sheeva performance graphs

[vds]

I found some nice rrdtool scripts on the wiki at http://computingplugs.com
and installed them (lightly tweaked) to try to get an idea of how hard
the Sheeva is working running wview-5.5.0. Short answer is the Sheeva
is almost idle running wview.

Graphs are at http://skahan.homedns.org:8080/index.html

There's also a link to the weather pages wview creates which are
unaltered other than turning the clouds background off, and putting a
statistics link at the bottom of the current weather page.

fyi - the hourly network write spikes are a cron job scp'ing a gzipped
copy of the archive sdb file to another linux box on the home
network. I rotate the filenames based on the hour of the day, so I
have the last 24 hourly snapshots stashed on the other box, in
addition to the 6+ days of raw data on the Davis VP serial adaptor
just in case the SD card on the Sheeva dies. The read spikes are
every-30-min grabs of a couple files off Internet via wget. Other
than syncing the time and sending the data to wunderground and cwop
there's essentially no network usage.

Not on the graphs (yet) is memory usage. Unfortunately the rrdtool
script I found doesn't initialize the database on the version of
rrdtool I downloaded with apt-get. Getting that working is next on my
list.

Quick look at memory usage with 'top' and 'free' says that there is
very little free memory running wview+apache but almost all of it is
in the buffers+cached category which to me means it's not really out
of memory in the traditional sense so I don't see any problem there
either.

Other than the question of lifetime writes to the SSD disk (or SD card
as I'm doing), the Sheeva almost seems to have too much performance.
Nice problem to have :-)

[Mark S. Teel]

Nice info, thanks.

wview should not ever require much CPU bandwidth during normal
operation. The file generation is somewhat intensive but is short in
duration and periodic. I like the file generation to take 100-200 ms but
it takes more like 6-8 seconds on the old slug platform. Not a big deal
for most people.

Where the 5.X.X versions of wview really taxed the slug were the initial
data conversions from WLK to SQLite. This was brutally long and slow on
the slug. There are ways to do this conversion off the slug then
transfer it over.

Let's be glad that the sheevaplug has plenty of CPU and memory bandwidth
left - by the time others add webcams, lightning detectors, post
processing of data, etc. we'll be glad we are starting out with ample
headroom.

Mark

[Graham Eddy]

anyone aware of where i can get a sheevaplug in australia?
(the listed supplier's website not responding, and phone entry in white
pages no longer present)
------------------------------------------------------------------------
*Graham Eddy*