Jump to content
IGNORED

Optical Networking & SFPs


Recommended Posts

On 6/18/2020 at 7:33 PM, ray-dude said:

With the Planet Tech SFPs, I preferred copper NIC on the Extreme (fiber to the opticalModule+PlanetTech to copper to the Extreme).  With the Finisars, I'm back to preferring the optical NIC, esp. now that I have my audio network isolated from my home network, but it isn't a clear preference.  

 

FWIW, subjectively, reduced network traffic to the my music server NIC does have an impact.  This is contributing to my hypothesis of laser switching-induced electrical noise being the mechanism in play.

 

 

It is hard to know with "Startech" and "Planet Tech" equipment. Since you are not cost constrained, is there are reason not to use major brand 10Gbe or faster NICs, switches which have known engineering and are known to have gone through the stressed eye-pattern testing? Intel (I have x520), Solarflare (I have 6322f and 7322f), Mellanox (I have ConnectX-4 and ConnectX-5) hmmm .... let's see I know the jitter budget for the cards goes in reverse order (the new Mellanox is crazy low) but I can't personally hear a difference among those three NICs. I mean sure Finisar (and get a new SFP28 module for the Mellanox) but I can't vouch for any objective performance numbers on the equipment you list, nor do I know who makes the ASICs, lasers etc. My current path up to the endpoint is Mellanox FWIW.

 

But you are correct, some of the older, off-brand/generic fiberoptic hardware may not have been adequately tested (newer standards are tighter and tighter) so you can't make blanket statements about old fiber being automagically better.

 

What I *can* say is that leakage currents through the fiber will be zero!

Custom room treatments for headphone users.

Link to comment
On 6/18/2020 at 4:34 PM, ray-dude said:

Thank's for the pointer to this topic Ken.  My hypothesis is that switching to the laser is causing a correlated pulsing on the reference voltage and/or ground planes that is having an indirect impact on the DAC.


reasonable theory however,

 

1) the lines to/from the SFP(+) are differential, so constant current.

2) compliance testing for 10Gbe+ disproves significant ground bounce

 

so available measurements fo not support thus theory, nor have I seen measurements of legacy 1Gbe SFPs which demonstrate this.

 

That said I would be surprised if a generic SFP or NIC failed to meet the standards ascribed to by major manufacturers. For me the differential cost of going with a well known manufacturer is low enough that I do that and don’t worry about this.

Custom room treatments for headphone users.

Link to comment

It’s interesting that in this so called “objective” thread we are hearing many speculations and philosophical arguments — remember that real world equipment has to pass compliance testing so perhaps we should consider the specifications. How much ground bounce does a 10 Gbe eye pattern allow? 100 Gbe? He he not much 😉 I know that certain manufacturers throw this out as a speculation but seriously given the known compliance testing, let’s see the data! 
 

From my perspective this equipment is all very cost effective, so like my XLR cable for which I go with well known manufacturers such as Neutrik connectors and brand name cable (pick) I also use brand name SFP(+) modules  and I consider Finisar/Avago to be brand names. 

Custom room treatments for headphone users.

Link to comment
9 minutes ago, The Computer Audiophile said:

Are there links to any of that, that I can read up on?

Do you mean the Tektronix stressed eye pattern testing primer — I’ve posted the link

 

SFP(+) I/o https://en.m.wikipedia.org/wiki/Small_form-factor_pluggable_transceiver

 

lines 12/13 are receive +|- and 17/18 are transmit +|-

Custom room treatments for headphone users.

Link to comment
17 hours ago, ray-dude said:

I remember looking at this page when I first heard differences.  On pins 15 and 16, the laser is actually turning on and off (even if fed with a differential signal).  

 

Why do you assume that in all or any implementations the laser actually turns on and off? https://www.optcore.net/transceiver-laser-types/ the more modern lasers are DML/DFB and EML https://community.fs.com/blog/silicon-photonics-and-lasers-in-100g-optical-transceivers.html

 

In any case if there is an irregularity on the power/ground planes, this would be evident in the eye pattern, no?

Custom room treatments for headphone users.

Link to comment
10 hours ago, kennyb123 said:

 

 

Whether it's on and off there is electrical activity that brings about whatever the laser is doing.  
 

 

of course that’s silicon photonics.

Quote

 

 

I'm really trying to understand what lasers are doing if not on and off.  Any chance you could explain what is actually occurring instead of providing links that don't clearly explain this either?

 

To simplify, the silicon modulates the light after it leaves the laser. It might disperse the light, decreasing its intensity, or absorb the light. 
 

Think about this like a gate on a transistor, a small gate current switches a much larger current between the collector and emitter. 

Quote

 

The article from FS that you provided mentions "injecting current".  The link mentions this too.  If current is being injected, why is it wrong to assume that this could result in a "pulsing load on the power lines"?

 


Anything could happen in theory. The question is whether there are fluctuations in the power and ground planes. The stressed eye pattern testing should pick this up, no?  People make all sorts of assumptions about what is going on in an electrical circuit. For example there is constant current switching, so an increase in current draw by one part in a circuit is met with a decrease in a complimentary part. 
 

Bottom line is whether there is noise on the data lines (electrical). That is what is measured. Consider a 60 femtosecond jitter Budget for a clock. You can’t very well have a pulsing power supply and meet that, can you? The stressed eye pattern testing is where the rubber meets the road.

