Ethernet Routers for Audio

[Guest rmpfyf]

58 minutes ago, danrey said:

See here as to why it's voodoo science:

https://www.audiosciencereview.com/forum/index.php?threads/uptone-audio-etherregen-switch-review.10232/

 

Please dude. Amir's jitter measurements and visualisations are a joke, that's coming from someone with a professional background in digital signal processing (me). 

 

I'd launch a website called AudioBetterScienceReview so I could talk down to people that didn't know better on a regular basis but I'm not that insecure. 

[Stereophilus]

22 minutes ago, pwstereo said:

Why is the terminology “ground plane”?

It doesn’t seem right in this situation.

The Ethernet signal is on multiple twisted pairs.

Once again, I will defer to the article.  I don’t want to copy and paste too much, but he does explain this concept quite clearly.  Beyond that, I’m not much use.  I am interested to hear valid arguments against the theory he presents though. Not so much interested in the device he sells, just the theory in this article. So far it still seems very logical to me.

[danrey]

10 minutes ago, rmpfyf said:

 

Please dude. Amir's jitter measurements and visualisations are a joke, that's coming from someone with a professional background in digital signal processing (me). 

 

As a software engineer who has worked in digitising tv broadcast environments in this country, i'm all ears as to why it's badly done.

[Guest rmpfyf]

Just now, danrey said:

As a software engineer who has worked in digitising tv broadcast environments in this country, i'm all ears as to why it's badly done.

Pull up any one of his jitter plots and show me the relevant part, and how it was acquired. 

[Guest rmpfyf]

1 hour ago, recur said:

Audio is a long way down the list of essential services IP networking provides and offers no unique problems to solve that haven't already been solved for other purposes.  Audiophiles are about as confused about networking as that kid with Beats headphones on his iPhone is with his Dad buying a $20k vinyl rig.

 

I would go a little easy and offer another perspective, not least as (from a scientific perspective) I don't think @dbastin, @Assisi and others are hearing things. I run similar experiments and the differences are palpable. 

 

We need to get beyond this being about the data or anything in the actual networking stack. The data arrives fine.

 

We also need to get beyond a notion of 'there's a buffer therefore it's a moot point'. That's not what buffers are for. They are not infinite in size. They store data, they don't offload interrupts. 

 

Fundamentally we're dealing with how the endpoint behaves relative to how the data arrives. 

 

Build a toolstack to measure some direct and not-so-indirect measures - OS jitter is a good thing to try to capture. It'll change on MTU, speed, pretty much anything you can throw at it. Not all changes are large.

 

This isn't like async USB where there's a timed relationship to when and how data arrives, though (AES67 argument again) the capability exists within networking - it is not implemented here.

 

1 hour ago, recur said:

There isn't a problem that the traditional issues in the audio realm (noise, interference etc) have that doesn't have a compensating control in the networking stack to ameliorate it.  If there was, it would have been fixed for much more important time critical or data sensitive applications like high frequency trading, financial settlements, video distribution networks or the plethora of IoT centric devices like PLCs that make your food, make your electricity or filter your water supply.  

 

I'd take issue here as in ultra-time-critical stuff there's network tech that isn't part of usual Eth transfer. Though it could be added, and emerging standards exist to deal with this. Till then we have punters trying to make things better (very) indirectly.

 

There's also this -

 

34 minutes ago, Stereophilus said:

I agree that none of the upstream components, CPUs, processors, are themselves jitter or timing sensitive. 

 

Very much untrue. A modern CPU is extremely timing sensitive, and in particular the way it manages power states (very) quickly and the knock-on effects for upstream power system design mean a modern PC is a rig designed for flexible power scaling, not for rock-steady repeatable periodic performance - and the two really are quite opposing design goals. Now it's not timing-sensitive in a way that a minor timing issue will throw out functionality, it's the inverse - it's inherently designed to be highly tolerant of timing variance. That doesn't mean timing is important, it means that if you want a CPU to deliver an outcome in a highly resolute and dependent way, it'll be done indirectly and with very specific controls. 

 

These problems are ultimately well towards the bleeding end. Optimised network and non is not a gulf between awesome and unlistenable unless something is really, really wrong. Electrical noise isn't making its way to DAC ICs unless there's a cascade of very sloppy design at play.

