no, i measure creation, read time and deletion, and i mainly look at read time as creation/deletion is not so common in real life.<br>i also used bonnie++ and it showed even more tragic results for xfs, btrfs without any options:<br>
<br> ------Sequential Create------ --------Random Create--------<br> -Create-- --Read--- -Delete-- -Create-- --Read--- -Delete--<br>files:max:min /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP<br>
DO 20:10000:1/10000 3289 87 17424 100 3594 96 4393 85 33271 98 2888 7<br><br>xfs, tweaked to the max (mkfs with lazy-count=1, size=128m, and mount options noatime,nodiratime,logbufs=8,logbsize=256k (also tried nobarier):<br>
------Sequential Create------ --------Random Create--------<br> -Create-- --Read--- -Delete-- -Create-- --Read--- -Delete--<br>files:max:min /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP<br>
DO 10:20000:1/10000 650 3 11524 28 382 4 693 4 8242 14 552 5<br><br>xfs was half of what i posted above when created default, or even 1/3.<br><br>if it's hard to read, i will just compare one , probably most important thing, number of read random created files: 33271 on btrfs, 8242 on xfs (something like 3000 without optimizations on xfs i think)<br>
<br><br><div><span class="gmail_quote">2008/3/8, Tuncer Ayaz <<a href="mailto:tuncer.ayaz@gmail.com">tuncer.ayaz@gmail.com</a>>:</span><blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">
On Sat, Mar 8, 2008 at 7:50 PM, Rekrutacja119 <<a href="mailto:rekrutacja119@gmail.com">rekrutacja119@gmail.com</a>> wrote:<br> > running postmark with numbers set to 20 000, transactions to 10 000 and<br> > subdirectories to 20 000 i have following results with btrfs:<br>
> 24 seconds total<br> > 11 seconds of transactions (909 per second)<br> > Data:<br> > 26.48 megabytes read (1.10 megabytes per second)<br> > 136.18 megabytes written (5.67 megabytes per second)<br>
><br> > same for xfs is:<br> > Time:<br> > 70 seconds total<br> > 52 seconds of transactions (192 per second)<br> > Data:<br> > 26.48 megabytes read (387.33 kilobytes per second)<br>
> 136.18 megabytes written (1.95 megabytes per second)<br> ><br> > it looks MUCH worse with more realistic setup, of 100 000 subdirs, 100 000<br> > files and 20 000 transactions. xfs slowed to 65KB/s !! on 4 drive RAID5 (new<br>
> drives, just bought, seagete 500GB ones)<br> > btrfs was stable<br> ><br> > ext3 tests not possible because it would take like a day. (it is creating<br> > subdirs very very slow)<br> > ext4 is the same as ext3<br>
> reiserfs is comparable speed to xfs<br> > reiser4 is comparable speed to btrfs<br> ><br> > i only tested with postmark, as this is going to be users file array, with<br> > files up to 3MB, and mostly in 5-50KB range<br>
> so no need to test how many MB/s it can copy with large files<br> <br> <br>Am I right to assume that your issue is the creation<br> speed and not accessing the created 100.000 entries?<br> I have to admit that I didn't look at what postmark<br>
effectively measures.<br> <br><br> > also - how can i turn on dir_index on xfs? i think it's mkfs.ext3 option<br> > (and it is default on in /etc/mke2fs.conf)<br> ><br> <br> <br>AFAIK xfs like ReiserFS uses a B+ tree.<br>
</blockquote></div><br>