The real performance drivers behind XML lock protocols

Abstract

Fine-grained lock protocols should allow for highly concurrent transaction processing on XML document trees, which is addressed by the taDOM lock protocol family enabling specific lock modes and lock granules adjusted to the various XML processing models. We have already proved its operational flexibility and performance superiority when compared to competitor protocols. Here, we outline our experiences gained during the implementation and optimization of these protocols. We figure out their performance drivers to maximize throughput while keeping the response times at an acceptable level and perfectly exploiting the advantages of our tailor-made lock protocols for XML trees. Because we have implemented all options and alternatives in our prototype system XTC, benchmark runs for all "drivers" allow for comparisons in identical environments and illustrate the benefit of all implementation decisions. Finally, they reveal that careful lock protocol optimization pays off. © 2009 Springer Berlin Heidelberg.