First off, thanks a lot for this detailed feedback! The Ubuntu packages in the GeoNode PPA are for a pre-release version of the upcoming release. There is not much difference between them apart from a few small bugfixes; I don't anticipate much of a performance difference.
The most effective thing you can do to make GeoNode work better in a resource-constrained environment is to reduce the number of worker threads for Tomcat. This will cause GeoServer to attempt fewer tasks in parallel, reducing RAM and CPU usage. http://docs.geonode.org/en/latest/deploy/production.html#constrain-geoserver-worker-threads discusses how to modify this setting, although I would try an even lower value (10 or 20 instead of the 50 listed there). -- David Winslow OpenGeo - http://opengeo.org/ On Wed, Feb 1, 2012 at 8:20 AM, Reinier Battenberg < [email protected]> wrote: > Hi, > > Under the Oil for Development program, NORAD is helping the Ugandan > Environmental sector to monitor the changes in the environment that are > related to the recent findings of oil in some of the most sensitive areas > of > Uganda. > > Last week there was a training with GIS profesionals from the Forest > Authority, Ministry of Water, Environment Authority etc. About 20 people > participated. Ragnvald Larsen from NORAD, Ketty Adoch and myself from > Mountbatten shared our Geonode knowledge. > > Instead of the recommended server with 6GB of memory, we opted for an atom > based portable media center PC with 2GB of memory. > http://www.advance.no/Item.aspx?id=4901512 > We installed the current geonode release from the apt-packages. > We connected everyone to a local LAN with a wireless router and started. > The > router was connected to a (very slow, maybe 256k) internet connection. > > With this setup, we experienced the following issues. We are aware that > there > is a brandnew version soon to be released, and that some of the issues > mentioned could very well have been solved already. Still, for people who > would like to train in similar conditions, this might be worthwhile: > > - Not suprisingly, we had performance issues. We had most performance > issues > on the first day. Especially in the morning, when we althogether uploaded > shapefiles. Processortime on the server maxed out at 400% and kept there > for a > long long time. A second factor might have been the slow internet > connection > and the overload of timed out http sessions to the Internet and the > Geoserver > that used all resources on the Wireless Router. The router didnt have any > statistics, so this was not so straightforward to diagnose. (tip: use WRT > on > your wireless router) > After lunch, when we started on creating maps and styling them, the system > became much more responsive. > > - During the styling session people created their own maps, and started > styling layers. During this exercise at some point, layers that where > shared > between maps got corrupted. People might have edited eachothers styles > (they > looked very similar), but what surely happened was that 2 layers lost their > default SLD in geoserver. This renders some pages in geonode useless and > causes all sorts of errors. Going into geoserver and assigning a (pretty > random) default style to the layer fixes this. > > - On other geonode installations we had already seen that sometimes when an > upload fails, a geonode layer might be in the geonode database, but is not > present in geoserver (the whole table is missing). I think Jude is working > on > some code to straighten this out when it happens. Would be great if an > admin/user could do this from the geonode gui. > > - On the second day, we brought a second server. It ran fine for about 30 > minutes, then there was a powersurge (yeah, those happen in Uganda.) After > the > power came back, tomcat would not finish starting. The attached > catalina.out > sample shows the full startup sequence. The server never finishes indexing. > Could it be that: > -- the indexing is not 100% necessary during training? > -- the indexing is causing the very high load during uploading data > sessions? > If so, is there a way to switch off the indexing? That would save a lot of > trainers in locations where you can not use "the cloud" during training > sessions a lot of headaches of running around with server-size hardware. > > Enough about problems. At the end of the second day, we workshopped around > uses of geonode, and one of the questions asked was: How could geonode be > improved? > > The top 5 was: > > 1. More data analysis tools > 2. display attribute table > 3. enable layer grouping > 4. connection to social sites like facebook and linkedIn > 5. minimise error bugs. > > And in random order the rest of the list was: > > Improve tools for styling > more admin control in user editing > application compatibility > include a timeline for map edits > include print layout > ability to download backdrop for a region(facilitates for offline) > use standard GIS terms > make it easier to make groups > help feature for frequently asked questions > having the legend directly on the map. > > These are the raw items, they also tell something about the training we > did. > We did not cover the printing part, some people might have missed the > current > legend, we later showed some advanced users how to load peoples own styled > WMS > layers in ArcGIS & QGIS (to the exitement of the trainees). > > And, while at it, maybe i could at 2 of my personal favorites (Withouth > knowledge of 1.2): > - Import Geoserver layers into Geonode. This is especially useful for very > big > geotiffs and for remote WFS layers. > - Slightly more userfriendly errormessages would be great. I know Tomcat > spits > out lists and lists of errors, but having a slight clue of where in the > process things broke helps diagnose issues a lot. > > In general, we love Geonode. Its very relevant in the Ugandan context. A > lot > of the discussion here is still about the actual sharing of data. There is > a > great level of reluctance. Using Geonode to show how easy it can become to > mix > & match data makes the advantage of sharing your data a lot more concrete > for > people. > > Commercial drilling for oil is not expected to start before 2017. > > until then, > > Reinier Battenberg > > > > >
