OpenSNB Core - a general discussion

[sfx2000]

Would the Silverstone case imped wireless signal? Would it need modification to add external antennas?

The anti-spam plugin is holding a post where I do go into more detail - but if putting wifi in the box, one would want to breakout the drill bits and mount the antennae in a good location - alternate would be to not put wifi in the box, and run dedicated AP's

 

[System Error Message]

external antennas are always a good option as it lets you use the ones you want.

GPU based router http://shader.kaist.edu/packetshader/

using the intel IGP on an i7 IGP with 16 shaders has enough compute power to forward many Gb/s of traffic as in comparison to a GTX 480 each intel IGP shader or GPU core can do more than a single shader or GPU core of the nvidia GPU although it doesnt mean the intel i7 IGP is faster than a GTX 480 in routing or compute in total. Aside from intel, AMD APUs have a good IGP. I think the maximum forwarding capacity of the IGP is 1/4 of the RAM bandwidth unless only the cache or on die memory are only used.

Nvidia has been boasting of their GPUs being independent of their CPU but only their kepler GPUs and above can initiate tasks without CPU intervention and only their tesla GPUs can communicate with other hardware components on PCIe without CPU intervention but this feature is locked for non teslas. In short only Tesla GPUs that support a modern CUDA version can do this but all consumer variants have this feature locked out.

 

[sfx2000]

Going to kick this one back up to the top - and perhaps start another discussion - I've had a few conversations with others that might lead to something very interesting...

 

[chadster766]

I suggest using a common high powered and inexpensive hardware platform for initial development; the Linksys WRT1x00AC\S Series.

Also McDebian is a full Debian system implementation for these routers. Debian is the basis of many new system platforms. McDebian is completely stable running long term Linux kernel 4.4.16 release. Debian can be ported to almost any hardware platform that supports the Linux kernel with the right kernel build options.

https://github.com/Chadster766/McDebian
https://github.com/Chadster766/McDebian/wiki

 

[sfx2000]

I suggest using a common high powered and inexpensive hardware platform for initial development; the Linksys WRT1x00AC\S Series.

I've built an OpenCloud Stack on RPi3 with Ubuntu 16.04LTS Mate - DevStack is running clean...

The OpenFlow/SDN/NFV side right now is running OpenWRT as a base, with OpenFlow integrated on top...

Looking now into VPNaaS - so Neutron can do IPSec/L2TP now, and kicking OpenVPN to the curb as it doesn't integrate well...

Looking at rolling this all into one box with the ADI 2440 board - this is an Intel Silvermont box, with four independent Intel NIC's - which then we can use OpenFlow to configure them as virtual Switches/Routers/etc...

@chadster766 - Way ahead of you...

 

[sfx2000]

Need to get a good/clean recipe for building LXC into the RPi builds, which might work, or just move to the 2440, and be done with it, as the debian and ubuntu builds already have lxc there - jessie is good to go, so is xenial..

Why is lxc important? Containers...

Apps - file services, domain name services, vpn, orchestration, etc...

I'm not porting debian to a router, that's be done, and it's an enabler perhaps...

I'm proposing a long term stack that is HW independent of any vendor that will fit SNB needs...

 

[System Error Message]

Regarding hardware you can use examples but avoid suggesting a platform as a baseline. This is because as time goes the requirements also grow and the featureset increases.

What is important however is in reducing waste. This means routers that can be upgraded (ram, storage, CPU?). Modules for modem, wifi and perhaps new features (like 2.5 and 5Gb/s)?
The other important factor is power use as well.

 

[sfx2000]

Anyways - there's another forum member that will likely start a thread to kick off this project. We've been chatting offline - and we're agreed that while some of this might be over the head of many, there's enough critical mass to make this happen...

@chadster766 - you're more than welcome to join the effort - this is my public reach out - others will be contacted privately - we're going to take an agile process management direction, and as such, I'm the Product Owner, and we'll sort out the user stories, and build the scrum team around it...

We're looking at Storage, Network, and Compute functions as it applies to the SNB Audience, and the target goal here is to define a specification that any vendor can meet...

 

[sfx2000]

Regarding hardware you can use examples but avoid suggesting a platform as a baseline. This is because as time goes the requirements also grow and the featureset increases.

Agreed - and needs differ depending on application - some might want routing, some may consider storage as a priority, and others see STB functionality as a must.

And we can break these down into components - the big clouds are doing this already - I'm looking to see if we can work this into a Product that someone can deploy into their home, perhaps multiple boxes (e.g. the "router" the "NAS" the "STB") into a holistic approach where one can manage all from a single dashboard...

I want to build a box where we can stash the lego's into - the blocks by themselves have utility, but if you want to build something - put the blocks in order...

To this end - it's architecture, it's API's, it's what makes sense within those constraints... it's about making sense of abstractions in how our networks do what they do... and how they meet our needs.

In an Agile Development Process - that vision turns into user stories - and having that vision... it's not a political thing, but in Agile that makes the vision the product owner...

 

[sfx2000]

One has to grok where SNB is in the small network business...

We've got a lot of attention from vendors - and it's our chance to tell them what we want - actually perhaps demand, if they want our patronage...

So the spec needs to be OPEN, it needs to be platform agnostic, it needs to be flexible to meet our needs...

The work is already done - while it might seem to be a daunting task, it really isn't... it's writing a spec that scales to the needs of the community here.

We just need to tell them what we want... and give the vendors some room to improve...

 

[chadster766]

What do you think of requesting network switch or network adapters compatible with netmap like:

e1000
e1000e
forcedeth
igb
ixgbe
r8169
virtio

This would allow for excellent throughput with IDS\IPS software like DAQ and SNORT:

Netmap Module
=============

The netmap project is a framework for very high speed packet I/O. It is
available on both FreeBSD and Linux with varying amounts of preparatory
setup required. Specific notes for each follow.

./snort --daq netmap -i <device>
[--daq-var debug]