Melco S100 Ethernet Switch Measurements

[jabbr]

It would be far more interesting to repeat the comparison against a 10G fiberoptic switch that costs a bit over $100

[jabbr]

31 minutes ago, R1200CL said:

I suppose if a buffer is to be used in order to block phase noise (jitter), this buffer must be located after last network interface, so it can’t be inside a switch.  Maybe it can be inside the opticalRendu (I know there is one, but not sure how big, or if it does anything to reduce jitter). 
 

10Gbe Ethernet has very low phase noise/jitter *as specified* and moreover jitter is *not allowed* to travel from one hop to the next. This is written into the compliance testing specifications. I’ve posted on this exhaustively.

[jabbr]

8 hours ago, R1200CL said:

And as you know there is in general accepted that a PC is the worst endpoint. With expensive exceptions. 

 

I'm neutral on this ... I don't generally accept.

 

Is a NUC a PC?

 

What about a low power celeron board with PCI-e?

 

You can get an inexpensive low powered celeron micro-ATX and use an LPS to supply, cheap 10G fiber NIC and perhaps a fancy USB card ...

[jabbr]

3 hours ago, R1200CL said:

Talking about measurements.
 

An entire village lost its broadband at the same time every day for 18 months. Now we know why

 

Second-hand TV wipes out broadband for entire village

 

A lightning strike took out a friend's home network.

 

I have fiber to the home and fiber in the home ;)

 

[jabbr]

7 hours ago, The Computer Audiophile said:

I'm late to this party, but for many of the measurement nuts, it means something is better engineered because it has better numbers so it must be better. 

 

I look at it as a difference that was measurable at the output of a DAC, which the objective crowd says is impossible. We can see here that it actually is possible. 

 

If the objective crowd was really into science and discovery, they'd try to determine why/how this happened. 

 

Its certainly not impossible -- at all.

 

leakage current is a simple explanation ... a ground loop.

 

[jabbr]

7 minutes ago, R1200CL said:

Clock phase noise is another issue. Somehow it seems to simple that it’s just gone by adding buffers. Eliminating the purpose of etherRegen if you use fiber and have a buffer as well. 
I may speculate in why those magic buffers then wasn’t added to any of John’s design.

 

10Gbe fiber does not have appreciable phase noise! By specification. By "appreciable" I mean that no audio manufacturer has ever demonstrated that they can improve over it ... its all in the sub-picosecond range (i.e. femto) moreover it cannot travel from link to link (as specified)

[jabbr]

40 minutes ago, R1200CL said:

On the other hand, shouldn’t it be possible to do 1GB with same specifications?

Well, maybe etherRegen is close to that. We would need an upgraded version with fiber B-side.

 

10Gbe switches and NICs work at 1Gbe also. You get the same engineering, clocks etc. I’ve run 10, 25 and 100Gbe interfaces at 1Gbe 😉

[jabbr]

17 minutes ago, plissken said:

 

10GB modules only work at 10GB. So you can't have a 10GB tranceiver on one end and 1GB on the other.

 

SFP only supports 1GB

SFP+ supports 1GB and 10GB modules

 

Many newer modules are multi rate ie the Intel 10/1Gbe

 

Most SFP28 modules do 25/10/5/1 

[jabbr]

1 hour ago, R1200CL said:

Does there even exist a consensus about how to measure a switch?

Is there a standard for it, and if so, will measurements fully cover what we’re after ?

Yes of course there are standards. The 1Gbe specs are ancient/pre-historic really. The industry realized over two decades ago that measurement standards were necessary. The 10Gbe specs provided that. You cannot claim compliance with 10Gbe without doing measurements! 
 

These specifications were specifically designed to prevent jitter from traveling across network hops. These issues are only a discussion because we are dealing with $10 home switches using generic parts otherwise.

[jabbr]

4 hours ago, R1200CL said:

On 9/24/2020 at 10:00 PM, jabbr said:

 

Are you sure a SFP28 will fit into a normal SFP cage ?

