We have several mountain sites with 40+ subscribers each. All running
KarlNet Flash ROMs for base units and a mixture of WG+, NDIS, Ubicom
units, AP-1000's for subscribers. Our base units are 1+Ghz CPU's with
256mb of RAM. Each of our mountain sites are configured in Turbocel ISP
base mode. We have experienced high latency spikes evident in any ping
test you perform from a subscriber's link to the mountain site itself.
The pings would return >10 for 8 or so pings then spike to over 300ms
then return to >10. This "spike" is even evident even when there is
little or no traffic on the site. We have tested turning poling off and
the spikes completely disappear. The only bad thing is the new Ubicom
units are a variant of NDIS client code that requires the base units to
be in ISP mode. ISP mode by default enables poling. If we disable poling
we loose the ability to support the new low cost CPE's. This spike is
across the board, each and every subscriber exhibits the spike.
In our lab we were able to recreate the problem by adding a
subscriber with a poor SNR (to the point where we began receiving
failures in the remote monitor utility) It seems as long as ALL
subscribers are receiving a 12db SNR consistently the "spikes" remain
acceptable at 30 or 40ms. The second ANY subscriber drops below 12db SNR
the latency spike begins to increase. The more unstable the link the
greater the latency, again this latency is exhibited across the board.
It seems that if even one subscriber is receiving a poor wireless link
IT WILL affect the entire subscriber base of that site. Disabling poling
seems to remove this cross susceptibility.
Has anyone seen this? Is it true that a KarlNet site with poling enabled
performs at its least common denominator?
|