[Cerowrt-devel] LCA 2018 talk available

[Cerowrt-devel] LCA 2018 talk available

[Joel Wirāmu Pauling]

[-- Attachment #1: Type: text/plain, Size: 469 bytes --]

Kia Ora (Hi in Māori).

Today I delivered my talk on 10Gbit(+) in the home at Linuxconf
Australasia. Some specific shout outs to those on the list who helped form
some of the content and especially for the continued efforts with FLENT
which I have been making extensive use of both professionally and privately.

Hopefully this is of some interest and use to people on the list.

https://github.com/aenertia/lca2018-talk/tree/talk


Kind regards

-Joel

[-- Attachment #2: Type: text/html, Size: 1004 bytes --]

Re: [Cerowrt-devel] LCA 2018 talk available

[Mikael Abrahamsson]

[-- Attachment #1: Type: text/plain, Size: 3497 bytes --]

On Thu, 25 Jan 2018, Joel Wirāmu Pauling wrote:

> Kia Ora (Hi in Māori).
>
> Today I delivered my talk on 10Gbit(+) in the home at Linuxconf
> Australasia. Some specific shout outs to those on the list who helped form
> some of the content and especially for the continued efforts with FLENT
> which I have been making extensive use of both professionally and privately.
>
> Hopefully this is of some interest and use to people on the list.
>
> https://github.com/aenertia/lca2018-talk/tree/talk

Great presentation, thanks.

Some feedback. I have been told MOCA is widely used in USA, and this is 
in-house coax cabling used for providing IP based services in multiple 
rooms.

http://www.instructables.com/id/How-to-set-up-a-coax-MoCA-network/

Now, this doesn't have much to do with your 10GE talk as it's not going to 
be that fast, but anyway. So back to > 1GE speeds.

It seems to me that 1GE is good enough for a lot of user needs. It's over 
100 megabyte/s, most HDDs won't even transfer faster than this. Most 
devices do not have anything faster than 1GE, so it's a chicken and egg 
problem. I have a 100EUR fanless managable 24 port switch with 4 SFP 
ports. I imagine anything faster than this would require fans and would 
bring up the cost a lot.

It would be ideal to have a 24 port 1GE + 4 (or 8) ports of 1/2.5/5/10GE 
for incremental migration, but I have 0 things in my home that speaks 
anything faster than 1GBASE-T (with RJ45 connector). I do have SFP+ based 
NIC cards and DAC cables, but I don't even use them (apart from occasional 
testing).

The upgrade was from 100BASE-T to 1GBASE-T was fairly cheap and addressed 
a wide need, since 10-11 megabyte/s was slower than most HDDs even 15-20 
years ago. Today, 100-110megabyte/s at 1GBASE-T speeds is actually still 
quite decent, and most people don't have huge amounts of data to move 
around. So for most people, anything faster than 1GBASE-T doesn't address 
a problem they actually have. Yes, for people handling 4k footage and 
doing video editing etc, they need faster. But most people don't. For them 
a 8-24 port 1GBASE-T switch is fine, and provides a networking solution 
that is not bottlenecking them in any significant fashion.

2.5G and 5G would be a good compromise, but it seems to be stuck in 
chicken/egg problem space. Most people actually don't even wire their 
computers today, it's all wifi, and even if they do wire them, the only 
NIC available is 1GBASE-T based.

The iMac Pro is the first prosumer device I have seen that actually 
supports faster networking. If Apple or someone else actually released a 
thunderbolt based NIC that was decently sized/priced that did support 2.5G 
or 5GBASE-T, then this chicken/egg problem could perhaps be solved. Most 
people don't feel the need to connect these kinds of things to their 
laptop:

https://www.startech.com/uk/Networking-IO/Adapter-Cards/thunderbolt-3-10-gbe-nic-chassis~BNDTB310GNDP
https://www.akitio.com/adapters/thunder2-10g-network-adapter
https://www.promise.com/Products/SANLink/SANLink2/10G-BaseT

http://www.tehutinetworks.net/?t=LV&L1=3&L2=0&L3=0&L7=157 is interesting, 
as this is not huge. It also does 2.5G and 5G.

https://www.anandtech.com/show/12422/akitios-thunder3-10g-adapter-now-available

300USD is still a significant chunk of money compared to the 29USD 
1GBASE-T thunderbolt2 adapter that Apple sells.

