I've been seeing some flaky performance in inexpensive switches. I'm seeing
connectivity interrupted for tens of minutes at a time or perhaps until a
reboot of the switch or perhaps until port use is changed.

In one case, the switch is being used as a convenient place to plug into the
downstream feed from a server. Otherwise, the switch is being used with one
port in and one port out - carrying all the server's traffic.

At one port ("upstream") there is but a single destination. On the other
port ("downstream"), all clients on the network appear.

Now, I have a rudimentary understanding of switches. So, I can imagine that
this configuration will cause the switch to store something like a table
that associates a long list of MAC addresses or IP addresses with a single
port. Either that or the switch "forgets" older associations as new traffic
appears from a variety of clients - and is continually updating its
associations on the downstream port. Or, maybe the switch reverts to being
a hub and "acts dumb". I have no idea.

I'm going to assume that the switch keeps a list. If this is the case then
there must be a finite limit on the size of the list. One can imagine a
LIFO stack of MAC associations with finite stack depth. One can imagine a
bug in LIFO management. etc. etc.

Might it be possible for a switch to get confused and stop all
communications and then take a noticeable amount of time to "wake up" or
simply fail altogether? Well, theoretically anything is possible. But
might this be a practical issue that is somewhat common / known in the
community?

Does anyone have experience with this sort of thing?
Any brand / model / bug correlation? It appears something like this could
be happening with a Belkin 8-port switch supporting 12 clients and a router
as described above.

Thanks,

Fred