Context: free cloud servers, a workshop and an idea

In this post I look at some work we’ve been doing at the University of Western Sydney eResearch group on visualizing metadata about research data, in a geographical context. The goal is to build a data discovery service; one interface we’re exploring is the ability to ‘fly’ around Google Earth looking for data, from Research Data Australia (RDA). For example, a researcher could follow a major river and see what data collections there are along its course that might be of (re-)use. True, you can search the RDA site by dragging a marker on a map but this experiment is a more immersive approach to exploring the same data.

The post is a quick update on a work in progress, with some not very original reflections on the use of cloud servers. I am putting it here on my own blog first, will do a human-readable summary over at UWS soon, any suggestions or questions welcome.

You can try this out if you have Google Earth by downloading a KML file. This is a demo service only – let us know how you go.

This work was inspired by a workshop on cloud computing: this week Andrew (Alf) Leahy and I  attended  a NeCTAR and Australian National Data Service (ANDS) one day event, along with several UWS staff. The unstoppable David Flanders from ANDS asked us to run a ‘dojo’, giving technically proficient researchers and eResearch collaborators a hand-on experience with the NeCTAR research cloud, where all Australian University researchers with access to the Australian Access Federation login system are entitled to run free cloud-hosted virtual servers. Free servers! Not to mention post-workshop beer[^[i]^](#_edn1). So senseis Alf and and PT  worked with a small group of ‘black belts’ in a workshop loosely focused on geo-spatial data. Our idea was “Visualizing the distribution of data collections in Research Data Australia using Google Earth”[^[ii]^](#_edn2). We’d been working on a demo of how this might be done for a few days, which we more-or less got running on the train from the Blue Mountains in to Sydney Uni in the morning.

The demo

We will work some more on this little very part-time project, but here’s a snapshot of where we are at the moment with our mission of getting RDA data into Google Earth.

Figure 1 Screenshot of data sets from Research Data Australia mapped onto the earth (visualized via their bounding boxes from the data collection metadata) from a long way up (the sets with the (claimed) broadest geographical coverage show up first, more appear as you zoom in)

Right now the experience is basic, but engaging, just using the geometry provided in RDA collections and without changing any of the default ANDS styles you can see a ‘heat map’ of where data have been collected. Speaking of heat it seems Aussie researchers prefer the more northerly latitudes to the South pole, as a place to take a few readings, or catch a few rays.


Figure 2 Zooming in on research hotspots like The Reef shows how crowded it is there – we’re limiting this to 40 boxes or points per screen, so keep drilling to find more

Figure 3 You can click on a marker to pull up a description


Why are we doing this?

  • To test out the NeCTAR Research Cloud, so we can advise researchers at UWS on how and when to use it – we’ll cover this in future posts and publish advice on the UWS eResearch site. Initial observations include:

    • It’s good for testing out ideas quickly – you can fire up a tool, pull in data from a networked source and do something to it, but some planning is required to package the tools and get them to operate as turnkey appliances. At the moment you need command-line skills and patience.

    • This infrastructure is a game-changer – it can take months for university IT departments to deliver a virtual machine to a research or dev group and policy prevents many of us from buying commercial cloud services on corporate credit cards.

    • The lack of easy-to access persistent cloud storage which had not yet come on line from NeCTAR’s sister project Research Data Storage Infrastructure RSDSI is a bit of a problem, there is no simple way to mount a device with all your data on it.

    •  A specific lesson from our demo: being able to fire up servers on demand to handle peak loads, such as building a big geo-spatial index (we only built a tiny one) seems promising but we’d like to see this made as easy as possible for people who want to get research done rather than mess with computers.

  • To see how Alf’s considerable expertise in practical, easy to deploy low-end visualization combined with my experience in repositories, harvesting metadata etc might be applied to resource discovery. Being able to fly around the landscape means you can follow geographical features and maybe discover useful data sets along the way.

  • To explore how the ReDBOX Research Data Catalogue application might be able to geo-index collections (see notes on this below).

The technical stuff

The current architecture of our demo is, to say the least, baroque. There is a lot of backwards and forwards between systems to get to the interaction between a user and Google Earth (bottom right of the diagram). That’s perfect for exploring cloud services but maybe not ideal for a ‘real’ service. It does show, however that using existing systems with open interfaces you can stitch something together fairly quickly. This will improve as packaged eResearch appliances start appearing at NeCTAR.

We have now deployed part of this on our own server so we can leave it running and shut down the NeCTAR servers so they don’t use up our allowances.

