fs.com 100m SFP-10G-T100 / ubiquiti UACC-CM-RJ45-MG

[Jabes]

Hi,

Just wanted to give some feedback here for some long link 10gb connections.
I have approx 70m of cat6 cabling going between my house and garage and wanted to try running it at 10Gb even though it is "out of spec". I have TP-Link Omada switches with SFP+ ports at each end.

I first tried the new ubiquiti UACC-CM-RJ45-MG, which is quoted as supporting 100m and is at the competitive price of £70. This is listed as maximum power consumption of 1.9W.
The great news is that it immediately synced to 10gb; which I verified using iperf3.
The not so good news is that there appears to be a compatibility issue between this module and my TP-Link switches. On a full power off recycle the port doesn't come up, requiring the module to be removed and re-inserted. TP-Link support (who were excellent) did some testing with me, and procured their own module for a detailed look. Their engineers identified a particular non-spec issue with this module which they claim will also mean it doesn't work with other switches that bring their ports live after boot (I can find their exact description of the issue if anyone is interested).
If you choose this module for non unifi hardware I would suggest you verify that it continues to work when your switch is power cycled whilst in your return period.

Now I had learnt it was possible to get this link up and running I turned to the fs.com SFP-10G-T100 - this is more expensive (£156) - but is likely the same chipset as the ubiquiti as is also a 100m supporting module, at max 1.8W of power consumption.
I inserted these modules and the link immediately came up and has been working for over 2 months with no problems. This is a main link in my networking so I would immediately notice if there were any drop-outs.

It'd be great to see sth review these two modules to highlight that longer distances at lower power are now possible, and also to bring to the fore the compatibility issues on the ubiquiti module. I hope some other modules appear in this lower price range!

 

[Jabes]

I found the response I got from tp-link about the unifi module

Thank you for your patience! We have now completed our analysis of the UBNT optical module, and have concluded that its hardware design has a compatibility issue with our switch.

According to our R&D analysis, there is a pin (TX_Disable) on the UBNT's optical module directly connected to its PHY chip, unlike ours and most of those from other brands. Our switch ports are disabled during the system startup, and are only activated after the system is fully booted. However, the UBNT pin design requires the ports to be activated during the system boot process, otherwise the optical modules will not be recognized after rebooting the switch, which causes the problem.

We are not sure why UBNT designed the module in this way, but we have tested other brands of switches and found the same issue. In addition, we believe that our port design is reasonable and is followed in all our products, i.e., the ports are activated after the system has finished booting. Otherwise, it may lead to many data packets entering the network device during the system startup, resulting in abnormalities

 

[Gaara]

@Jabes I use the UACC-CM-RJ45-MG in a Zyxel XGS2220-30HP switch and have no problems. Even after a full power cycle it is working just fine. Maybe this problem is just specific to TP-Link? My UACC-CM-RJ45-MG shows revision U07, how about yours?

 

[Jabes]

@Jabes I use the UACC-CM-RJ45-MG in a Zyxel XGS2220-30HP switch and have no problems. Even after a full power cycle it is working just fine. Maybe this problem is just specific to TP-Link? My UACC-CM-RJ45-MG shows revision U07, how about yours?

TP-Link didn't make it sound like it would be specific to them; but it can't generally apply or it wouldn't work even in a unifi switch!
There may also be a newer hardware revision if Ubiquiti care about broader compatibility and this has hit their radar (I did raise it directly with them).

I don't have the devices any more (my supplier accepted the returns), but the log extract from the tplink cli has the device identifying as a U07

Te1/0/28 Copper-SFP Ubiquiti Inc. 850 0 0 0 0 AK22117501293 UACC-CM-RJ45-MG 103 U07 8 RJ45

I had four of these; and one of the ones I had connecting directly to an RJ45 connection from an intel card would work on boot, none of the others did - I put this down to timing given what tp-link said

I think the basics of my post remain - the UACC-CM-RJ45-MG is a great choice - but beware if you aren't using ubiquiti. It'd be great to hear other success stories to guide other people about compatibility of this module

Hope this helps

 

[oneplane]

I can somewhat confirm that UBNT is a bit odd here in that the tx_disable pin is supposed to be a GPIO-controlled pin, so if they do it differently (i.e. connect it to a static output pin on the PHY) it will not be software-controlled. That means that either the entire SFP cage needs to have additional internal power control (some switches don't do that so save some money), or you have to manually re-power it by re-plugging it.

