Date: Sat, 8 Apr 1995 22:14:25 +1000 From: Bruce Evans <bde@zeta.org.au> To: Ollivier.Robert@keltia.frmug.fr.net, rgrimes@gndrsh.aac.dev.com Cc: freebsd-current@FreeBSD.org Subject: Re: Label/slices : how to add a disk ? Message-ID: <199504081214.WAA26653@godzilla.zeta.org.au>
next in thread | raw e-mail | index | archive | help
>Even the 128 512 iozone result is amazing : Not really. >206 [18:22] root@keltia:/mnt# ~/Src/C/iozone_2.01/iozone 128 512 > IOZONE writes a 128 Megabyte sequential file consisting of > 262144 records which are each 512 bytes in length. > ... >Writing the 128 Megabyte file, 'iozone.tmp'...86.078125 seconds >Reading the file...48.976562 seconds >IOZONE performance measurements: > 1559254 bytes/second for writing the file > 2740448 bytes/second for reading the file As explained in other mail, clustering does a good job of turning the too-small 512-byte i/o's into 64K i/o's. The system has to do a lot more work to do the clustering and other things for many more buffers, but if there are enough cycles for this then the i/o speed doesn't suffer. Your speed is slower so you must not have enough cycles :-). This is not surprising for a 486/33. How fast is your controller for `dd if=/dev/rsd0 bs=512 count=4000'? Clustering doesn't apply to the raw device so a block size of 512 really is used. OTOH, unclustering applies to the block device /dev/sd0: for `dd if=dev/sd0 bs=64k count=4000', the specfs driver turns the nice 64K block size into the stupidly small size BLKDEV_IOSIZE = 2K. Bruce
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199504081214.WAA26653>