Getting off topic in this thread but, no,  SFP cages need SFP modules.

 

You can use a single rate 1Gbe SFP module on one end and connect to a multi rate SFP(+) or SFP28 module at the switch.

 

[jabbr]

32 minutes ago, R1200CL said:

 @jabbr says different. Let’s be patient. As he stats those modules can also give 1GB. 
 

 

SFP(+) modules fit in SFP(+) ports in 10Gbe devices. SFP modules fit in SFP(+) ports but SFP(+) modules do not fit in 1Gbe SFP ports.

 

You need a switch with SFP(+) ports i.e. 10Gbe.

 

I don't know if the EtherRegen nor Melco S100 switches meet the 10Gbe standards because they are not 10Gbe devices nor have I seen measurements against the 10Gbe and up stressed eye pattern tests that are required and implemented by e.g. Agilent / Keysight or Tektronix.  I cannot assume that either of these devices, or any other device which does not claim compliance with the standards, meet these standards.

 

Obviously 10Gbe and faster switches, typically designed for the professional environment, have such testing done as part of the design process and before the products are sold.

[jabbr]

1 hour ago, firedog said:

I think that's the wrong question. It doesn't matter how the switch itself measures. It matters how the switch does or doesn't change the output at the DAC. What if the switch is "low noise" but that changes nothing at the output of the DAC when compared to a standard "noisey" switch? 

I'm not interested in the measurements of the switch. I'm interested in how it does or doesn't change the output of the DAC. That's part of my problem here. Why is it so hard for manufacturers of such devices to provide such measurements? And if they somehow need "special devices" to measure DAC output, why can't they contract out for their devices to be measured when in line before the DAC? Doesn't seem like something that should be especially expensive or impossibly difficult. 

 

The switch measurements are required for newer technology switches, e.g. those using less than 20 year old technology.

 

How can you expect a company which doesn't at least do the industry standard measurements, be able to measure the effects of switch jitter on the output of a DAC?

 

They haven't and presumably either can't or don't want to. Fiction and stories sell more than documentaries. Using dragonglass to banish jitter is a sexy story. The nightwalkers are coming ...

 

This is entertainment

[jabbr]

1 minute ago, R1200CL said:

Based on @plisskenand @jabbrinsight we can conclude it’s impossible to implement any of John’s present optical design into a 10GB network, even if some optical modules can operate under 1GB speed. 

 

I have no idea what "John's present optical design is", but consider that a 10Gbe switch can operate at 1Gbe, and realize that its the same circuit, clocks, PSU etc.

 

1 minute ago, R1200CL said:

For those using computers as streamers/players/endpoints can enjoy experimenting with 10GB network.

 

So maybe we can hope to see audio devices with SFP+ and SFP28 (10GB) some day.

 

If Uptone / Sonore / SGC is first, l wonder if some other companies will follow.

I think it has been said other companies is coping them. 

 

The opticalRendu, the clearfog that I've been using since ?2016 and any endpoint including PCs with NICs, can have the benefit of a 10Gbe network -- which works with 1Gbe endpoints just fine using either 1Gbe or 10/1 Gbe modules.

 

1 minute ago, R1200CL said:

Meaning above 1 GB ?

I’m asking, cause otherwise I could start trying to get measurements from many manufacturers of switches. Including my ISP. 
 

Now let’s have some entertainment 🤩

 

 

No doubt if you become a customer of Cisco or Mellanox etc, they will provide compliance testing. @plissken do you insist prior to a major install? I'm fine with trusting Intel, Cisco, Mellanox, HPE, Dell etc ...

 

[jabbr]

Just now, plissken said:

 

In the medical field you can't install something that the vendor hasn't supplied a conformance/compliance certificate.

 

FS.COM modules are even FDA certified.

 

The take home message is that if you use a 10Gbe+ compliant switch, you can be assured that the stressed eye pattern conformance testing has been done. That's the take home message for switch measurements.