 

But it's not right to suggest it doesn't exist.

[DSD_Fan]

1 hour ago, Hydrology said:

For now Audio Connection in NSW.

Umm. I'll pass lol

 

 

[dbastin]

48 minutes ago, rmpfyf said:

Dale, it was in my post last night - good enterprise-grade switch > fibre > etherregen > very short CAT6 UTP lead > streamer.

Ah, so you are suggesting no router!

[Guest rmpfyf]

4 minutes ago, dbastin said:

Ah, so you are suggesting no router!

There's to be a router above all this, sure, but it's not nearly as essential to your areas of interest as what happens downstream.

[PCOWandre]

2 hours ago, :) Go Away (: said:

Fibre cables: https://www.fs.com/au/products/40192.html

Yellow patch lead? Nope, you want aqua:

https://www.fs.com/au/products/40220.html

[:) Go Away (:]

17 minutes ago, PCOWandre said:

Yellow patch lead? Nope, you want aqua:

https://www.fs.com/au/products/40220.html

Aqua! Sure.

but yellow presents a more natural warmer sound over longer distance. Say from Spotify server to a DAC.

 

Joking aside, yes you are right, aqua cable is needed.

[PCOWandre]

I was going to try to add an arty snapshot of some happy home fibre, but it's cramped and bloody dark in the back of the rack, so you can have this poorly lit blur-pocalypse instead:

[Stereophilus]

1 hour ago, rmpfyf said:

Very much untrue. A modern CPU is extremely timing sensitive, and in particular the way it manages power states (very) quickly and the knock-on effects for upstream power system design mean a modern PC is a rig designed for flexible power scaling, not for rock-steady repeatable periodic performance - and the two really are quite opposing design goals. Now it's not timing-sensitive in a way that a minor timing issue will throw out functionality, it's the inverse - it's inherently designed to be highly tolerant of timing variance. That doesn't mean timing is important, it means that if you want a CPU to deliver an outcome in a highly resolute and dependent way, it'll be done indirectly and with very specific controls. 

If I read you correctly, you are saying CPUs are extremely robust in their jitter management.  But are you saying that jitter at, say, the router CPU is potentially audible?

 

My (probably over-simplified) argument was intended to state that jitter incurred at these CPU points upstream of the DAC is not the problem.  If I read your statement correctly, we agree that the mechanisms built into these CPUs ensure that timing errors / jittery signals are inconsequential to the integrity of the signal transmission.

Edited September 16, 2020 by Stereophilus
Punctuation only

[PCOWandre]

Ethernet frame timing isn't jitter. Badly formed ethernet frames will be discarded, but they are created by a MAC that has nothing else to do so if the actual frames having a timing problem the switch is faulty and should be discarded. 

 

To dig for analogy, talking about the impact of frame timing on ethernet would be like saying your CD sounded different because auspost sent it on an odd journey to your house. It either arrived on time and you had music, or it didn't show up on time and you had no music. That's just dragging the time scale out. 

 

Carrying the analog a little further, you listen to one CD a day. On average, one CD a day turns up in the mail. Sometimes no CD, sometimes two. You keep a small stack of CDs next to the player so you never run out because the post doesn't come on weekends or bad weather. That's readahead, buffering, caching, etc. Now, the council tears up the road past your house (cable broken) and the postman skips your delivery for a week and your stack is empty. Now you've got an audible drop-out and you're miserable (like most auspost customers). So being a forward-planning individual, you figure you'll solve this with more read-ahead, buffering and caching and keep a month's supply of CDs. Now your supply line can be interrupted and as long as it catches back up to the overall rate of 1 CD a day, you'll never be high and dry.

 

Your neighbour is a bit skeptical of the postal service, and keeps a stack of 200 CDs next to his player at all times. It takes a little while tidy up the stack at the start of playing (filling a deep buffer) but he'll be listening to the latest Beatles re-issue while you've got dead air because the new postal worker has given your CDs to the next street over for the last week. That's using tweaked-up overkill buffering and readahead to make sure there's no problems and this is the scenario where you can pull the cable and the music keeps on coming because the entire track and part of the next one are already in RAM. 

