We have the same exact problem on ALL our mountain sites. We currently have
five mountain sites with 40+ subscribers each. All are running Karl Net
flash ROM units with at least 1 GHz CPUs and 256mb of system RAM. We have
been running with 40+ clients on at least two of the sites for over three
years with no "ping jitter" whatsoever. All of our sites are monitored and
logged via SMTP. The "ping jitter" problem appeared on our sites once we
moved from the 3.x version (3.88) to the 4.x (4.43) version. Once we noticed
this we began a small lab test to narrow down where the problem resides.
What we found was polling was the cause. Whenever we would disable the
polling functionality of the base unit the ping times would return to the
normal 10ms to 30ms range. Once we enabled pooling the ping times would
change dramatically from 10ms to over 400ms. Poling by itself would increase
the latency, then when we created some weak SNR signals to our test lab the
latency would increase dramatically. It seems that the pooling of the base
station is effected by the SNR of each subscriber. Once you get several weak
subscribers (15db and under) the latency seems to increase exponentially.
This problem does not occur when we flash our lab unit with the older 3.88
version. The latency remains consistent between 10 and 40 ms. it is affected
by poor SNR bur not as dramatically as the 4.x versions. It would seem to
eliminate the "ping jitter" you have two choices, one "turn of poling" this
fixes the problem immediately but then you loose the ability to support the
NDIS version of the Karl Net client software (this includes the new Ubicom
units since they are a variant of the NDIS client). Or downgrade to the 3.x
version and loose a whole bunch of new base software features. At this time
we are doing everything we can to maximize SNR on all our subscribers to at
least reduce the latency and wait for Karl Net to address the issue in a
future BIN release.
Timothy Steele
e-ISCO Internet
MCSE, MCP, MCI
tsteele@e-isco.com
www.e-isco.com
|