Not the usual ethernet problem on PINE64 (solved)

[winterflaw]

(One liner added later : the problem turned out to be the Cisco SOHO switch.  They're "broken from the factory".  Not what you'd expect, normally or in particular from Cisco.)

 

Hi, everyone.

 

I've taken the liberty of posting here with regard to my problem as I have the understanding the PINE64 forums proper are run by crazy people; my thought being to stay away from there.

 

So, I have a 1 GB PINE64.  I have ethernet performance problems - but NOT, I believe, in any way related to the early production models with their hardware fault or with gigabit ethernet being flakey.

 

My PINE64 is in fact connected to a 100mbit hub, and is as such running a 100mbit, not 1gbit, and it a fairly later model (as my original order was accidentally sent the wrong continent, and it took some time for this to be sorted out =-)

 

I'm testing performance using nuttcp.

 

This is a representative result;

   10.9375 MB /   1.00 sec =   91.7470 Mbps     0 retrans
   11.2500 MB /   1.00 sec =   94.3755 Mbps     0 retrans
   11.1875 MB /   1.00 sec =   93.8451 Mbps     0 retrans
   11.2500 MB /   1.00 sec =   94.3732 Mbps     0 retrans
   11.1875 MB /   1.00 sec =   93.8461 Mbps     0 retrans
   11.1875 MB /   1.00 sec =   93.8466 Mbps     0 retrans
   11.2500 MB /   1.00 sec =   94.3731 Mbps     0 retrans
    5.1875 MB /   1.00 sec =   43.5160 Mbps     0 retrans
    0.0000 MB /   1.00 sec =    0.0000 Mbps     0 retrans
    0.0000 MB /   1.00 sec =    0.0000 Mbps     0 retrans
    0.0000 MB /   1.00 sec =    0.0000 Mbps     0 retrans
    0.0000 MB /   1.00 sec =    0.0000 Mbps     0 retrans
    0.0000 MB /   1.00 sec =    0.0000 Mbps     0 retrans
    0.1875 MB /   1.00 sec =    1.5729 Mbps     0 retrans

I've tried changing the cable and I've tried changing the port and I've tried changing the power supply (I have a USB multimeter, and the load is fine - much less than the 2 amp the socket is rated for).

 

I also have a Raspberry Pi 2 and a Createo Ci20 plugged into the same hub, and they work fine - the Pi shows a consistent 93/94, with no drop outs, the Ci20 about 70 (which is also odd, but seems to be the way it is on that board).

 

I also see these results on the PINE64 on the Debian distro.

 

Googling for other people's experiences is hard because you can't get away from all the posts about the gigabit ethernet problems.

 

Does anyone have any advice?

 

 

 

 

 

 

 

[winterflaw]

UDP performance is perfect, BTW - 90mb/sec sustained no problem.

 

So I'm guessing now this is something to do with TCP settings.

 

This of course opens up a massive can of worms.  Where to even begin?

 

[tkaiser]

3 minutes ago, winterflaw said:

Where to even begin?

 

Honestly I don't know. Might be ten years ago that we banned Fast Ethernet equipment here. This is definitely one of the problems I would try to solve by throwing some hardware at it. Eg. the cheapest TP-Link Gigabit Ethernet switch with the 'smart managed' buzzwords in its name (since then you can at least (mis)configure settings with a web browser  -- a friend recommends TL-SG108E, they had them as desktop switches at the office, start to replace them now and he managed to implant a NanoPi NEO 2 into the switch enclosure trying to convert these switches to WAN/LAN routers since VLAN capable)

[winterflaw]

It turned out to be the Cisco switch.

 

Cisco make a range of SOHO products.  I can't speak very much for the products in that range outside of the line which I bought from, but I bought an SG100D-05 and it turns out they are basically, to quote one comment I read, "broken from the factory".  This is a far description, in fact.  The switch was loosing contact with the dev boards plugged into it (and needed to be rebooted) was unable to communicate normally with the PINE64.  There are plenty of other problems which you can find by Googling - including, for example (although I can't remember if it's my model or the ones in the next range up from this, the SSG220s) blocking all SYN packets on ports <= 1024 (which is staggering for an *unmanaged level 2 switch*).  I think Cisco never did release a working firmware which fixed the problem - a broken firmware came out after about two years...(!)

 

If you're going to buy Cisco SOHO, make bloody sure you have a good Google first, because what you find will probably convince you not to buy Cisco SOHO.  That was surprising, of course - I *assumed* if it's Cisco, it's going to be solid kit.  It looks however like they just don't take the SOHO market seriously.

 

[tkaiser]

23 hours ago, winterflaw said:

It turned out to be the Cisco switch.

 

It would be nice if you adjust thread topic (adding 'Solved' to it) and edit your very first posting inserting a single line at the top descibing the problem 'switch fault, check your switch first' or something like that). Might save other people some time googling around...