On 07.11.2014 17:46, Mark Phippard wrote: >> On Nov 7, 2014, at 8:30 AM, Branko Čibej <br...@wandisco.com> wrote: >> >> On 07.11.2014 16:02, Mark Phippard wrote: >>>> On Nov 7, 2014, at 6:46 AM, Branko Čibej <br...@wandisco.com> wrote: >>>> >>>>> On 07.11.2014 14:07, Ivan Zhakov wrote: >>>>> Actually, I have used my veto on the log addressing feature two months >>>>> ago [1]. >>>> How many times do how many people have to explain that saying "-1" >>>> without substantiating that with technical reasons is not a valid veto? >>>> >>>> -- Brane >>> I think it is fair to ask whether it makes sense to introduce a fairly >>> risky feature that adds marginal benefits into a mature and stable area of >>> the code. >> Performance test results seem to imply that the benefits are far from >> marginal. > The bias I bring to the table here is that my view of a "typical" SVN > deployment is an Apache SSL server that is serving at least dozens of > repositories though often it is hundreds. With fairly random access across > those repositories.
This is exactly the kind of deployment that can benefit from FSFSv7: the changes in data layout and other details make cold-cache performance much better than FSFSv6. Performance test results appear to corroborate that; though I don't think anyone actually tested this by running hundreds of clients against dozens of repositories. -- Brane