Normally you'd be able to tell the disable pin to go high or low with an I2C or SPI write. This is often also managed using an FPGA (or CPLD in old hardware) where many things like the LEDs and per-cage control are multiplexed so you don't need to have a management CPU or fabric ASIC with a billion control lines just to mess with each SFP.

 

[Jabes]

I can somewhat confirm that UBNT is a bit odd here in that the tx_disable pin is supposed to be a GPIO pin, so if they do it differently (i.e. connect it to a static output pin on the PHY) it will not be software-controlled. That means that either the entire SFP cage needs to have additional internal power control (some switches don't do that so save some money), or you have to manually re-power it by re-plugging it.

Normally you'd be able to tell the disable pin to go high or low with an I2C or SPI write. This is often also managed using an FPGA (or CPLD in old hardware) where many things like the LEDs and per-cage control are multiplexed so you don't need to have a management CPU or fabric ASIC with a billion control lines just to mess with each SFP.

I don't know if you are on the side of the module implementing it wrongly, or tp-link should be more accepting of this design - but it sounds like what they have said makes some sense

 

[Pholostan]

I got a couple of UACC-CM-RJ45-MG and they do not work in any of my Mellanox ConnectX-3 cards. Pro or non Pro, older or newer firmware. Just nothing. Cheap Qnap switch works fine with them.

Edit: Huh, I must have made some mistake, but after running them in my QNAP I got them to work in my Mellanox cards. Dunno why it didn't work the first few times, but they work now. Just that the RJ-45 jack in my apartment don't want to play ball with it, I suspect that has to do with 2.5G, another converter that is 1/2.5G does not get carrier either. A 1/10G module works fine.

 

[WhiteNoise]

@Jabes Does the FS.com SFP-10G-T100 work at 2.5G and 5G? what about 1G?

 

[Jabes]

The specs say so but I have not verified this myself.

 

[WhiteNoise]

The page that I saw on fs.com do not say so. Which spec page are you referring to?

 

[Jabes]

https://www.fs.com/uk/products/154916.html

 

[WhiteNoise]

Thank you very much, now I see it.
This from FS.com and UniFi are the only 100m (or even 80m) that explicitly claim to support 1G, 2.5G, and 5G.

It would be great if somebody can empirically confirm. Anyway, I'll try to send them an email.

 

[Jabes]

I haven't tried mine at lower speed (it is running a long distance 10G connection). However, I also have the Wiitek modules and for me they do work at 2.5G since I do that every day.

 

[WhiteNoise]

the Wiitek 100m? good to know!

 

[Jabes]

the Wiitek 100m? good to know!

Yes, I have the Wiitek 100m modules as well (I posted about them elsewhere on here)

 

[donileo]

I am also have the same issue using the UACC-CM-RJ45-MG on an ICX-6610. On restart of the switch the sfp port using this module fails to power on and I have to unplug and replug the module to get the port to work. Is the WiiTek 100m the SFP-10G-T-S? The part number on Wiitek's website shows it as SFP-10G-T-S but on Amazon.com (US Store) I see it listed as SFP-10G-T-100 not sure if they are the same. Also is this module from Wiitek 1.8w at 100m like the Ubiquiti or 2+ w?

