I am trying to be open about everything so here it goes.
OK first off I apologize for the delay on the release of this switch but I did not want to release something that is known faulty.
We are narrowing this problem down and making headway though!
FIRST we thought it was related to the switch MAC table which was incorrect, dead end there and a loss of several days.
SECOND we thought it was related to the switch VLAN configuration which was also incorrect.
We thought these things because originally we were using all 24 ports and did not make to connection that it was port based but assumed it was VLAN based, long story....
We have been working with the chipset manufacturer to resolve the issue and we have a conference call Monday again with them.
What we have discovered is if we only use the first 12 ports of the 24 port switch it works fine.
There is a second PHY chip to handle ports 13-24. And if we use ports 13-24 without POE there is also no problem.
When we use the higher ports the problem exhibits itself randomly if we are using POE.
Now the PHY's in the SOC for ports 1-12 are Luton and the PHY's in the second PHY chip for ports 13-24 are ATOM, both are made by VITESSE.
Also it only happens when the switch powers devices with the high ports ports 13-24 from the switch.
The POE circuits are identical across 1-24 so it is not the way the POE is done as it works fine on 1-12. It has to be something in the way the high PHY's may handle noise, ESD, EMF, or ground potential differences? We also had designed in but not populated filter circuits for the Ethernet ports.
If we use power bricks on ports 13-24 then it is also all fine keeping in mind that 10/100 UBNT POE bricks do NOT pass pins 4,5,7,8 through so we are looking at the fact that UBNT bonds the digital negative to earth ground in their devices and since they do this we may be forced to as well?
We "think" it has something to do with one of the following:1) Ubiquiti bonds digital negative to earth ground in all their 10/100 devices like ToughSWITCH, Rocket, and so on and we do not.
2) ESD or stray current from above from 150 feet of cable in the air or the fact that the Rockets have bonded neg to ground.
3) there are some software fixes for later revisions of the PHY's (13-24) that we may not be implementing, will find out Monday.
Possible solutions for corresponding issues above:1) Bond them the same as Ubiquiti, we did leave discrete population points for CAPs or 0 ohm resistors to do just that, we may try this today.
2) Possibly same as 1
3) DUH
We may be able to start shipping the 12 port since it does not appear to suffer the same issue.
But as you can see in the picture below I ran the switch all night and we only lost ONE ping to a customer when using only ports 1-12.
When using the TS we lost a considerable amount of pings, mind you less than 0.01% on average but our switch appears to preform much better but one would assume this since the TS BCM53118 switch core was meant for small home or office environment and our core was designed for enterprise.
Now the edgeMAX has a core that is also suitable for Enterprise installations as well and the way they implement the feature is a better solution for an Enterprise deployment but we feel ours is better for tower locations because of the higher operating temperature and the fact that we did not allow all the advanced Enterprise features that 99% of WISPs do not need at a tower site and just make the configuration more complex.
- vlan pings.png (103.33 KiB) Viewed 25051 times
I am sooo jealous!