Evgeny Kotkov <[email protected]> writes: > I'll rerun my tests (try /head for the FreeBSD repository, test the high load > situation, etc.) and will update my vote — hopefully, in the next couple of > hours.
I ran the tests one more time — against a local server, against a remote Microsoft Azure server and between two remote Azure servers located in Northern and Western Europe. All tests were performed over http://. I tried different repository paths in the repositories that I had nearby; they include /, /head, /stable in FreeBSD repository, /trunk in TortoiseSVN repository, /avalon/branches/, /httpd/httpd/branches/2.0.x in a part of the ASF repository. I also permutated other possible test conditions, such as eating all available CPU resources on the server, using --xml --with-no-revprops, enabling and disabling authorization checks and restarting the server machine. I did not see any difference for the tests that were run against a remote server (when I tested a remote Azure server from my workstation and when I tested one Azure server from another). Here are the sample timings for 'svn log' commands when they were executed from one server to another. Target server was idle, and the log was done against /head of the FreeBSD repository: Unpatched: 171 ms 187 ms 206 ms 218 ms 172 ms 174 ms 183 ms 187 ms Patched: 218 ms 203 ms 172 ms 171 ms 218 ms 195 ms 203 ms 171 ms When I tried the same with 100% CPU load on the target server, I got the following results. The numbers aren't exactly stable, but, again, I cannot say that I see the difference: Unpatched: 959 ms 189 ms 162 ms 290 ms 306 ms 808 ms 316 ms 634 ms Patched: 560 ms 773 ms 410 ms 242 ms 191 ms 262 ms 472 ms 987 ms I thought that the difference could show up for a local server when the log command is being executed right after the reboot — that is, when no relevant data is in the HDD cache. So, I tried it, but received controversial results, i.e., sometimes the patched version was faster, sometimes not. For example, for /asf/avalon/branches the results could look like 734 ms (unpatched) vs 434 ms (patched), 350 ms (unpatched) vs 591 ms (patched) and 641 ms (unpatched) vs 709 ms (patched) in different test runs, where I rebooted the machine every time before running 'svn log'. Sorry, guys, but based on these observations, I am going to keep my current vote unchanged. Regards, Evgeny Kotkov
