here are my results of the comparison of the OSGi-R4 implementations: <http://apache-felix.18485.x6.nabble.com/file/n5006870/DiagrammVergleichStartzeit_-_Kopie.png>
I distinguish between a best case and a worst case. The best case is if no package dependencies exist between the specific bundles. The worst case is if each bundle depends on one parent bundle and the parents dependencies. I derived the worst case from the 'Single-Responsible-Principle'. I assume that each bundle provides one functionality and needs all low-level-functionality provieded by the parent bundle and its dependencies. The number of dependencies grows with the number of bundles like triangular numbers (1, 3, 6, 10,...1275). All measurements are performed on a System with a PowerPC processor (266MHz, Single-Core), 128MB RAM and embedded Linux. The slope of the Apache Felix starttime grows linear and two times faster than that of Knopflerfish! Eclipse Equinox is optimized for big Applications with hundreds of bundles. Apache Felix is optimized for very dynamic Applications (frequently installing or updating bundles at runtime). Knopflerfish starts very fast and can handle a lot of bundles in an efficient way. I measured the Data Acquisition rates of a SCADA-System, too: OSGi increases the number of object-instanciations by a classloader during data acquisition up to 30% due to the smaller and limited classpathes of the bundles in comparison to normal java applications (laboratory conditions !!!!!). Under real conditions, you would not get this result, e.g. if a object pool is used. OSGi does not affect the data acquisition in a bad manner! Regards Roland -- View this message in context: http://apache-felix.18485.x6.nabble.com/How-to-improve-the-start-time-of-Apache-Felix-tp5004833p5006870.html Sent from the Apache Felix - Users mailing list archive at Nabble.com. --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