But still, with these kinds of products, there might be hope!

-- 
Mikael Abrahamsson    email: swmike@swm.pp.se

Re: [Cerowrt-devel] LCA 2018 talk available

[Joel Wirāmu Pauling]

Aquantia 10GBase-T TB3 self powered, adapters are now available. They
support 803.11bz.

Again it's not the speed, it's the throughput. TB3 delivers near to
what my local x86 can do in terms of throughput. Also network should
never be slower than disc. Since NVME has been around this is
no-longer true. It's an unnatural order of things.

Interestingly the NVME stuff came about by a competing teams during
the same time as Lightpeak. They share many underlying philosophies.

--

Cabling is the issue in my mind right now. Every laptop with tb3 ports
has 10G+ capability, if passive optical long run was cheap and easily
available for tb3 then half the problem would already be solved.

Maybe 10G over cat6a will be ok as the evolution. But you have to go
to cat8 to get anything beyond 10G... so the cabling situation and
incentive to upgrade to future-proof isn't there.

On 15 February 2018 at 00:33, Mikael Abrahamsson <swmike@swm.pp.se> wrote:
> On Thu, 25 Jan 2018, Joel Wirāmu Pauling wrote:
>
>> Kia Ora (Hi in Māori).
>>
>> Today I delivered my talk on 10Gbit(+) in the home at Linuxconf
>> Australasia. Some specific shout outs to those on the list who helped form
>> some of the content and especially for the continued efforts with FLENT
>> which I have been making extensive use of both professionally and
>> privately.
>>
>> Hopefully this is of some interest and use to people on the list.
>>
>> https://github.com/aenertia/lca2018-talk/tree/talk
>
>
> Great presentation, thanks.
>
> Some feedback. I have been told MOCA is widely used in USA, and this is
> in-house coax cabling used for providing IP based services in multiple
> rooms.
>
> http://www.instructables.com/id/How-to-set-up-a-coax-MoCA-network/
>
> Now, this doesn't have much to do with your 10GE talk as it's not going to
> be that fast, but anyway. So back to > 1GE speeds.
>
> It seems to me that 1GE is good enough for a lot of user needs. It's over
> 100 megabyte/s, most HDDs won't even transfer faster than this. Most devices
> do not have anything faster than 1GE, so it's a chicken and egg problem. I
> have a 100EUR fanless managable 24 port switch with 4 SFP ports. I imagine
> anything faster than this would require fans and would bring up the cost a
> lot.
>
> It would be ideal to have a 24 port 1GE + 4 (or 8) ports of 1/2.5/5/10GE for
> incremental migration, but I have 0 things in my home that speaks anything
> faster than 1GBASE-T (with RJ45 connector). I do have SFP+ based NIC cards
> and DAC cables, but I don't even use them (apart from occasional testing).
>
> The upgrade was from 100BASE-T to 1GBASE-T was fairly cheap and addressed a
> wide need, since 10-11 megabyte/s was slower than most HDDs even 15-20 years
> ago. Today, 100-110megabyte/s at 1GBASE-T speeds is actually still quite
> decent, and most people don't have huge amounts of data to move around. So
> for most people, anything faster than 1GBASE-T doesn't address a problem
> they actually have. Yes, for people handling 4k footage and doing video
> editing etc, they need faster. But most people don't. For them a 8-24 port
> 1GBASE-T switch is fine, and provides a networking solution that is not
> bottlenecking them in any significant fashion.
>
> 2.5G and 5G would be a good compromise, but it seems to be stuck in
> chicken/egg problem space. Most people actually don't even wire their
> computers today, it's all wifi, and even if they do wire them, the only NIC
> available is 1GBASE-T based.
>
> The iMac Pro is the first prosumer device I have seen that actually supports
> faster networking. If Apple or someone else actually released a thunderbolt
> based NIC that was decently sized/priced that did support 2.5G or 5GBASE-T,
> then this chicken/egg problem could perhaps be solved. Most people don't
> feel the need to connect these kinds of things to their laptop:
>
> https://www.startech.com/uk/Networking-IO/Adapter-Cards/thunderbolt-3-10-gbe-nic-chassis~BNDTB310GNDP
> https://www.akitio.com/adapters/thunder2-10g-network-adapter
> https://www.promise.com/Products/SANLink/SANLink2/10G-BaseT
>
> http://www.tehutinetworks.net/?t=LV&L1=3&L2=0&L3=0&L7=157 is interesting, as
> this is not huge. It also does 2.5G and 5G.
>
> https://www.anandtech.com/show/12422/akitios-thunder3-10g-adapter-now-available
>
> 300USD is still a significant chunk of money compared to the 29USD 1GBASE-T
> thunderbolt2 adapter that Apple sells.
>
> But still, with these kinds of products, there might be hope!
>
> --
> Mikael Abrahamsson    email: swmike@swm.pp.se

