I am in favor of a regular layer 2 uplink.  I think the main reason I feel this way is it is industry standard, and then the switch has its own forwarding table.  We have seen issues in the past where if the link on a high volume port goes down, the MAC gets removed from the forwarding table, and then all of that traffic gets flooded to all other ports until BGP times out.  This seems to be especially problematic when it hits the 1G ports.

My vision of a future upgrade would be something along the lines of a 5200, or equiv from another manufacturer that has all of the other switches uplinked to it.  Then we would have separate forwarding tables, a place to connect remote switches, as well as something that has some 40G, and 100G capabilities.  All of the uplinks would then be easy to monitor/graph without excessive load on the switch.  Copper SFP+/QSFP+ cables can be used between switches and they are affordable/easy to come by.

Jeremy

-----Original Message-----
From: MICE Discuss [mailto:[log in to unmask]IPHOUSE.NET] On Behalf Of Doug McIntyre
Sent: Tuesday, September 20, 2016 3:15 PM
To: [log in to unmask]
Subject: Re: [MICE-DISCUSS] EX4200 (1G Switch) Packet Loss -> 10G Participants

On Tue, Sep 20, 2016 at 02:23:00PM -0500, Jeremy Lumby wrote:
> That would make sense, then we are just down to how traffic can leave the 4200 without going across a vcp

Most likely we'd have to kick off (umm. move) the three members that
are fiber 1G connections in the expansion slot on the 4200, get a 10G
card, optics for both sides, etc.  Then decide if we are going to run
virtual chassis protocol, or just layer-2 uplink into the two other switches.

Or we move the 19 active 1G copper members + 3 1G fiber members off the 4200
altogether and eat up ports on the 4550 with 19 copper SFPs. Then we wouldn't
have to do "Mixed-mode Virtual-Chassis" then either.

> Although- it looks like so far all the paths with loss seem to have "vcp-1" in them?

As I've stated multiple times, I think since VCP is active/passive,
there is only one active ring around the the VC through VCP-1 now,
and I would expect to see vcp-1 in all paths through the EX4200.

We can swap active and passive with the command I posted previously.
That way we can troubleshoot if there is a port/cable issue, or if it is
an issue with the device itself.

Or it could be a defect in this coderev (although I doubt it).


--
Doug McIntyre                            <[log in to unmask]>
                    ~.~ ipHouse ~.~
       Network Engineer/Provisioning/Jack of all Trades