EXT3, EXT4, Btrfs Ubuntu Netbook Benchmarks

[phoronix]

Phoronix: EXT3, EXT4, Btrfs Ubuntu Netbook Benchmarks

Last month we published benchmarks of EXT4 comparing this file-system's performance when it was first marked stable in the mainline kernel and then where it is at now in the Linux kernel while testing every major release in between. This article was followed up by a Btrfs versus EXT4 comparison using the Linux 2.6.33 kernel to see how the two most talked about Linux file-systems are battling it out with the latest kernel. After those Linux file-system benchmarks were published, we received a request from Canonical to look at the EXT3 performance too. With that said, we have done just that and have published EXT3, EXT4, and Btrfs benchmarks from Ubuntu 9.10 and a Ubuntu 10.04 development snapshot from an Intel Atom netbook.

http://www.phoronix.com/vr.php?view=14600

[Kano]

Those benchmarks does not show the real speed improvement from ext4 to ext3. For me the DELETE performance of huge files beginning wit 1 GB up to over 25 GB is more important - and that takes ages with ext3. Compared to that ntfs (via ntfs-3g) is fast as hell - and also needs much space for itself. Especially when you only need to store data ntfs is a valid choice.

[cl333r]

I wouldn't use ntfs for storing data for it isn't posix compatible so you loose meta info (like whether the file is executable or not) when storing files, not to mention that writing to ntfs (for me) is (a lot) slower than to ext4 under Linux. Also, it's enemy's territory. Can't wait for the final btrfs and see whether it delivers what it promises.

[portets43]

Darnit Kano!

you're on a linux site! you're supposed to pretend that microsoft has never existed and that even if they did exist, everything linux is WAY more perfect than that fictional company could ever make!!

hahaha. sorry..

[Kano]

I use ntfs-3g since the beginning and have got absolutely no fat partitions left for anything bigger than 4 gb. For data storage they are really good if you like the company behind it or not. It can not be used for every kind of data - i would not compile on it or install Linux apps. But in lots of other cases it is fine. You should think more practial - especially for external storage.

[sabriah]

Here is a summary of the numbers



Raw data

BENCHMARK;Karmic_EXT3;Karmic_EXT4;Karmic_BTRFS;Luc id_EXT3;Lucid_EXT4;Lucid_BTRFS
SQLite_12000_inserts_sec;181,88;187,89;929,76;179, 98;183,02;600,41
Compile_Bench_Initial_Create_MBps;9,7;18,48;16,7;9 ,11;24,78;15,47
Compile_Bench_Compile_MBps;10,39;23,77;28,21;5,26; 29,08;61,67
Compile_Bench_Read_Compiled_tree_MBps;81,23;85,33; 73,86;58,46;83,6;73,72
IOZone_4GB_write_MBps;33,88;66,31;51,82;33,9;62,96 ;63,25
IOZone_4GB_read_MBps;91,01;102,1;107,8;101,21;99,7 5;88,4
IOZone_8GB_write_MBps;32,96;62,85;50,5;33,15;59,76 ;54,33
IOZone_8GB_read_MBps;65,93;97,44;107,89;100,99;98, 8;90,06
Dbench_12_clients_MBps;59,06;59,92;43,94;58,26;64, 08;41,64
Dbench_128_clients_MBps;2,7;6,87;21,66;21,34;10,86 ;16,33
FSMark_Files_ps;8,8;17,6;9,9;6,8;17,7;12
TIOT_64MB_rand_read_32t_MBps;16,23;27,07;17,13;16, 1;20,36;14,38
TIOT_64MB_rand_write_32t_MBps;1,2;1,36;7,06;1,2;1, 32;6,44
PostMark_Disk_transaction_Tps;139;196;206;126;321; 236
Kernel_unpacking_sec;64,16;61,26;68,51;63,6;64,15; 70,37


Logarithmic_transformations;;;;;;

BENCHMARK;Karmic_EXT3;Karmic_EXT4;Karmic_BTRFS;Luc id_EXT3;Lucid_EXT4;Lucid_BTRFS
SQLite_12000_inserts_sec;-2,259784946;-2,273903666;-2,968370858;-2,255224247;-2,262498551;-2,778447917
Compile_Bench_Initial_Create_MBps;0,986771734;1,26 6701967;1,222716471;0,959518377;1,394101302;1,1894 90314
Compile_Bench_Compile_MBps;1,016615548;1,376029182 ;1,450403086;0,720985744;1,463594402;1,790073948
Compile_Bench_Read_Compiled_tree_MBps;1,909716453; 1,931101746;1,868409303;1,766858811;1,922206277;1, 867585327
IOZone_4GB_write_MBps;1,529943402;1,821579028;1,71 4497409;1,530199698;1,799064719;1,80106053
IOZone_4GB_read_MBps;1,959089114;2,009025742;2,032 618761;2,005223425;1,998912904;1,946452265
IOZone_8GB_write_MBps;1,517987203;1,798305282;1,70 3291378;1,520483533;1,776410589;1,735039705
IOZone_8GB_read_MBps;1,819083076;1,988737275;2,032 981193;2,004278372;1,994756945;1,954531943
Dbench_12_clients_MBps;1,771293443;1,777571805;1,6 42860053;1,76537048;1,806722503;1,619510721
Dbench_128_clients_MBps;0,431363764;0,836956737;1, 335658452;1,329194415;1,035829825;1,212986185
FSMark_Files_ps;0,944482672;1,245512668;0,99563519 5;0,832508913;1,247973266;1,079181246
TIOT_64MB_rand_read_32t_MBps;1,21031852;1,43248825 6;1,233757363;1,206825876;1,308777774;1,157758886
TIOT_64MB_rand_write_32t_MBps;0,079181246;0,133538 908;0,848804701;0,079181246;0,120573931;0,80888586 7
PostMark_Disk_transaction_Tps;2,1430148;2,29225607 1;2,31386722;2,100370545;2,506505032;2,372912003
Kernel_unpacking_sec;0,015586035;0,016323865;0,014 596409;0,01572327;0,015588465;0,014210601