Re: [Cerowrt-devel] LCA 2018 talk available

[Joel Wirāmu Pauling]

it's late /throughput/latency/%s

On 15 February 2018 at 00:45, Joel Wirāmu Pauling <joel@aenertia.net> wrote:
> Aquantia 10GBase-T TB3 self powered, adapters are now available. They
> support 803.11bz.
>
> Again it's not the speed, it's the throughput. TB3 delivers near to
> what my local x86 can do in terms of throughput. Also network should
> never be slower than disc. Since NVME has been around this is
> no-longer true. It's an unnatural order of things.
>
> Interestingly the NVME stuff came about by a competing teams during
> the same time as Lightpeak. They share many underlying philosophies.
>
> --
>
> Cabling is the issue in my mind right now. Every laptop with tb3 ports
> has 10G+ capability, if passive optical long run was cheap and easily
> available for tb3 then half the problem would already be solved.
>
> Maybe 10G over cat6a will be ok as the evolution. But you have to go
> to cat8 to get anything beyond 10G... so the cabling situation and
> incentive to upgrade to future-proof isn't there.
>
> On 15 February 2018 at 00:33, Mikael Abrahamsson <swmike@swm.pp.se> wrote:
>> On Thu, 25 Jan 2018, Joel Wirāmu Pauling wrote:
>>
>>> Kia Ora (Hi in Māori).
>>>
>>> Today I delivered my talk on 10Gbit(+) in the home at Linuxconf
>>> Australasia. Some specific shout outs to those on the list who helped form
>>> some of the content and especially for the continued efforts with FLENT
>>> which I have been making extensive use of both professionally and
>>> privately.
>>>
>>> Hopefully this is of some interest and use to people on the list.
>>>
>>> https://github.com/aenertia/lca2018-talk/tree/talk
>>
>>
>> Great presentation, thanks.
>>
>> Some feedback. I have been told MOCA is widely used in USA, and this is
>> in-house coax cabling used for providing IP based services in multiple
>> rooms.
>>
>> http://www.instructables.com/id/How-to-set-up-a-coax-MoCA-network/
>>
>> Now, this doesn't have much to do with your 10GE talk as it's not going to
>> be that fast, but anyway. So back to > 1GE speeds.
>>
>> It seems to me that 1GE is good enough for a lot of user needs. It's over
>> 100 megabyte/s, most HDDs won't even transfer faster than this. Most devices
>> do not have anything faster than 1GE, so it's a chicken and egg problem. I
>> have a 100EUR fanless managable 24 port switch with 4 SFP ports. I imagine
>> anything faster than this would require fans and would bring up the cost a
>> lot.
>>
>> It would be ideal to have a 24 port 1GE + 4 (or 8) ports of 1/2.5/5/10GE for
>> incremental migration, but I have 0 things in my home that speaks anything
>> faster than 1GBASE-T (with RJ45 connector). I do have SFP+ based NIC cards
>> and DAC cables, but I don't even use them (apart from occasional testing).
>>
>> The upgrade was from 100BASE-T to 1GBASE-T was fairly cheap and addressed a
>> wide need, since 10-11 megabyte/s was slower than most HDDs even 15-20 years
>> ago. Today, 100-110megabyte/s at 1GBASE-T speeds is actually still quite
>> decent, and most people don't have huge amounts of data to move around. So
>> for most people, anything faster than 1GBASE-T doesn't address a problem
>> they actually have. Yes, for people handling 4k footage and doing video
>> editing etc, they need faster. But most people don't. For them a 8-24 port
>> 1GBASE-T switch is fine, and provides a networking solution that is not
>> bottlenecking them in any significant fashion.
>>
>> 2.5G and 5G would be a good compromise, but it seems to be stuck in
>> chicken/egg problem space. Most people actually don't even wire their
>> computers today, it's all wifi, and even if they do wire them, the only NIC
>> available is 1GBASE-T based.
>>
>> The iMac Pro is the first prosumer device I have seen that actually supports
>> faster networking. If Apple or someone else actually released a thunderbolt
>> based NIC that was decently sized/priced that did support 2.5G or 5GBASE-T,
>> then this chicken/egg problem could perhaps be solved. Most people don't
>> feel the need to connect these kinds of things to their laptop:
>>
>> https://www.startech.com/uk/Networking-IO/Adapter-Cards/thunderbolt-3-10-gbe-nic-chassis~BNDTB310GNDP
>> https://www.akitio.com/adapters/thunder2-10g-network-adapter
>> https://www.promise.com/Products/SANLink/SANLink2/10G-BaseT
>>
>> http://www.tehutinetworks.net/?t=LV&L1=3&L2=0&L3=0&L7=157 is interesting, as
>> this is not huge. It also does 2.5G and 5G.
>>
>> https://www.anandtech.com/show/12422/akitios-thunder3-10g-adapter-now-available
>>
>> 300USD is still a significant chunk of money compared to the 29USD 1GBASE-T
>> thunderbolt2 adapter that Apple sells.
>>
>> But still, with these kinds of products, there might be hope!
>>
>> --
>> Mikael Abrahamsson    email: swmike@swm.pp.se

