsp mailing list juniper-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/juniper-nsp
a***@netconsultings.com
2018-07-08 22:03:08 UTC
Permalink
Sent: Saturday, July 07, 2018 3:58 PM
https://github.com/OAI/OpenAPI-Specification/issues/568
can we change the topic of the thread to an informative one, instead of a
leaked video or not, to why exactly do network engineers are often
confused by the abusive marketing all over the place of what is open and
what is not and other computing terms.
I guess this is happening in networking more often than other domains
because networking people didnt get a chance in their career to learn about
the world of computing, their heads were somewhere else, learning about
complex networking protocols and not the common computing interfaces,
the open source world, existing frameworks and paradigms, this video helps
https://vimeo.com/262190505https://vimeo.com/262190505
has anyone here seen list of topics that network engineers usually miss on
their journey ? i know they never get exposed to software development
and engineering in general, databases, web technologies, operating system
fundamentals.
Well I guess if you stick around in networking for long time you kind of get exposed to some of these to a certain level on a day job, some of it was covered in school in various levels of detail, and to some of these concepts we (networkers) get a specific very narrow filed exposure I'd say, like in your example of databases -well various protocol tables are good examples of decentralized distributed databases, then some Network OS-es are good examples of distributed operating systems. So I guess it then just boils down to the willingness of and individual to understand these concepts on an ever more fundamental level -with every next interaction with these. Maybe it draws one more towards the software development side or perhaps more towards the somewhat holistic understanding of the networking discipline through graph theory and complex adaptive systems.


adam

netconsultings.com
::carrier-class solutions for the telecommunications industry::


_______________________________________________
juniper-nsp mailing list juniper-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/juniper-nsp

Loading...