> I will endeavor to re-run these comparisons with RELENG_7 within the > near future (regardless of the answer to the above question). So to follow-up, with RELENG_7 the performance is now significantly better than 6.2 for me too. I guess it was the debugging options in userland. If someone wants details anyway I'll provide them, but since it is no longer that interesting I'm leaving it as is if no one speaks up. I knew about it and wanted to turn it off, but I was under the impression it was enabled by malloc.conf, which was non-existent, rather than at compile time. Also the phrasing in the recommentation to turn things off was such that it did not necessarily imply that malloc debugging was actually turned on. Or I believed so anyway in my possibly dream state. Would it be useful to keep a list of *all* things one needs to do to turn off all the debugging options and whatnot? I have always disabled the INVARIANTS/WITNESS stuff, but was unaware of the details of how to turn off malloc debugging (and I did actively look for more information on it, but obviously not hard enough). In the interest of benchmarks that people publish perhaps it's worth it to spoon feed instruction for what needs to be done, as part of the usual warning in /usr/src/UPDATING? Regarding the time source choice, I attempted to switch to TSC in my original benchmarks but never got it working. I tried all the following (one at a time), but none had any effect: kern.timecounter.tc.TSC.quality="1000" kern.timecounter.hardware="TSC" kern.timecounter.choice="TSC(1200) ACPI-fast(1000) i8254(0) dummy(-1000000)" kern.timecounter.smp_tsc="1" Perhaps because it's not an SMP machine. -- / Peter Schuller PGP userID: 0xE9758B7D or 'Peter Schuller <peter.schuller_at_infidyne.com>' Key retrieval: Send an E-Mail to getpgpkey_at_scode.org E-Mail: peter.schuller_at_infidyne.com Web: http://www.scode.org
This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:39:19 UTC