Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

[8023-POEP] My thoughts on L2 protocol for PoE



Guys,

 

I have some thoughts regarding the proposed Layer 2 power management scheme.  Basically, the more I think about it, the more I’m convinced the midspans need to implement L2, not just the endspans.

 

The reason is that, if dynamic power negotiation is actually going to be used, then both PSE need L2 capability because endspan-only negotiation seems too limited to be useful.  So I think we either mandate that the midspans also do L2, or we drop the dynamic power negotiation idea all together.

 

Of course the problem is cost.  I did my own crude cost estimates, and here is what I came up with.  I estimated the total costs of a 12-port 802.3af midspan to break down as follows:

 

  1. Main power supply                                50%
  2. Housing                                                15%
  3. Bare boards                                           7%
  4. Labor (assy & test)                                13%
  5. Port circuitry, including magnetics          12%
  6. Other circuitry (microprocessor, etc.)       3%

 

Obviously, I don’t make midspans, so I’m guessing.  But I think these numbers are in the ball park.  (If anyone thinks these are way off, please let me know.  I’m interested.)

 

Next, I looked at what it would cost to add a PHY/MAC and 10/100 magnetics to each port, and an Ethernet switch so that the micro could talk to all ports.  I estimate the additional circuitry would cost about as much as the existing port circuitry (PSE controller, associated discretes, magnetics for power injection).  So item 5 above is essentially doubled, which would increase the total system cost by approx 12%.  But this doesn’t count the additonal assy or test labour.  So the actual impact may be more like 14%.

 

But I have an idea how we could add L2 capability to midspans without extra hardware.  Suppose the PD negotiates with the midspan indirectly.  Packets from the PD would pass thru the endspan into the LAN, and then into the management port on the midspan.  All PDs connected to the midspan would take the same path, so just one Ethernet management port is required instead of one PHY/MAC for each midspan port.

 

The problem is that we loose some info this way.  The port associations, which are inherant if every midspan port has a PHY, are lost in this scheme.  For example, if we don’t know that port-N on the endspan is powering the same PD as port-M on the midspan, then we can’t do dynamic power negotiation.  The endspan could tell the midspan via the management port “some PD needs 5 more Watts”, and the midspan could allocate the additional power, but if the midspan port is unkown, then how would it know when to deallocate the extra negotiated power when the PD is disconnected?  How would the midspan do power-policing if it doesn’t know which PDs have power allocations different from their L1 classification levels?

 

So my idea is, somehow encapsulate the port mappings into the MIB.  For example, the sys admin could enter (via command line or GUI) the info as the midspan-to-endspan cables are connected during intial system setup.  When a PD requests more power from endspan port-N, the N=M association stored in the MIB would tell the endspan to copy the packets to the midspan via the management port, and the midspan would know that it’s talking about port-M.

 

Any comments?

 

Steve