|
+
+
+ Some tips for running digilib
+
+
+ Java VM settings for digilib
+
+ The Java virtual machine (Java-VM) only uses a fixed amount of memory for
+ its operations. When an operation needs more memory than available it aborts
+ with an error ("out of memory error").
+ digilib can need a lot of memory depending on the size and type of images.
+ Since digilib runs as a servlet under Tomcat its in the same VM as the Tomcat
+ server.
+ The amount of memory Tomcat (version 5.0) uses is configured by creating a
+ "setenv.sh" script with a line
+ CATALINA_OPTS="-Xmx512m"
+
+ in Tomcat's "bin" directory (giving 512MB in this case).
+
+ You can check the amount of memory your digilib instance has available on the
+ bottom of the web page /server/dlConfig.jsp in your digilib instance (e.g. http://digilib.mpiwg-berlin.mpg.de/digitallibrary/server/dlConfig.jsp).
+
+
+ Prescaled images
+
+ You can provide any number of scaled-down versions of your images that
+ digilib uses when a smaller version of an image is requested. Since less data
+ has to be read and processed this can speed up digilib's performance quite a
+ lot.
+ The actual process is that the client requests a certain target size,
+ digilib scans all available scaled-down versions of the same image, selects the
+ smallest image that is larger than the requested size and scales it down to the
+ requested size.
+ The scaled-down versions of the image have to have the same file name as
+ the original hi-res file. They can have a different type and extension (e.g.
+ img002.jpg for img002.TIFF)
+ The scaled down images have to be in the same subdirectory (the part that
+ shows up in digilib's "fn" parameter) as the hi-res file. The first part of each
+ directory tree is configured by the basedir-list parameter in
+ digilib-config.xml.
+ The sequence of directories in basedir-list is from hi-res to lo-res.
+ Images must be present in the hires directory but they need not be present in
+ all lower-res directories.
+ e.g. if digilib-config.xml contains
+ <parameter name="basedir-list" value="/images:/scaled:/thumb" />
+
+ and a user requests the image "books/book1/page0002" digilib looks for
+ /thumb/books/book1/page0002.*
+ /scaled/books/book1/page002.*
+ /images/books/book1/page002.*
+
+ and uses the first image that is bigger than or equal to the requested
+ size.
+
+ For batch-prescaling our images we use a script called "scale-o-mat" that uses a
+ lot of freely available imaging libraries (ImageMagick, libtiff, netpbm) and is
+ available in our public CVS [1]. The script is given a
+ hi-res base directory, a destination base directory, a destination size and a
+ starting directory. It then processes all files in the starting directory and
+ all its subdirectories and creates scaled images in corresponding directories
+ under the destination base directory.
+
+ We currently use prescaled thumbnails of 100 pixels and images for browser
+ display of 1500 pixels. Remember that the prescaled image has to be larger (or
+ the same size) than the requested image size!
+ There is another optimization in digilib: if the requested image is exactly
+ the same size and type as the prescaled image then the prescaled image is sent
+ unmodified to the client which is a lot faster. So it makes sense to produce
+ thumbnails of exactly 90 pixels width when they are used in an HTML page where
+ all images are 90 pixels wide.
+
+
+ Installing JAI
+
+
+ In principle you can install the Java Advanced Imaging JAR files (and
+ native library files if available) in the /WEB-INF/lib/ directory of the digilib
+ web application.
+
+
+ I found that in some cases digilib stopped reading TIFF files after a
+ period of running. In these cases it helped to install the JAI files directly in
+ the local Java installation (i.e. in the Java's /jre/lib/ext/ directory on
+ linux).
+
+
+ Sample setup
+
+ The current digilib setup at the MPIWG (as of November 2005):
+
+ -
+ One frontend server running the lightweight web-multiplexer pound
+ on port 80 that distributes requests to two servers runnning digilib
+
+ -
+ the two servers run digilib under Tomcat on port 8080 without Apache
+
+ - one server is the same machine as the frontend server (Linux, Dual
+ 2.4GHz Xeon, 2GB RAM)
+ - the other server is a separate, newer machine (Linux, Dual 1.8GHz
+ Opteron, 2GB RAM)
+
+
+ -
+ the digilib instances (digilib 1.5.9c as of 22.11.2005) run on Tomcat on Java
+ 1.5.0_05 (or 1.4.2_06) with 1GB of Java VM memory
+ for digilib (-Xmx1024m) with JAI (1.1.2_01) and JAI-ImageIO (1.0_01) installed in the JRE(!)
+
+ - both digilib servers access all image files over NFS (over GBit
+ Ethernet) from a central file server (Solaris 8, Sun Fire 480, multiple RAIDs
+ on Fibrechannel)
+
+
+ |
+ 
-