 

Going further, though, suppose you wanted to see the actuall stressed eye pattern tests ... what would it take to get those?

[jabbr]

8 minutes ago, R1200CL said:

Can I have a link to one or several SFP modules that both support a 10GB switch as well as the opticalRendu? 

 

 

First, the SFP module on the switch need not be identical to that on the opticalRendu.

 

Yes, I do this. My 100Gbe switch  literally connects to my 1Gbe  clearfog. My 10Gbe Mikrotik connects to my 1 Gbe clearfog, both also connect to my Intel NIC.

[jabbr]

1 minute ago, MarkusBarkus said:

One of the primary issues was related to stacking tolerances. Individual pieces were quite precise, but their lack of knowledge caused them to miss the tolerance outcome when the pieces were assembled. Oops.

 

And no excuses for that other than hubris, BTW.
 

I recognize this is a little adjacent to your point, but it makes me want to leave the door ajar for well meaning manufacturers who may feel there is no way they can *predict" what impact every possible combination could create. 
 

I'm not defending anyone, or tolerating sloppy design, just leaving the door open a

bit.

 

 

Right, that's why the newer network standards not only specify things like SFP+ modules but *end to end* jitter, precisely so you *can* mix and match different switches and NICs and SFP(+) modules and it just works. Of course if a piece of equipment is not compliant or faulty or otherwise defective it won't work, and there are always little gotchas, but the very strict compliance testing was designed a long time ago precisely to avoid your old company's problem...

 

Isn't it miraculous that the Internet works!!! Lots of brilliant folks have been working on this.

[jabbr]

7 minutes ago, R1200CL said:

This is important information. I always assumed they had to be equal. 
 

Still I’m missing suggestions for one SFP+ / SFP28 that matches the suggestion below. 

 

This switch: https://www.servethehome.com/mikrotik-crs305-1g-4sin-review-4-port-must-have-10gbe-switch/

 

Single mode or multimode (need both on same end of link)

 

get 2 cisco SFP modules, stick one in the Mikrotik and another into the opticalRendu

 

.... that's the simple answer for a 10Gbe capable switch

 

I have also used Finisar 10/1 Gbe dual speed SFP(+) modules as well as Mellanox multispeed SFP28 with Mellanox NIC and several QSFP28 modules in Mellanox SN2700 switch

 

I am sure the "Optical Networking and SFP" thread has these details

[jabbr]

2 minutes ago, Superdad said:

 

Since the others here seem not to have stated it plainly enough to avoid confusing you, I’ll take a crack at it:

 

While one can order special dual-rate SFP+ modules that will work at 1 Gigabit in a standard SFP port, standard SFP+ transceivers are 10Gbe (or higher) and will not negotiate down when plugged into a 1Gbe SFP port.

However, one can use 1Gbe transceivers in SFP+ ports.

 

So the obvious and reasonable question still remains:

If you buy two of the special type, specifically dual-rate SFP+ transceivers and install one in a 1Gbps-only SFP cage (of say, your opticalRendu or EtherREGEN) and install the other in either:

Yeah this is confusion, let me clarify:

 

Don't do this. Don't put a dual rate 10/1 SFP(+) module into an SFP port.

 

Put an SFP module into the SFP port

2 minutes ago, Superdad said:

a) the SFP+ cage of a 10Gbps switch;

 

Do this: put the dual rate 10/1 SFP(+) module into the SFP(+) port 

Do this: put a 1Gbe SFP module into an SFP(+) port

 

2 minutes ago, Superdad said:

or

b) the SFP cage of a 1Gigabit switch,

will the pairing properly negotiate down to 1Gbps?

 

Don't do this.

 

2 minutes ago, Superdad said:

 

I suspect yes for b), but am not sure about a).

 

 

yes for a) ... don't do b)

[jabbr]

4 minutes ago, R1200CL said:

So still no SFP+ that’s support both 10GB as well as 1GB then. I suppose not important. 

