On Tue, Apr 2, 2013 at 11:45 PM, UkJung Kim
So, my question is that how I can improve the performance of aggregating, if it is possible to replace with link instead of copy?
Checking the source, I noticed bs_sched uses the API to get the packages, with view=cpio. Checking BSHTTP.pm to see how this all worked, there are parts that sound quite sub-optimal. Not sure it's the bottleneck, but for large files or large number of them it might. cpio_sender could be improved by use of sendfile, to avoid creating so many small (8k) chunks and instead creating one big chunk per file (HTTP supports it just fine), or many big ones, but also avoiding a lot of data copying. cpio_receiver could be similarly optimized, but that would require kernels newer than 2.6.33 (version at which to-file sendfile support was added). This could be especially efficient, if done both in sender and receiver, for local (loopback) connections, almost equal in performance to a "cp". Not sure if that is the case for b.o.o, but it sure is for appliances. How significant a win that would be, would require some profiling I'd imagine. -- To unsubscribe, e-mail: opensuse-buildservice+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-buildservice+owner@opensuse.org