 

On tiny UPNP/DLNA implementations -- with very small amounts of memory and no filesystem layer to add buffer and readahead -- an overloaded NAS can cause audible glitches because the buffer is empty. The cure for this is to take the tiny UPNP/DLNA implementation and put it in the bin while sighing gently, OR to make sure you have a storage head that doesn't get overloaded. 

 

The moral of the story: always have a deep stack of CDs to listen to, then you can ignore ethernet and be happy. 

 

[Rosco8]

Beware ... a  slight rave   

 

Geez, I have been thru some Networking gear over the past 30 years ....  and I was in IT for 45 years up until Dec last year.   I used to buy the latest and greatest Networking gear for my kids and me to play with from Fry's in the Silicon Valley and in LA when i was travelling for work.  My kids learnt heaps and we had lots of technical fun ...  I weirdly got into a serious cease and desist argument about homework with their Primary school Headmaster because my boys were using computers to research and then to print assignments out.   I would argue that Computers will become part of everyone's job in the future to no avail.  Eventually I found out the problem was that they had no access to computers, and felt threatened until they had experience.  In high school my oldest ended up running their first School network  

 

Both my sons are sparkys and both did a lot of Networking jobs.

 

A new home 6 years ago and Cat 6 to every room.  Yep was wonderful .... for a while ...

 

My main problem has always been getting enough bandwidth because as a family we became more and more bandwidth hungry, exponentially I reckon, gaming by the boys as teenagers where they would arrange gaming nights in my house, rooms and corridors filled with kids on computers and switches and cables running everywhere ...  playing against kids on the other side of the world ... amazing !!!!!   The nasty word was 'we are lagging'  !!!!!   

 

Then they grew up and it was businesses run from home, cloud hosting and required better upload speeds.

 

More recently connected monitoring devices resulted in implementing 2xADSL2 and a 4G data hub multiplexed into different groups so we didn't compete with one another .... finally NBN finally arrived this year and I now run a 100Mb plan with Aussie Broadband and thought all will be good ..  well it was for a short time.   Mmmm .. I feel a repeating theme ...

 

You see everybody in the family keeps buying and upgrading devices that want MORE wireless access.    Next step was a booster (satelite) hub (all TP-Link gear) and adding 4 wifi boosters that just plug into powerpoints to cover dead spots, including my shed, outside on the alfresco and within the house.  For a while we were travelling nicely again  ... however that recently reached a new limit ...  a quick wireless survey and there is a minimum of 21 devices connected, Hey Google devices, Nest smart devices, RING cameras, smart phones, smart watches, smart TV's, IQ cable boxes and the list goes on !!!!!   When the boys hit home we have up to 36 devices all chattering on the network.  My oldest then discovered that our monitoring devices were struggling to upload to their respective clouds.  You must be kidding is what I thought !!!!!

 

Audio .. so I have 4 devices that I use for Tidal, 1 is on Cat 6 and the rest are wifi, though I only use a max of 2 concurrently whilst the boys use Spotify, a max of 3 concurrently at times as we also have a stray staying with us as well ....  so lots of competition and we have lag problems,drop outs and excess buffering, which as we all know is REALLY ANNOYING on Audio.   I rarely have any problem on the Cat 6 network, its that damm wifi world !!!!

 

So this week we installed a Mesh wifi system ... Core unit plus Satellite, Orbi Wifi-6.  Tricky as we discovered its also a router and we had DHCP conflicts with the existing fixed Network until my son made the Orbi the only router in the house, so we disconnected the existing router and added the fixed Cat 6 switches to it as well.  Now we have no lagging, no buffering and the existing Ethernet (Cat 6) network  has now become the secondary network in the house, I was not expecting that to happen so fast.  The Mesh happily stretches up to the back shed as well and I just checked now and the Mesh Network is supplying 102.77 Mbps download and 38.07 Mbps upload.   In my office I am getting 68Mbps, in my Audio/theatre room I am getting 97Mbps which is overkill to stream Tidal and watch Netflix. I am very impressed with its performance and the way it efficiently manages so many devices concurrently.  No lagging on video, no buffering on Tidal, monitoring is now in real time.  Mesh is certainly the way to go for us ....

 

My thoughts for what ever they are worth ....