Custom room treatments for headphone users.

Link to comment
  • 2 months later...

We have focused extensively on the properties of different SFP, SFP(+) and SFP28 modules and lasers etc but that is only part of the overall story.

 

As of introduction of 10Gbe 20 years ago, end to end compliance testing was recognized as essential to prevent, among other things, jitter from propagating down a network. The "stressed eye pattern" test was introduced and is implemented by Tektronix and Keysight measurement systems, among others.

 

Switches and NICs compliant with 10Gbe+ must pass these tests.

 

A 1Gbe switch which was designed to pass this stressed eye pattern test would similarly not transmit jitter across its link. I have not seen a 1Gbe switch vendor provide such testing (its not required), but that said, I have not seen evidence that *any* 1Gbe switch of any design, perform better in this regard than the cheapest 10Gbe switch which is also compliant to this specification.

Custom room treatments for headphone users.

Link to comment
  • 2 weeks later...
  • 9 months later...
11 hours ago, tategoi said:

You can get one for less than £4 from China before shipping 😏


https://m.tb.cn/h.4xpTySQ?sm=a7aa28

 

He he but handling the shipping through China, to the export port, onto the ship and to America can be, shall we say challenging. I've ordered products, tracked them to have them disappear just before being packed into a container at the port ... and there is no one to call. The business that Amazon is in is entirely logistics and I appreciate what it takes to get many products to me within 24 hours -- its really impressive as much as everyone complains.

Custom room treatments for headphone users.

Link to comment
13 hours ago, MarcelNL said:

If it is anything like the Solarflare card I'm using getting HOT is normal, hey what do you expect ...it uses a laser ;-)

FWIW I've found the Intel x520 to get less hot. Check out the power consumption of the cards. What goes in comes out as light and heat and electrons. Assume the light and electrons to be the same across boards. Boards that consume more power, generate more heat.

Custom room treatments for headphone users.

Link to comment
21 hours ago, MarcelNL said:

I read the specs  for the Solarflare forwards and backwards its, saying it uses 7 Watts, it amazes me how much heat can be generated by 7 Watts....

 

Well … the heat generated is guaranteed to be <= 7 watts if that’s the consumption. The Intel is 1/2 that. Get one of those LED cooking thermometers and check the chip temps on the running boards …

Custom room treatments for headphone users.

Link to comment
  • 6 months later...
17 hours ago, MarcelNL said:

Forgot to add: DO check compatibility with your gear, as the 1475 operates at 40Gb/s whereas the previous one maxxed out at 2.7Gb/s...both plenty fast but my previous switch did not cope with the 1475

The finisar FTLX1475 devices are 10GbE ... there are devices which do 25GbE (SFP28) which is the current max for single lane, the 40Gbe are multiplexed 10x4 and 100Gbe are 25x4

 

[EDIT] no I stand corrected: the current Max is 50Gb/s per lane in this 400Gbe device is 8x50 multiplexed over a single mode pair! https://ii-vi.com/product/400gbase-lr8-qsfp-dd-optical-transceiver/

Custom room treatments for headphone users.

Link to comment
  • 2 weeks later...
On 2/19/2022 at 8:26 PM, PYP said:

I started with the eR to reduce noise, then added an opticalModule.  Interestingly, the opticalModule did further improve sound quality until I added a second eR at which point the optical solution wasn't needed -- two eRs sounded better.  Just my setup, but the optical solution I tried did not eliminate all the noise.  Perhaps it is different if one runs fiber from the router...or perhaps the eR eliminates other kinds of noise.

 

Since we are on the objective subforum, let's try to look at this:

 

1) there isn't a direct correlation between network noise and the noise you hear.

2) I can't guess as to what noise either of the devices you listened to injected into your audio system, or if they did without measurements.

3) compliant 10Gbe and faster network devices have to be tested using expensive equipment to ensure that the do not transmit jitter and differential mode noise (in the case of copper). Fiberoptic does not transmit common mode noise (directly)

4) without measuring, its hard to speculate why two devices in series would be better than one.

 

I don't worry about this and use professional grade equipment. Also isolate my audio area with a robust transformer, and no cheap SMPS  power supplies in the audio area.

Custom room treatments for headphone users.

Link to comment
  • 2 months later...
On 4/30/2022 at 7:16 PM, Archimago said:

So guys...

 

We're 8 pages into optical networks / SFP networking discussions. Is there any objective evidence (I include controlled listening A/B testing) that this makes a difference?

 

Considering that we don't listen to ethernet data, but analogue output from DACs, has anyone ever shown objective benefit after all this time?

 

 

It is rather easy to show that fiberoptic ethernet does not transmit common mode noise which results in leakage currents. Such leakage currents are widely considered relevant in hospital/ICU settings where there is the theoretical risk of cardiac events.

 

In any case Ethernet leakage currents are measured by industry. There are electrical measures (filters, isolation transformers) to mitigate, however fiber trivially eliminates.

 

Do leakage currents aka ground loops cause audible problems? Obviously. They have been measured forever and the audible effects well documented.

 

So its more than measureable. We don't need controlled A/B listening tests to want to eliminate ground loops. There are many ways to eliminate ground loops. I, for one, use multiple methods including fiber ethernet, isolation/balanced power transformers and balanced audio circuits.

 

 🚫 common mode noise

 

Custom room treatments for headphone users.

Link to comment

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now



×
×
  • Create New...