Tak som to vyskusal, naozaj to spocivalo v tom barrier/nobarrier. Pre uplnost: testovane na 5400 rpm notebook disku Hitachi HTS541616J9SA00, ext3 59% z 15GB plna, xfs particia 93% plna z 11 GB pred testom (tj. prejavilo sa aj seekovanie, na skutocny benchmark by bolo treba rovnake podmienky):
xfs nobarrier: time cp -al ../base.cow .; time sync cp -al ../base.cow . 0.04s user 0.54s system 16% cpu 3.444 total sync 0.00s user 0.04s system 25% cpu 0.144 total cp -al ../base.cow . 0.02s user 0.55s system 17% cpu 3.244 total sync 0.00s user 0.04s system 13% cpu 0.306 total #v dalsom ma uz nacachovane bloky, kde su subory na hardlinkovanie cp -al ../base.cow . 0.06s user 0.45s system 22% cpu 2.221 total sync 0.00s user 0.05s system 23% cpu 0.216 total time rm -rf base.cow ; time sync rm -rf base.cow 0.02s user 0.28s system 10% cpu 2.716 total sync 0.00s user 0.03s system 22% cpu 0.147 total rm -rf base.cow 0.02s user 0.33s system 12% cpu 2.822 total sync 0.00s user 0.05s system 18% cpu 0.256 total xfs barrier: time cp -al ../base.cow .; time sync cp -al ../base.cow . 0.11s user 0.87s system 4% cpu 20.038 total sync 0.00s user 0.05s system 15% cpu 0.327 total time rm -rf base.cow ; time sync rm -rf base.cow 0.02s user 0.62s system 5% cpu 10.872 total sync 0.00s user 0.00s system 0% cpu 0.315 total ext3: cp -al ../base.cow . 0.01s user 0.21s system 28% cpu 0.756 total sync 0.00s user 0.01s system 0% cpu 0.738 total rm -rf base.cow 0.01s user 0.11s system 83% cpu 0.143 total sync 0.00s user 0.00s system 0% cpu 0.394 total cp -al ../base.cow . 0.02s user 0.18s system 96% cpu 0.215 total sync 0.00s user 0.01s system 1% cpu 1.280 total rm -rf base.cow 0.00s user 0.14s system 94% cpu 0.148 total sync 0.00s user 0.00s system 0% cpu 0.644 total cp -al ../base.cow . 0.02s user 0.21s system 99% cpu 0.227 total sync 0.00s user 0.01s system 0% cpu 1.108 total rm -rf base.cow 0.01s user 0.11s system 97% cpu 0.122 total sync 0.00s user 0.00s system 0% cpu 0.868 total cp -al ../base.cow zase 0.03s user 0.19s system 95% cpu 0.226 total sync 0.00s user 0.01s system 0% cpu 1.201 total rm -rf zase 0.01s user 0.14s system 99% cpu 0.145 total sync 0.00s user 0.00s system 0% cpu 0.794 total
Tj. xfs vysiel do 2x pomalsi pri copy, do 3x pomalsi pri rm. BTW ext3 pri mazani poloziek z adresarov trocha "podvadza", pretoze nezmensuje adresare (da sa to jednoducho otestovat vytvorenim adresara s 5000 subormi a potom tie subory zmazat, velkost adresara sa nezmensi). Toto spravanie je v ext3 uz hodne dlho, v podstate neviem, ze ci je to zamerna featura, ale davalo by to zmysel - moc miesta to nezabera a ked sa uz v nejakom adresari casto obmenuju subory, tak je pravdepodobne, ze jeho kapacita sa vyuzije.