 

I'm trying to help, read this:

https://www.mouser.com/datasheet/2/610/finisar_ftlx8574d3bcv_1g-10g_850nm_multimode_datac-1530955.pdf

 

4 minutes ago, R1200CL said:

So then if using  a 1GB SFP module, I won’t have more jitter compared to using a 10GB SFP+ module?

It’s crucial that the SFP module won’t degrade the quality of the switch. 

 

The "Optical Networking and SFP" thread extensively discusses the differences between SFP(+) modules

 

4 minutes ago, R1200CL said:

As the idea is of cause to see if this switch could be equal to etherRegen. 

 

Equal in what way? Its a 10Gbe switch

 

4 minutes ago, R1200CL said:

 

And how about using this module?

https://mikrotik.com/product/S-RJ01
Still no jitter at 100GB. (Yes, I could used it against etherRegen B side😀)

 

That is a module that allows you to use a copper RJ45 cable with the switch. You can make this as combinometrically complex as you like.

[jabbr]

3 hours ago, R1200CL said:

As far as I understand there exist only one multimode 1/10 GB Finsair module. (And one single mode). 

https://www.finisar.com/sites/default/files/downloads/ftlx8571d3bcv_1g_10g_850nm_multimode_datacom_sfp-plus_transceiver_product_specification_revb.pdf

 

I use this selection chart to find it:

https://www.finisar.com/sites/default/files/resources/finisar_optical_transceiver_product_guide_3_2015_web.pdf
 

Do not insert this product into an etherRegen or an opticalRendu. It will get destroyed. 
 

Now @plissken or @jabbr

Will this product automatically communicate with 1GB SFP if that SPT is dual (LX) and 850nm ?

(Intended to be in opticalRendu/etherRegen/Cisco 2960). 

 

The LX version is: https://www.finisar.com/sites/default/files/downloads/finisar_ftlx1471d3bcv_rohs-6_compliant_1g-10g_10km_datacom_sfp_transceiver_product_specification.pdf

 

read the datasheets, the one you quote clearly states that its multimode (SX), LX is singlemode

 

3 hours ago, R1200CL said:

 

Is jitter Per IEEE 802.3-2005 Table 38-10 equal to almost none as stated earlier by you @jabbr ?

(Jitter ref. as pr. data sheet).

By using this CRS305-1G-4S+IN and the suggested SFP+ module, should I then expect a jitter free / phase noise free equal or even better than etherRegen ? 

 

Fiberoptic does not transmit common mode noise, much better isolation than any electrical "moat"

 

 

3 hours ago, R1200CL said:

If we assume ground loop noise etc is eliminated by the moat in the etherRegen, will inserting a RJ45 SFP into a 10GB switch also ensure elimination of jitter and phase noise ?

 

CSS/CRS305-1G-4S+ -- supports up to 2 simultaneous S+RJ10 modules.

 

 

 

You can use the RJ45 in and fiberoptic out.

[jabbr]

4 hours ago, R1200CL said:

 

Now @plissken or @jabbr

Will this product automatically communicate with 1GB SFP if that SPT is dual (LX) and 850nm ?

(Intended to be in opticalRendu/etherRegen/Cisco 2960). 

 

Is jitter Per IEEE 802.3-2005 Table 38-10 equal to almost none as stated earlier by you @jabbr ?

(Jitter ref. as pr. data sheet).

By using this CRS305-1G-4S+IN and the suggested SFP+ module, should I then expect a jitter free / phase noise free equal or even better than etherRegen ? 

 

Time for you to do an experiment and report results.

 

Get 3 finisar LX/LR modules (single mode)

 

2x 1000base-LX (SFP)

1x 10g/1g dual LX/LR SFP(+) (per the sheet I linked to)

 

put one LX sfp module in your opticalRendu

put the other in the Mikrotik

 

listen

 

put the 10g/1g SFP+ module in the Mikrotik

 

listen

 

can you hear a difference?

