[Skimmertalk] My experience with my skimmer system

[Rick Walker]

All,

Just a few notes on my experience with my skimmer system.

First, skimmer version 1.3 is extremely stable.  In version 1.2 I had
constant problems with the application hanging and also with it producing
huge volumes of floating point errors.  My version 1.2 skimmer applications
would produce hundreds of megabytes of error files (some of which I zipped
and sent to Alex).  However, since switching to version 1.3 I have
experienced only one floating point error to date.  That's "one" version 1.3
error in 24 days versus hundreds of version 1.2 errors per week.  The
version 1.3 skimmer applications sit quietly and perform their mission
day-after-day with little to no attention required.  I would give Alex a
major gold star for stability improvement.

Second, the decoding seems to be much more snappy in version 1.3.  I have
not done any sort of formal study to ascertain actual performance metrics
between the two versions, but my general impression is that the latest
version produces much more timely spots.  As to spot accuracy, I have not
been able to test the new version under any sort of significant load, but
users of my skimmer telnet cluster report that the skimmer spots are of
excellent quality.  The addition of the SNR parameter in the remarks field
does allow one to develop a certain amount of initial confidence in the spot
even prior to visiting the spot frequency to apply the "grey matter" decoder
to it.  Another gold star to Alex on this front as well.

I recently switched over to the Wintelnetx application, developed by K1TTT,
to perform the fusion and aggregation function for my six skimmer
applications.  Dave, all I have say about this application is "WOW!!!"  It
is extremely easy to set up and maintain.  The application has the
capability to create and store different .ini files that enable me to
reconfigure it to accomadate changes within my station IT infrastructure.
The auto-reconnect functions work as advertised and quickly restore lost
connections making my system much more robust and "self-healing" than the
application I was using previously.  I highly recommend this application and
give K1TTT a major gold star on it as well.

I have found that skimmer performance is directly correlated to processor
performance.  This obviously comes as no shock, but the take-away point is
that one's enjoyment and utility of the application is marginalized when
running it on machines that aren't equipped to handle the load.  I have
found that my main skimmer server (using Intel Xenon 2.5 GHz dual quad-core
processors) will easily handle six concurrent skimmer applications at 96 kHz
bandwidths.  The system runs at around 25 to 30 percent CPU utilization
during normal days and at 70 to 80 percent utilization on peak loads.
(Note:  At peak loads the system is producing between 1950 and 2000 spots
per hour on average).  During the initial setup and checkout of my server, I
experimented with manually setting the affinity of the six individual
skimmer applications to specific virtual CPUs.  I noticed that in this
configuration the CPU utilization actually went down enough to be
significant (a drop of approximately 5 percent overall), most of which was
associated with less OS time spent in kernel space rather than in
application space.  I tried setting the affinities of version 1.2
executables using the "imagecfg.exe" utility (enabling persistent affinity
on application start-up) but found this wouldn't work.  I was told that this
might be due to the way the skimmer application was compiled so I mention it
in the hopes that future skimmer versions would be compiled to support this
capability.  I have found that a computer using a 2.4 GHz Intel core(tm)2
quad processor will easily handle the load of two SDR-IQ receivers (again at
96 kHz bandwidths) as well as running my main SO2R station configuration.
With the help of George (NW4G) at Gigaparts, I have recently finished
upgrading the computers at my station operating positions to this hardware
baseline.  Another interesting and recent improvement to my station's IT
infrastructure is the addition of a Windows Home Server.  This machine
(sufficiently equipped to handle the data throughput requirements) has
proved to be an excellent addition.  It gives me the capability to access
skimmer replay files from any computer on my network.  I expect this to
significantly improve my ability to store and retrieve data for both
real-time and after-action analysis of station performance.

Speaking of storage and retrieval, one of the most disappointing aspects of
the current skimmer application, in my opinion, is its recording and replay
function.  I have not found a way to break the recordings into smaller
increments than one hour.  At the receiver bandwidths I use, it means that
each replay file is about 2 GB in size.  This makes them impractical to
share and/or move except within the station on the GB LAN.  It would be
beneficial to be able to "tune" the size of the replay files by selection of
smaller recording intervals.  In addition, the replay function would be more
usable if it were to allow me to jump to a specific time via some form of
direct input instead of the "hunt and peck" method I employ with the current
slider bar.  It would be helpful to be able to store the output of the
native skimmer telnet server along side its associated recording file.  This
could serve as a basic "index" of the recording.  Another beneficial
capability would be to enable the placement of "markers" within the replay
files that could be used to highlight items of interest for off-line review
or analysis.  I realize most of this is still "over the horizon," but
hopefully not too far over...

Finally, I, for one, do not utilize the skimmer GUI to any great extent.  In
fact, the only time the skimmer windows ever get expanded is when I am
engaged in some form of diagnostic effort.  I know there has been a great
deal of positive comment regarding the use of a "blind mode" skimmer as a
visual bandmap aid.  Since I use SDR-IQ receivers, I prefer to use the
SpectraVue application to perform visual inspection of a "band."  The SDR-IQ
receiver comes with a software server application that enables remote access
by SpectraVue via a TCP/IP connection.  I am using a gigabit LAN in my
station so this sort of architecture works well and puts an additional
receiver on any desktop in my station.  So I would be interested in some
sort of "reverse blind mode" skimmer that has little in the way of GUI and
devotes more of it's processing time to signal detection and decoding.

I hope each of you who are planning to experiment with skimmer during the
CQWW this weekend have an enjoyable experience.  I am looking forward to
reading comments on the various experiences and "lessons learned" that we
will certainly gather.

73,

Rick
K4TD