- Its about getting the bandwidth with minimal lagging (QoS) that your device requires.  Video (downloading and especially uploading) I found in our house is a lot more demanding that audio downloading.

- Wireless is by far the easiest to deploy, no dragging cables thru the house, and the technology just keeps improving.  I will only invest now in Wireless.  Mesh and Wifi-6 will rapidly become cheaper, plenty of new systems forecast towards the end of the year.

- Wireless is coming with just about every device, everything is being branded/marketed as Smart, you don't really have a choice than to embrace Wireless.  I was surprised at how many devices in my house now need wireless today.

- Audio to me is just another device requiring good QoS i.e,  guaranteed priority bandwidth with minimal lag.

- Audio; I just don't believe in buying special audio Ethernet cables, routers, switches. Your end devices take care of sequencing.  Just buy decent Networking gear that more than covers what you need today, as tomorrow you will likely need more. 

 

Last thought - I was fortunate to spend a day at a Network supplier forum many moons ago with a gentleman named Robert Metcalfe (worked for Xerox, who remembers the Star workstation running on Ethernet, it was the future of end user computing, took my breath away when I first saw it), now the forum was at the time of the great Token Ring vs Ethernet debate ... I can still remember Robert telling me over lunch that at that time Ethernet 10Mb 'is like somebody pissing in a 6ft pipe'  and that over time it will just keep increasing  .. yep ... Ethernet won out, its Gbit now and labs are building Terabit !!!!    Copper should be dying in Networking but ...   Optic fiber and Wifi ....        Cao .....

Edited September 16, 2020 by Rosco8

[Guest rmpfyf]

1 hour ago, Stereophilus said:

If I read you correctly, you are saying CPUs are extremely robust in their jitter management.  But are you saying that jitter at, say, the router CPU is potentially audible?

A modern CPU can do its job irrespective of timing inaccuracies. It is accordingly not inherently designed to spit out data or a complex process in a temporally perfect manner.

 

Jitter is inconsequential to signal transmission. It is very consequential to managing a process that requires time-accurate outcomes. 

 

Ask a modern CPU to do one thing, and one thing only in a very explicitly programmed way. You'll get close to that with a very, very and I mean very stripped back operating system. Forget graphics. 

 

Ask it to do many things and those other things better happen regularly if you don't want an audible impact.

 

E.g. 

 

Play music

Play music

Play music

Play music

Deal with an Ethernet packet

Play music

Play music

Play music

Play music

Deal with an Ethernet packet

 

Sounds quite different (in a general sense) to something more irregular:

 

Play music

Play music

Deal with an Ethernet packet

Play music

Play music

Play music

Play music

Deal with an Ethernet packet

Deal with an Ethernet packet

Play music

 

And that's vastly oversimplified.  Though you might get the drift here. Your modern CPU and OS ensures the Ethernet packets are dealt with and the music is played - eventually, ideally as close as possible to now or in some specified order of priority - but what it generally isn't is 

 

Play music

Play music

Play music

Play music

Play music

Play music

Play music

Play music

Play music

 

That'd be your average CD transport. Any interrupts there are genuine to your cause - hitting pause interrupts your playback which is fine, that's why you asked as much. 

 

The best way around this is to limit what your CPU needs to and quite literally tune out everything else. What @PCOWandre describes above and in an earlier post is just that - in not making Ethernet a problem/area where any improvement would bring about a benefit you can save a ton and do it smart. Upcoming standards literally force best behaviours and will work wonders IMHO when it's network direct to a DAC in an active rig, though good design downstream helps much.

 

I did a decent Eth system in mine and spend the most cash reclocking I2S... that's where it counts, IMHO.

Edited September 16, 2020 by rmpfyf

[Guest rmpfyf]

6 minutes ago, Rosco8 said:

Last thought - I was fortunate to spend a day at a Network supplier forum many moons ago with a gentleman named Robert Metcalfe

 

We're only 20 years in, sir, and I'd already like to nominate this comment for Casual Namedrop of the Century.

 

Here we are having a casual yarn over networking and you slide in with 'I was chatting with God recently'. 

 

Xerox my ass BTW. He worked a Xerox PARC, as you know. Those are some hallowed walls. 