 

further discussion on the "Optical Network Configurations" thread which exists for exactly these issues

 

To close this discussion in this thread: I have never seen any objective data to demonstrate that *any* 1Gbe Ethernet switch has either lower phase noise/error/jitter not lower common mode noise transmission/leakage current, than the Mikrotik 10Gbe switch using fiber, nor any other 10Gbe switch using fiber. The 10Gbe switches have gone through extraordinary testing over the last 20 years which has been extensively documented in the literature.  There is no equivalent actual data known to me for the 1Gbe series of switches.

[jabbr]

6 hours ago, R1200CL said:

What is more interesting to understand and learn more about is my question about using a RJ45 SFP+ port into that Mikrotik switch. Will the clock jitter still be gone ?

 

Yes, as specified.

[jabbr]

4 hours ago, charlesphoto said:

I tried a 10gb Mikrotik. Went back. I didn't find it anything special compared to my cisco switch or opticalModules. I probably didn't have it set up optimally or with the right SFP's, but I also didn't need that many fiber outlets. YMMV. 

 

That's fine. I have *never* claimed that 10Gbe fiber "sounds  better" than 1Gbe fiber. What I am saying is that *all* the compliant 10Gbe switches and NICs have been designed such that end to end jitter *is not* transmitted across links. Nor do any of the fiber devices transmit common mode noise (e.g leakage current) which is an electrical phenomenon.

 

The user whether professional or home, can be assured that the compliance sticker for 10Gbe indicates that the device has been tested for and does not pass along upstream jitter. Its a non issue. It may not be an issue for your 1Gbe devices but I haven't seen any testing.

[jabbr]

4 hours ago, Miska said:

 

Yes, certainly. I sort of always wonder why people pay attention to network interface jitter things with something like NAA. Since as protocol it would work even over internet between continents...

 

There are a considerable number of people who believe that server components generate different amounts of electrical noise and that this noise somehow makes it across the network, ultimately to the DAC.

 

No doubt NAA is a critical component because it runs on a low powered electrically quiet device.

 

The fiber ethernet ensures that no electrical noise can cross the network, so use a high powered server without concern and an electrically quiet NAA and voila the music comes out. 

 

[jabbr]

5 hours ago, R1200CL said:

It seems there is an agreement (silenced members included) that a 10GB network with fiber SFP+ cage will be almost an equal solution to what the etherRegen does.

 

What do you mean "almost equal solution"? I have not seen measurements to demonstrate whether the EtherRegen can handle the stressed eye pattern test required of all compliant 10Gbe switches.

 

Quote

 

But so far it’s only possible to build computers with 10GB fiber interface. So until SPF+ is the preferred standard for streaming.....

However I think testers, like @charlesphoto did find John’s design to sound better.

 

Maybe there will be an upgraded etherRegen with SFP+ as well as an opticalRendu / module with SFP+
But ops, this is where the Mikrotik should do the trick instead. My understanding this is what @plisskenand @jabbr is trying to say with so many words.
So I don’t expect an upgrade soon.

 

To be clear it is possible (and I have done this with every 10Gbe switch I have used) to use 1Gbe SFP modules with these switches. You get the benefit of the 10Gbe electronics which are known to have very low jitter and noise.

 

Quote

 

I’m still not sure how to understand that Yes from @jabbr as he added “as specified”, was with the 10Mhz standard or the 10GB standard as that RJ45 SFP+ module supports a wide range of speeds.

 

If that Mikrotik is powered by an LPS 1.2, ground loops shouldn’t be an issue. 

 

Per the 10Gbe specification, with a copper in and fiber out link, the outgoing link will not allow upstream jitter across the down stream link. No reason the 10Gbe switch would not perform equally well for a 1Gbe downstream link.

 

Now none of this says anything about "SQ" or what any individual's perception is, rather only about objectively measurable jitter and common-mode noise/leakage -- for both are a non-issue for all compliant 10Gbe fiber switches. Everyone should choose whatever they wish, but we are talking in the objective subforum here and what I say is regarding measurable parameters.