 

[Jabes]

The wiitek ones are the SFP-10G-T-100 - I think the T-S might be the shorter range ones? I have these, and also the fs.com 100m modules (which are more expensive). I believe they are 1.8w like the ubiquiti (same chipset), and are much lower temperature

```
Temperature(C) Voltage(V) Bias Current(mA) Tx Power(mW) Tx Power(dBm) Rx Power(mW) Rx Power(dBm) Data Ready Rx Los Tx Fault
Te1/0/1 46.808594 3.323300 6.000000 0.500000 -3.010300 0.400000 -3.979400 True False False

Te1/0/5 45.472656 3.366500 6.000000 0.500000 -3.010300 0.400000 -3.979400 True False False

Te1/0/9 53.000000 3.250000 6.016000 0.380800 -4.193031 0.510800 -2.917491 True True False

Te1/0/11 53.425781 3.296500 6.000000 0.500000 -3.010300 0.400000 -3.979400 True False False

Te1/0/14 55.812500 3.257000 0.000000 0.000000 -inf 0.000000 -inf True False False

```

```
Transfer Distance Bit
Wavelength SMF OM1 OM2 OM3 Rate
Port Transceiver Type Vendor Name [nm] [km] [m] [m] [m] Serial Number Part Number [Mbps] Rev Compliance Connector Type
---------- ------------------ ---------------- ---------- ---- ---- ---- ---- ---------------- ---------------- ------ ---- ---------- ------------------
Te1/0/1 10G_BASE_SR_SFP Wiitek 850 0 20 80 300 WAMZ012409X25C SFP-10G-T-100 103 1 3 LC
Te1/0/4 Copper-SFP OEM 256 0 0 0 0 S180500340398 SFP-H10GB-CU2M 103 0 Copper pigtail
Te1/0/5 10G_BASE_SR_SFP Wiitek 850 0 20 80 300 WAMZ012409X10C SFP-10G-T-100 103 1 3 LC
Te1/0/9 10G_BASE_SR_SFP FS 850 0 40 100 0 G2310005048 SFP-10G-T100 103 3 LC
Te1/0/11 10G_BASE_SR_SFP Wiitek 850 0 20 80 300 WAMZ012412X006C SFP-10G-T-100 103 1 3 LC
Te1/0/13 Copper-SFP OEM 256 0 0 0 0 S180601540116 SFP-H10GB-CU2M 103 0 Copper pigtail
Te1/0/14 Copper-SFP TP-Link 0 0 0 0 0 222C9K1000663 TL-SM5310-T(UN) 103 1.0 0 LC
Te1/0/16 Copper-SFP OEM 256 0 0 0 0 CSC210601480435 SFP-H10GB-CU1M 103 R 0 Copper pigtail
```

 

[Halken]

Is it only the Wiitek that uses the new Broadcom chipset, or are there some from Qsfptek as well, as they are more readily avalible here in Europe?
Do the 80m use less power than the 30m or? I only need this for 4 meters...
Also, I have a China SFP+ switch & intel nics. What coding do you need or does it not matter?

 

[blunden]

Is it only the Wiitek that uses the new Broadcom chipset, or are there some from Qsfptek as well, as they are more readily avalible here in Europe?
Do the 80m use less power than the 30m or? I only need this for 4 meters...
Also, I have a China SFP+ switch & intel nics. What coding do you need or does it not matter?

Lots of other brands use them now. Pretty much all the 80 and 100 meter ones use it, with the 100 meter ones supposedly using slightly less power than the 80 meter ones despite using the same chip (seemingly binning).

Yes, the 30 meter ones draw significantly more power than the 80 and 100 meter ones. Unless you need the Aquantia based transceivers for some reason, I personally don't think anyone should buy 30 meter rated 10GBASE-T transceivers anymore now that the price from Chinese manufacturers has dropped so much. I managed to pick up the ZYOPM 100 meter model for $25 including 25% VAT, although that seems to have been an unusually good deal. $35-40 including VAT seems to be pretty standard though.

The thread below includes a number of other examples:

https://forums.servethehome.com/index.php?threads/10gb-sfp-rj45-100m.45817/

Your Chinese switch very likely doesn't care. For Intel NICs, you very likely need the transceiver to fake being a fiber transceiver, which many of them do. My XICOM 80 meter transceiver came like that by default. ZYOPM asked me what brand of equipment I intended to use it with and programmed it accordingly. Have you unlocked your Intel NICs btw.? They can be vendor locked by default.

 

[Jabes]

I absolutely agree with the above. No-one should buy the 30m modules.
all of the module makers should bin them off and only sell the 80/100 parts they are so much better (actually within spec temperature and power wise is a great start!)