 

I've lived and worked in Silicon Valley previously and would have given quite a bit for a casual yarn with Metcalfe. That dude is super impressive.

[dbastin]

On 16/09/2020 at 3:57 PM, rmpfyf said:

There's to be a router above all this, sure, but it's not nearly as essential to your areas of interest as what happens downstream.

Maybe not essential, well honestly none of this is essential, but it makes a difference.

 

I had very similar to what you suggest.

 

good enterprise-grade switch (Ubiquiti EdgeSwitch) > fibre > etherregen > very short CAT6 UTP lead (1m Synergistic Research Atmosphere X Ref ethernet) > streamer.

 

I changed from the stock iinet nbn router to Uniquiti EdgeRouter X (both using SMPS) and enjoyed an inprovement in SQ.  And this is just over Cat 5e UTP to the EdgeSwitch.    Based on my experience, I suspect the following will each provide SQ inprovements:

 

- better grade router

- LPS on router

- power for router via my power conditioner

- fibre from router to Switch or EtherRegen.

- ECT in router

 

And I am going to put a media convertor just after nbn box to run fibre to the router.

 

And have no wire ethernet into the router.

 

All this is pretty simple, low cost and likely provide quite good SQ for money.

 

So, some specific suggestions for enterprise grade routers with 4 or more SFPs would be very useful.

 

This is the realm of the network experts amingst us because there is only 2 audiophile routers, neither accept fibre.  One of these is also an audio component, also Waversa ...

 

https://www.kevalinaudio.com/product-page/waversa-wrouter

 

Edited September 19, 2020 by dbastin
typo

[Rosco8]

44 minutes ago, rmpfyf said:

 

We're only 20 years in, sir, and I'd already like to nominate this comment for Casual Namedrop of the Century.

 

Here we are having a casual yarn over networking and you slide in with 'I was chatting with God recently'. 

 

Xerox my ass BTW. He worked a Xerox PARC, as you know. Those are some hallowed walls. 

 

I've lived and worked in Silicon Valley previously and would have given quite a bit for a casual yarn with Metcalfe. That dude is super impressive.

I wondered whether anyone would know who he was ... years later i actually visited the Xerox labs in San Diego .. specializing in electronic publishing ... PARC was gone but got to sit in one of those hallowed bean bags that was saved from the original labs.

Edited September 16, 2020 by Rosco8

[Guest rmpfyf]

17 minutes ago, dbastin said:

Maybe npt essential, well honestly none of this is essential, but it makes a difference.

 

I had very similar to what you suggest.

 

good enterprise-grade switch Ubiquiti EdgeSwitch) > fibre > etherregen > very short CAT6 UTP lead (1m Synergistic Research Atmosphere X Ref ethernet) > streamer.

 

I changed from the stock iinet nbn router to Uniquiti EdgeRouter X (both using SMPS) and enjoyed an inprovement in SQ.  And thus is just over Cat 5e UTP to the EdgeSwitch. 

 

 

Neither the iinet gateway nor the EdgeRouter are enterprise grade. Neither is my Mikrotik. 

 

You will need a router though as per my comments, what you need sits upstream of what's already been suggested

 

1m is a long lead relative to what you're trying to achieve. Think of it this way - would you have a 1m lead from your DAC transport clock to where it's used if you could avoid it? I mean go really short. 

 

17 minutes ago, dbastin said:

So, some specific suggestions for enterprise grade routers with 4 or more SFPs would be very useful.

 

No. Not least as you're straight into solution mode, which is a dangerous place to be given what needs to be understood here.. Not least as there are very many solutions - how many fibre ports? What type? What support? What are your space requirement? Noise? Etc.. There's quite literally a plethora of solutions out here.

 

29 minutes ago, dbastin said:

Based on my experience, I suspect the following will each provide SQ inprovements:

 

- better grade router

- LPS on router

- power for router via my power conditioner

- fibre from router to Switch or EtherRegen.

- ECT in router

 

If you're going to try to hotrod any component it'd be the switch, not the router, though think of what you can actually change - there's no EMI down a fibre line, only packet timing.

 

38 minutes ago, dbastin said:

This is the realm of the network experts amingst us because there is only 2 audiophile routers, neither accept fibre.  One of these is also an audio component, also Waversa ...

