Hasivo Switch DAC Support ?

[oneplane]

It's mostly current leakage over the shielding and not-so-great internal power regulators, fiber has no electrical connection and thus the whole problem doesn't exist with a fiber SFP.

 

[blunden]

It's mostly current leakage over the shielding and not-so-great internal power regulators, fiber has no electrical connection and thus the whole problem doesn't exist with a fiber SFP.

Interesting. Did you measure this somehow?

I wonder why it works when I put my other cheap switch in-between. Is it more tolerant to current leakage?

 

[oneplane]

Interesting. Did you measure this somehow?

I wonder why it works when I put my other cheap switch in-between. Is it more tolerant to current leakage?

It was mostly just analysis based on the PCB and experiments a few pages ago in this thread. What normally happens is that external ports have galvanic isolation, especially in network equipment, and when that doesn't happen correctly you usually get some sort of current leakage which causes all sorts of power regulation issues (such as ports resetting or the entire ASIC resetting).

Generally, if it's software-only, the issues are limited to just a port and not an entire device. Because DACs tend to be electrically connected (because the ports are supposed to be isolated) they conduct, while optical connections do not. I think you can get optical directly attached SFP+ lines but at that point getting some SFPs and a patch fiber makes more sense. DACs on short runs are mostly supposed to be a cheap and power saving alternative (because there is no electricity-to-light required).

 

[blunden]

It was mostly just analysis based on the PCB and experiments a few pages ago in this thread. What normally happens is that external ports have galvanic isolation, especially in network equipment, and when that doesn't happen correctly you usually get some sort of current leakage which causes all sorts of power regulation issues (such as ports resetting or the entire ASIC resetting).

Generally, if it's software-only, the issues are limited to just a port and not an entire device. Because DACs tend to be electrically connected (because the ports are supposed to be isolated) they conduct, while optical connections do not. I think you can get optical directly attached SFP+ lines but at that point getting some SFPs and a patch fiber makes more sense. DACs on short runs are mostly supposed to be a cheap and power saving alternative (because there is no electricity-to-light required).

I see. That could certainly be the case.

It doesn't really explain why my Hasivo switch didn't show a link with my 2 meter DAC, but did with a 0.5 meter DAC from the same manufacturer (so they should be built the same way) though. That's at least how I remember it. If I'm not mistaken, that was part of the reason for why I thought it was a signal integrity issue. After all, the switch would only need to be able to send the signal intact a couple of centimeters (the trace distance from the switch chip and the SFP+ slot) since the transceiver handles producing the signal from there.

Yeah, AOC cables are basically fiber cables with transceivers fused directly to each end. Now that we have cheap fiber transceivers, the cost seems to be basically the same between an AOC cable and transceivers + a fiber cable so the latter usually makes more sense.

Besides the reasons you mentioned, DACs not only generate less heat, they are also marginally lower latency. Arista benchmarked it.

 

[oneplane]

One of the reasons grounding (or shielding) having potential differences was that having a separate earth wire solved it, but I wouldn't be surprised if there are other marginal signal/DC issues at play.

 

[blunden]

One of the reasons grounding (or shielding) having potential differences was that having a separate earth wire solved it, but I wouldn't be surprised if there are other marginal signal/DC issues at play.

Oh, it did? I didn't know that. Did some in this thread say so?

EDIT: I see you change your post slightly compared to what I saw in the email.

Perhaps I should've tried that. Don't know if I have a suitable wire to test with though. Oh well, I have the transceivers now anyway.

 

[blunden]

About a month ago when contacting the AliExpress seller for POEPLUS with some questions regarding DAC support for one of their other switches, I mentioned the issues I had with DACs on this switch (that they also sell a rebranded version of). That led to them contacting the firmware developers, presumably at Hasivo, to try to improve that instead since they claimed it was a firmware issue.

Earlier today I received a new firmware to try, which now includes the following new speed options under Port Setting: DAC50CM, DAC100CM, DAC300CM and DAC500CM


Using the DAC300CM setting with the switch connected with a 2 meter DAC to my Qotom box did indeed stop it from constantly flapping like it did before. Instead, the link was seemingly stable and allowed me to pass traffic over it, which it never did before. I didn't experience any packet loss when pinging a different host on the network over that link.

On the other hand, my Intel X710-DA2 still won't link up to the switch with that same DAC cable. Still, it's promising, even though I'm not sure exactly what those options do in practice.

I uploaded the firmware here if someone else is interested.

 

[blunden]

I think I found more details about what setting those DAC* settings actually does in practice, as described below.

This post in an OpenWrt thread mentioned that the Realtek SDK code contains signal calibration values for different cable lengths:

[...] What does not (yet) work is the 10GBit SFP+ ports on both models and the 10GBit mode on the 3 multi-gbit ports on the XGS1250. The reason is that the 10GBit mode needs a calibration based on the actual cable or fibre used. In the SDK there are even different sets of initial calibrations for direct attach cables of the SFP+ port depending on the cable length (.5m, 1m, 2m, 5m). Code for this is in the SDK but it is complicated and I do not understand it due to my lack of technical knowledge of eye-diagrams and timings for such ethernet speeds.[...]

This site by one of the active developers bringing support for Realtek switch chips to OpenWrt also mentions the need to calibrate the signal when using DACs:

[...] The realtek serdes controllers are quite flexible, which can be understandable, each PCB is designed differently and different parameters are needed to ensure clean signals between the chips. This is even more so important on Direct Attached Copper (DAC) SFP modules, which basically extend the SerDes signals over Copper wires. As such, the serdes needs to be calibrated/tuned to match its use.[...]

 

[gargravarr]

Something I can contribute - I'm using an S600WP-5GT-2SX_SE with a Dell/Amphenol 10Gb 0.5m DAC (P/N 0C6Y7M) to another switch, works no problem.