From mboxrd@z Thu Jan 1 00:00:00 1970 From: Chris Mason Subject: Re: btrfs vs ext4 benchmark Date: Tue, 07 Apr 2009 08:10:35 -0400 Message-ID: <1239106235.24527.5.camel@think.oraclecorp.com> References: <001001c9b52a$8ef00bc0$acd02340$@com> <897274600904070447s276a53c7teb5afbfb6ff95783@mail.gmail.com> Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Cc: linux-btrfs@vger.kernel.org To: "Morten P.D. Stevens" Return-path: In-Reply-To: <897274600904070447s276a53c7teb5afbfb6ff95783@mail.gmail.com> List-ID: On Tue, 2009-04-07 at 13:47 +0200, Morten P.D. Stevens wrote: > Update: >=20 > And here the compilebench with options -i 90 and -r 150 (the same as = oracle) >=20 The difference in initial create times is pretty surprising. My guess is that single spindle duplication of metadata hurts more on this raid array. =46or the other times compilebench was actually written to make btrfs l= ook bad. The main goal was to age and fragment the metadata, while constantly clearing out the caches used to make things fast. IOW, it was meant to be a worst case ;) Some of these numbers look lik= e the cache clearing wasn't hitting ext4. I'll rerun the test against ext4 next week. -chris > btrfs: >=20 > ./compilebench -D /btrfs/work2 -i 90 -r 150 >=20 > run complete: > =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D > intial create total runs 90 avg 45.18 MB/s (user 0.84s sys 2.89s) > create total runs 20 avg 13.30 MB/s (user 0.88s sys 2.93s) > patch total runs 24 avg 5.94 MB/s (user 0.34s sys 2.56s) > compile total runs 23 avg 37.04 MB/s (user 0.19s sys 1.60s) > clean total runs 14 avg 71.36 MB/s (user 0.03s sys 0.83s) > read tree total runs 17 avg 7.02 MB/s (user 0.75s sys 3.26s) > read compiled tree total runs 8 avg 16.27 MB/s (user 0.86s sys 4.83s) > delete tree total runs 14 avg 21.51 seconds (user 0.45s sys 4.07s) > delete compiled tree total runs 6 avg 29.23 seconds (user 0.57s sys 4= =2E66s) > stat tree total runs 14 avg 15.99 seconds (user 0.41s sys 1.31s) > stat compiled tree total runs 10 avg 19.68 seconds (user 0.41s sys 1.= 51s) >=20 >=20 > ext4: >=20 > ./compilebench -D /mnt/work2 -i 90 -r 150 >=20 >=20 > run complete: > =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D > intial create total runs 90 avg 92.42 MB/s (user 0.76s sys 1.38s) > create total runs 20 avg 52.06 MB/s (user 0.73s sys 1.26s) > patch total runs 24 avg 17.28 MB/s (user 0.37s sys 0.96s) > compile total runs 23 avg 85.64 MB/s (user 0.17s sys 1.05s) > clean total runs 14 avg 887.53 MB/s (user 0.02s sys 0.11s) > read tree total runs 17 avg 13.60 MB/s (user 0.79s sys 1.34s) > read compiled tree total runs 8 avg 24.65 MB/s (user 0.95s sys 2.07s) > delete tree total runs 14 avg 2.26 seconds (user 0.39s sys 0.78s) > delete compiled tree total runs 6 avg 2.82 seconds (user 0.43s sys 0.= 91s) > stat tree total runs 14 avg 1.71 seconds (user 0.38s sys 0.33s) > stat compiled tree total runs 10 avg 1.99 seconds (user 0.42s sys 0.3= 3s) >=20 >=20 > I=C2=B4m a little disappointed about the btrfs results. > I was hopefully that btrfs is faster than ext4.... >=20 > - >=20 > Morten >=20 >=20 > 2009/4/4 Morten P.D. Stevens : > > Hi, > > > > here are some tests on an IBM server with btrfs vs. ext4. > > > > Kernel: 2.6.29.1 > > Benchmark software: compilerbench with options -i 10 -r 30 > > CPU: Intel Xeon Quadcore E5310 > > Chipset: Intel 5000 > > Memory: 4 GB FB-DIMM DDR2-667 > > HDDs: 2x WD6400AAKS @ Raid0 > > Storage Controller: IBM Serveraid 8k > > > > btrfs Result: > > > > intial create total runs 10 avg 50.89 MB/s (user 0.85s sys 2.59s) > > create total runs 5 avg 23.62 MB/s (user 0.82s sys 2.55s) > > patch total runs 4 avg 11.35 MB/s (user 0.38s sys 2.22s) > > compile total runs 7 avg 66.33 MB/s (user 0.19s sys 1.32s) > > clean total runs 4 avg 195.76 MB/s (user 0.03s sys 0.50s) > > read tree total runs 2 avg 11.99 MB/s (user 0.66s sys 2.59s) > > read compiled tree total runs 1 avg 30.14 MB/s (user 0.88s sys 3.64= s) > > delete tree total runs 2 avg 10.79 seconds (user 0.43s sys 3.39s) > > no runs for delete compiled tree > > stat tree total runs 4 avg 9.62 seconds (user 0.41s sys 1.03s) > > stat compiled tree total runs 1 avg 10.51 seconds (user 0.49s sys 1= =2E19s) > > > > ext4 Result: > > > > intial create total runs 10 avg 96.09 MB/s (user 0.77s sys 1.34s) > > create total runs 5 avg 50.84 MB/s (user 0.82s sys 1.20s) > > patch total runs 4 avg 20.17 MB/s (user 0.28s sys 1.04s) > > compile total runs 7 avg 94.39 MB/s (user 0.17s sys 1.07s) > > clean total runs 4 avg 959.66 MB/s (user 0.03s sys 0.11s) > > read tree total runs 2 avg 14.67 MB/s (user 0.78s sys 1.26s) > > read compiled tree total runs 1 avg 31.96 MB/s (user 0.87s sys 2.31= s) > > delete tree total runs 2 avg 2.14 seconds (user 0.34s sys 0.81s) > > no runs for delete compiled tree > > stat tree total runs 4 avg 1.82 seconds (user 0.40s sys 0.35s) > > stat compiled tree total runs 1 avg 1.83 seconds (user 0.35s sys 0.= 33s) > > > > Best regards, > > > > Morten > > > > > -- > To unsubscribe from this list: send the line "unsubscribe linux-btrfs= " in > the body of a message to majordomo@vger.kernel.org > More majordomo info at http://vger.kernel.org/majordomo-info.html -- To unsubscribe from this list: send the line "unsubscribe linux-btrfs" = in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html