 

https://www.kevalinaudio.com/product-page/waversa-wrouter

 

See, for that much money I'd sooner have a MERGING+NADAC, run RAVENNA on some very standard Ethernet cables and hardware, have next-gen audio-over-Ethernet performance, run any application I like and have something to actually listen to. 

 

For a good bit less money I had a I2S reclocker built and integrated, and a lot less dependence on what happens upstream.

[Guest rmpfyf]

26 minutes ago, Rosco8 said:

I wondered whether anyone would know who he was ... years later i actually visited the Xerox labs in San Diego .. specializing in electronic publishing ... PARC was gone but got to sit in one of those hallowed bean bags that was saved from the original labs.

 

Fortunate in my career, I was doing Strategic Planning for IT from the early days so got to meet a lot of the great masterminds in IT.   l did my RDBMS database training under Dr Ted Codd (Author of the Relational data theory) and SQL training with Chris Date (Team lead of SQL at IBM).  I got pissed with Ted one evening and he told me how he obscured the Relational theory into a debate with Universities so IBM wouldn't catch on, as they would kill it to protect IMS.  It worked, the world got DB2.  I also did a half day session including an hour one on one with James Gosling in the fledgling years of Java ... he gave me a run down on where they pinched the different bits of Java from other languages.   I created Qantas.com and during that period spent a lot of time with Netwscape in the Valley and then with Sun Labs, which in those days was into some incredible initiatives, like interactive TV, the labs were managed by an ex United Airlines IT guy who had been the head of IT Aircraft Engineering.   So lucky ....

 

I'm going to PM you sometime to swap Silicon Valley war stories not least because yours are better than mine.

 

Sounds like you've had a hell of a ride.

[PCOWandre]

2 hours ago, Rosco8 said:

Robert Metcalfe

I vanish for a while to eat dinner and take all the blue LEDs out of a preamp, and I come back to name dropping the Father of Ethernet. Nice. The only member of the old guard I've met is McNealy.

 

Anyway, we're four pages in and still don't have a defined noise problem or network issue. Or any problem at all, other than a need to spend some moolah during these boring times.

 

1 hour ago, rmpfyf said:

you're straight into solution mode, which is a dangerous place to be given what needs to be understood here

This might be the most valuable contribution to the entire thread. The only time good IT decisions are ever made is when the problem is properly described and success/fail criteria. 

 

My setup? Cisco 3845 router, Cisco 3750G core switch, Procurve 1800 series in the office and loungeroom (managed, but fanless). Procurve 2810 48-port switches elsewhere. Storage heavy lifting on fibre channel, Brocade Silkworm 300. Medium/light lifting on NFS, with each storage head connected via quad gigabit LACP and compute on failover pairs of quad gigabit LACP. Mikrotik for VPNs to remote sites; 4G backup for those VPN links and BGP routing to make the magic happen.

 

And I hand terminated all my own long cables, and use 4cabling.com.au for all my patch leads. And my next network upgrade is waiting until I can find a 10 gig L3 switch that isn't going to cost the earth and isn't going to consume enough power to make me snarl any harder when the bill comes in. And talking of power, if I was going to upgrade anything that wasn't speakers/amp/dac to improve SQ, it might be more power management because in the Pariah State all six UPSes agree that the AC voltage swings between 215V and 249V every day. 

 

2 hours ago, dbastin said:

ECT in router

Being a router is depressing, but electro-convulsive therapy is just over the top. 

[keinesorge]

6 hours ago, DSD_Fan said:

Umm. I'll pass lol

At 2695 i'm afraid i'll also have to give it a pass.

[DSD_Fan]

1 hour ago, keinesorge said:

At 2695 i'm afraid i'll also have to give it a pass.

Yeah. Being $915 US RRP. That's a bit steep. 

[PCOWandre]

For less money on ebay:

[keinesorge]

29 minutes ago, PCOWandre said:

For less money on ebay:

With that amount of backplane capacity, the Bandwidth Envy Coefficient will disrupt the quantum transmission of the audio passband.

Not to mention the cooling fans without phase lock.  Think of the modulation of the schumann field around the listening area.

 

When will you people learn.