AVERAGE;1,004977471;1,176814991;1,162781742;1,0387 66564;1,208567959;1,184748775
MEDIAN;1,21031852;1,432488256;1,450403086;1,329194 415;1,463594402;1,619510721


RANK_per_benchmark;;;;;;

BENCHMARK;Karmic_EXT3;Karmic_EXT4;Karmic_BTRFS;Luc id_EXT3;Lucid_EXT4;Lucid_BTRFS
SQLite_12000_inserts_sec;2;4;6;1;3;5
Compile_Bench_Initial_Create_MBps;5;2;3;6;1;4
Compile_Bench_Compile_MBps;5;4;3;6;2;1
Compile_Bench_Read_Compiled_tree_MBps;3;1;4;6;2;5
IOZone_4GB_write_MBps;6;1;4;5;3;2
IOZone_4GB_read_MBps;5;2;1;3;4;6
IOZone_8GB_write_MBps;6;1;4;5;2;3
IOZone_8GB_read_MBps;6;4;1;2;3;5
Dbench_12_clients_MBps;3;2;5;4;1;6
Dbench_128_clients_MBps;6;5;1;2;4;3
FSMark_Files_ps;5;2;4;6;1;3
TIOT_64MB_rand_read_32t_MBps;4;1;3;5;2;6
TIOT_64MB_rand_write_32t_MBps;5;3;1;5;4;2
PostMark_Disk_transaction_Tps;5;4;3;6;1;2
Kernel_unpacking_sec;4;1;5;2;3;6

RANK_SUM;70;37;48;64;36;59

Whatever, but following the ranks, EXT4 is the winner using these benchmarks. Also using the averages of the logarithmic data sets (with timings in sec first inverted), EXT4 is the winner. However, using the median of the logarithmic data sets, BTRFS is the winner!

Sex, lies, and statistics.



.

[jakubo]

what i am wondering about is the introduction.
why does phoronix launch a test when canonical wants it?
i actually thought that FSs are kernel issue not ubuntu-bound.

maybe im going paranoid, but is there some kind of treaty? is the PTS based on ubuntu? or...
looks to me as if canonical was developing momentum im rather being afraid of and which has few of the linux spirit left, does it?
though it has a positive point: someone being behind it and giving developement a boost, although it might influence other projects...

[portets43]

I use ntfs-3g since the beginning and have got absolutely no fat partitions left for anything bigger than 4 gb. For data storage they are really good if you like the company behind it or not. It can not be used for every kind of data - i would not compile on it or install Linux apps. But in lots of other cases it is fine. You should think more practial - especially for external storage.

i didn't know ntfs was any good. but anyway, i've never been against windows.
microsoft actually makes some good quality solid products(with a few exceptions).

for example; god i loved windows 2000 and xp!! they were sooo solid.

i just hate microsofts company ethics. microsoft is a D***!

[portets43]

what i am wondering about is the introduction.
why does phoronix launch a test when canonical wants it?
i actually thought that FSs are kernel issue not ubuntu-bound.

maybe im going paranoid, but is there some kind of treaty? is the PTS based on ubuntu? or...
looks to me as if canonical was developing momentum im rather being afraid of and which has few of the linux spirit left, does it?
though it has a positive point: someone being behind it and giving developement a boost, although it might influence other projects...

i don't think there's any treaty kind of thing. but it looks like michael does have a thing for ubuntu.

but i do agree with you that canonical has lost the linux spirit, if they ever even had it..

lately they've just seemed waaayy too money driven.

[Michael]

what i am wondering about is the introduction.
why does phoronix launch a test when canonical wants it?
i actually thought that FSs are kernel issue not ubuntu-bound.

maybe im going paranoid, but is there some kind of treaty? is the PTS based on ubuntu? or...
looks to me as if canonical was developing momentum im rather being afraid of and which has few of the linux spirit left, does it?
though it has a positive point: someone being behind it and giving developement a boost, although it might influence other projects...

Hmm? If the kernel team, Novell, or any other major company asked for particular tests to be carried out, I would be happy to run them publicly on Phoronix. It's been done before for Red Hat.