Re: [Cerowrt-devel] LCA 2018 talk available

[Mikael Abrahamsson]

[-- Attachment #1: Type: text/plain, Size: 2257 bytes --]

On Thu, 15 Feb 2018, Joel Wirāmu Pauling wrote:

> Again it's not the speed, it's the throughput. TB3 delivers near to what 
> my local x86 can do in terms of throughput. Also network should never be 
> slower than disc. Since NVME has been around this is no-longer true. 
> It's an unnatural order of things.

Having done networking since mid 80-ties, having the network be slower 
than disk has been the reality, forever, for me. The only time this might 
not have been true would be in the beginning of 1GBASE time, where single 
HDDs were slower than network. With in 10BASE-2 days, HDDs were doing a 
magnitude higher transfer speeds compared to network. Running NFS was slow 
compared to local drive.

> Cabling is the issue in my mind right now. Every laptop with tb3 ports
> has 10G+ capability, if passive optical long run was cheap and easily
> available for tb3 then half the problem would already be solved.

Cabling fiber is unfortunately always quite a lot harder and more 
complicated than copper, that's why RJ45 won. Having factory-made fiber 
cable with USB-C connectors at each end might work, if the active 
electronics can be made small enough. Think pulling these through holes in 
walls, through cable management etc. Unfortunately I doubt these will 
reach enough volume in near time to really become widely used due to their 
initial high cost.

> Maybe 10G over cat6a will be ok as the evolution. But you have to go to 
> cat8 to get anything beyond 10G... so the cabling situation and 
> incentive to upgrade to future-proof isn't there.

If we need higher than 10G speeds, then yes, fiber is the next natural 
evolution. I don't know how we're going to make single-mode fiber 
something that the average user can handle without problems. There are 
advantages though.

I am getting FTTH now. The cable they're putting is looks like this:

https://www.facebook.com/photo.php?fbid=10154951833141595&set=p.10154951833141595&type=3&theater

It has 3 strands and it's single mode.

So if we can light up these at a good cost/power/size compromise, the 
cables can be made extremely thin. Still wondering how the connectors etc 
are going to look like to make this end user friendly.

-- 
Mikael Abrahamsson    email: swmike@swm.pp.se

Re: [Cerowrt-devel] LCA 2018 talk available

[Joel Wirāmu Pauling]

If  POF (Plastic Optical Fibre) like install methods can be scaled up
to Polymer/Glass runs (Sharpie knife slicing/jam into receptor). I
don't see this being the problem. Depending on the Sheathing fibre is
just as good as UTP cabling. Magnitudes cheaper too.

When I learnt of POF I was excited, until I learned how it's severely
limited in the bandwidth/frequency transmission department.

I guess if we could get some sort of Clamp-on USB-C style adaptor for
fibre would probably be the ideal. I don't really see why this
couldn't work with MPO style fibre.


