Hello all. I am using plain 2.4.7 with ext3-2.4-0.9.4-247, compiled with gcc-2.96-85 on a RH7.1 distribution with the relevant changes, i.e. e2fsprogs-1.22-1, mount-2.11g-4 and util-linux-2.11f-3 taken from rawhide. I am using sometimes my /usr/local partition for ext3 testing: /dev/hdc2 3842408 738176 2909044 21% /usr/local For my DIY testing, I am timing this trivial script which copies, decompresses and deletes the compressed source 2.2.17 kernel of 17563583 bytes (17M). The source partition /dev/hdc4 885468 792616 47872 95% /incoming is on the same disk. #! /bin/bash time cp /incoming/linux-2.2.17.tar.gz . time tar -zxf linux-2.2.17.tar.gz time rm -rf linux time rm -f linux-2.2.17.tar.gz The relevant harware: AMD Athlon 1.2 GHz, 512M ram, hda: QUANTUM FIREBALLP LM20.5, ATA DISK drive hdc: IBM-DJNA-371350, ATA DISK drive hda: 40132503 sectors (20548 MB) w/1900KiB Cache, CHS=2498/255/63, UDMA(66) hdc: 26520480 sectors (13578 MB) w/1966KiB Cache, CHS=26310/16/63, UDMA(33) The typical timing results for ext2: time ./test.sh real 0m0.159s user 0m0.000s copying sys 0m0.140s real 0m2.129s user 0m1.420s decompressing and writing sys 0m0.690s real 0m0.127s user 0m0.000s deleting tree sys 0m0.130s real 0m0.016s user 0m0.000s removing source file sys 0m0.010s real 0m2.439s user 0m1.430s total elapsed sys 0m0.970s i.e. a total of about 2.5" for the whole procedure. Now, for ext3 mounted with default parameters, i.e. data=ordered: time ./test.sh real 0m0.216s user 0m0.000s copying sys 0m0.220s real 0m3.008s user 0m1.530s decompressing and writing sys 0m1.030s real 0m7.658s user 0m0.020s deleting tree sys 0m0.260s real 0m0.026s user 0m0.000s removing source file sys 0m0.020s real 0m10.916s user 0m1.550s total elapsed sys 0m1.530s as you can see, the huge difference is probably due to the fact that, since the journalling mechanism starts a new transaction every 5", there is a good pribability of hitting a new transaction during the decompression. I'm not an expert, just a user. Resorting to data=writeback the total timing is around 3", quite close to ext2 timings. time ./test.sh real 0m0.186s user 0m0.000s copying sys 0m0.190s real 0m2.528s user 0m1.510s decompressing and writing sys 0m0.950s real 0m0.264s user 0m0.020s deleting tree sys 0m0.230s real 0m0.021s user 0m0.000s removing source file sys 0m0.010s real 0m3.006s user 0m1.530s total elapsed sys 0m1.380s In my opinion, if you're not working on large files, something of the order of, say, 50-100M on a modern machine (i.e. how much you write in less than about 5") perhaps you'll gain a lot by mounting data=writeback . Comments? I hope my small test could be useful. Sicnerely, Nicola Fabiano