Key to the diagram:

  • ReDBox is a research data catalogue application which I helped design.

  • GeoHarvester is a  python script ( that sucks data out of the Solr server in ReDBox and turns it into something that another Solr set up for geospatial queries can index. You can see the Solr index here if you’re interested via the default search interface. Or try a geo-spatial query, which could lead you to some KML.

  • KMLServer is a little web app that fields queries from Google Earth (what data sets are ‘in view’) and talks to the Solr server. Alf has coded it to give you stuff from the Great Barrier Reef by default.

Note that the only user-facing system is Google Earth all the rest is back-end stuff. None of the work we did needed any kind of user interface at all.

If  you’re really brave, you can check out the code and try to get it to run by following the readme file. Warning, there are hard-coded URLS everywhere that one bit talks to another that you WILL need to change. I’ll help if you email me but this is not a packaged application, it’s a demo.

Description: ReDBox -> RDA : Fetch collection data over\nOAI-PMH protocol in RIF-CS format ReDBox -> ReDBox : Index data into Solr index GeoHarvester -> ReDBox : Fetch all data with\ngeo-information from ReBOX Solr index ReDBox -> GeoHarvester : Return paginated JSON data GeoHarvester -> RDA : Fetch KML for each collection \n(not included in the RIF-CS) RDA -> GeoHarvester : Return KML (with default style) GeoHarvester -> KMLServer/Solr : Index collections and cache KML GeoHarvester -> KMLServer/Solr : Divide bounding-boxes\ninto points and index each point create KML KMLServer/Solr -> KML : Create KML file\nwith a dynamic data loader GoogleEarth -> KML : Load file User -> GoogleEarth : Change view GoogleEarth -> KMLServer/Solr : What collections are in view? KMLServer/Solr -> GoogleEarth : Return KML file with all in-view\nbounding boxes and\ncollection descriptions GoogleEarth -> User : <collections drawn on globe>

Figure 4 We could possibly have made this more complicated, but it’s hard to see how

Building this first-take has taken us a step closer to our goal of having an immersive demo for exploring data collections, ready for eResearch Australasia. Along the way we learned a few lessons.

  • Apache Solr (which is the text/data Indexer in both ReDBOX and RDA) has basic but working geo-search features. You can only have one point per index entry so to index large shapes we approximated them to a simple box, then filled in a set of points (at this stage chunking by whole degrees), indexing each point individually with a reference back to the container object. Solr’s faceting feature makes it easy to find the biggest boxes first and prevent Google earth from trying to draw 20,000 collections at once.

  • Yes, RDA has a search by bounding box, but it’s slow, we don’t want to barrage it with thousands of hits, and it returns multiple HTML pages. By indexing the data ourselves we get exactly what we want.

  • Yes, ReDBOX has Solr inside but it’s too old a version to support geo-searching well, and we’d also require some non-trivial changes to allow the aforementioned indexing of multiple points etc. It’s probably better to wait a bit for Solr to improve, which it does inexorably thanks to the Solr community, before trying to add this kind of search to ReDBox and the underlying Fascinator platform.

As usual, there’s lots TODO:

  • Make the KML styles customizable so we can provide better visualizations – we’d like to be able to fly through a data mountain-range or city or forest or something. Likewise, improve the heat-map that suggests where data might be concentrated.

  • Get rid of the Google Earth desktop app and use the new web version.

  • Get rid of the GeoHarvester code, by building it into a version of the existing ReDBOX harvester or take ReDBOX out of the tool chain and harvest directly from RDA.

  • Allow search by subject or keyword to select just some of the data.

  • See if we can auto-fetch actual data rather than just location metadata (especially maps and tracks).

  • Index some other data sources – which Alf is busy sourcing from UWS researchers.

  • But above all, see if this is useful enough to researchers or the eReseach/ANDS community to continue the experiment.


[^[i]^](#_ednref1) My partner said “Beer! Is it free?”, I said “Yes”, and she said “What, as in beer?”, she’ll be using Linux next.

[^[ii]^](#_ednref2) As it turned out we didn’t look at this demo in our Dojo, instead we tackled getting a resource hungry desktop application called Gplates, which does interactive visualization of plate-tectonics, running on the cloud. We got it going, but didn’t solve the problem of how to move the heavy number crunching to the cloud and have the result display back at the desk.

Copyright Peter Sefton 2012. Licensed under Creative Commons Attribution-Share Alike 2.5 Australia.


comments powered by Disqus