On 15 February 2018 at 02:21, Mikael Abrahamsson <swmike@swm.pp.se> wrote:
> On Thu, 15 Feb 2018, Joel Wirāmu Pauling wrote:
>
>> Again it's not the speed, it's the throughput. TB3 delivers near to what
>> my local x86 can do in terms of throughput. Also network should never be
>> slower than disc. Since NVME has been around this is no-longer true. It's an
>> unnatural order of things.
>
>
> Having done networking since mid 80-ties, having the network be slower than
> disk has been the reality, forever, for me. The only time this might not
> have been true would be in the beginning of 1GBASE time, where single HDDs
> were slower than network. With in 10BASE-2 days, HDDs were doing a magnitude
> higher transfer speeds compared to network. Running NFS was slow compared to
> local drive.
>
>> Cabling is the issue in my mind right now. Every laptop with tb3 ports
>> has 10G+ capability, if passive optical long run was cheap and easily
>> available for tb3 then half the problem would already be solved.
>
>
> Cabling fiber is unfortunately always quite a lot harder and more
> complicated than copper, that's why RJ45 won. Having factory-made fiber
> cable with USB-C connectors at each end might work, if the active
> electronics can be made small enough. Think pulling these through holes in
> walls, through cable management etc. Unfortunately I doubt these will reach
> enough volume in near time to really become widely used due to their initial
> high cost.
>
>> Maybe 10G over cat6a will be ok as the evolution. But you have to go to
>> cat8 to get anything beyond 10G... so the cabling situation and incentive to
>> upgrade to future-proof isn't there.
>
>
> If we need higher than 10G speeds, then yes, fiber is the next natural
> evolution. I don't know how we're going to make single-mode fiber something
> that the average user can handle without problems. There are advantages
> though.
>
> I am getting FTTH now. The cable they're putting is looks like this:
>
> https://www.facebook.com/photo.php?fbid=10154951833141595&set=p.10154951833141595&type=3&theater
>
> It has 3 strands and it's single mode.
>
> So if we can light up these at a good cost/power/size compromise, the cables
> can be made extremely thin. Still wondering how the connectors etc are going
> to look like to make this end user friendly.
>
>
> --
> Mikael Abrahamsson    email: swmike@swm.pp.se

Re: [Cerowrt-devel] LCA 2018 talk available

[Mikael Abrahamsson]

[-- Attachment #1: Type: text/plain, Size: 1755 bytes --]

On Thu, 15 Feb 2018, Joel Wirāmu Pauling wrote:

> If  POF (Plastic Optical Fibre) like install methods can be scaled up
> to Polymer/Glass runs (Sharpie knife slicing/jam into receptor). I
> don't see this being the problem. Depending on the Sheathing fibre is
> just as good as UTP cabling. Magnitudes cheaper too.
>
> When I learnt of POF I was excited, until I learned how it's severely
> limited in the bandwidth/frequency transmission department.
>
> I guess if we could get some sort of Clamp-on USB-C style adaptor for
> fibre would probably be the ideal. I don't really see why this
> couldn't work with MPO style fibre.

Problem with multimode fibers (that I imagine all above are) is that they 
have really low reach at higher speeds. So if your goal is to support 100G 
cheaply, I doubt the above will work very well.

Looking at:

http://ieeexplore.ieee.org/document/6192181/

"100G transmission over GI-POF is demonstrated by using 112-Gb/s 
PolMux-QPSK modulation and digital coherent detection. The transmission 
distance over 100m and simplified coupling method is suitable for optical 
data-center applications."

PolMux-QPSK is never going to be cheap'n'easy, it requires substantial 
amount of components and processing power at both ends of the fiber.

For cheap GI-POF 1000BASE-RH has been defined 
http://www.ieee802.org/3/GEPOFSG/email/pdfKiCuEsVuMv.pdf but it's only 
1GE. I doubt POF is a solution to your 10GE/100GE transmission speeds 
requirement.

So I think the conclusion basically is: 10GE and up is hard. There are 
real physical limitations here. I would be very happy if we could prove 
Shannon wrong. :P

https://en.wikipedia.org/wiki/Shannon%E2%80%93Hartley_theorem

-- 
Mikael Abrahamsson    email: swmike@swm.pp.se