Date: Wed, 26 Apr 1995 16:40:48 -0700 (PDT) From: Julian Elischer <julian@ref.tfs.com> To: jdli@CCCA.NCTU.edu.tw (醉生夢死 無言以對) Cc: freebsd-hackers@FreeBSD.org Subject: Re: benchmark hell.. Message-ID: <199504262340.QAA13778@ref.tfs.com> In-Reply-To: <199504261859.CAA06439@CCCA.NCTU.edu.tw> from "醉生夢死 無言以對" at Apr 27, 95 02:59:13 am
next in thread | previous in thread | raw e-mail | index | archive | help
> > > hi : > > I did this test on the two machines... > > Linux 1.2.5 : > P90, 2940, AIR54CEP, 16MB RAM, 1G SCSI (slower than next HD) > > FreeBSD 0412-SNAP : > P90, 2940, AIR54CEP, 32MB RAM, 1G SCSI > > Only RAMs and HDs are different... > > Though the HD on linux is 1/3 speed of the one on FreeBSD base on iozone test, > but I found that Linux's disk-io is faster and more "quiet" than freebsd ... for SCSI devices UFS performance can be inproved and 'made more quiet" by disklabelling the drive as though it has only 1 head and 4096 sectors per track... (phk has checked this.....) this is because on modern drives the UFS code it TOO SMART and tries to take into account the "non-existant" geometry and actually ends up slowing down and rattling hte heads a lot. > > When un-tar the gcc-2.6.3, linux use almost the same time as freebsd, > and when delete the tree, linux is 3 times faster than freebsd... > (the HD on linux is 1/3 speed of the one on FreeBSD base on iozone !!) > > I guess maybe Linux will allocate as much memory as it can to be buffer. so will FreeBSD now.. > > Especially when the memory is more than 32mb or 64mb, linux will be much > more faster than freebsd on the same machine... nope, not for that reason.. modern SNAPs will do the same.... > > Anyway, I still love FreeBSD and hope it will be really fast someday !! > :-) > > ==== Linux test ==== > > Linux > free > total used free shared buffers > Mem: 14864 14608 256 5156 848 > -/+ buffers: 13760 1104 > Swap: 33772 1892 31880 We could do with this sort of thing :) > Linux > ./iozone 4 > Writing the 4 Megabyte file, 'iozone.tmp'...4.440000 seconds > Reading the file...2.230000 seconds > IOZONE performance measurements: > 944663 bytes/second for writing the file > 1880853 bytes/second for reading the file > FreeBSD > ./iozone 4 > Writing the 4 Megabyte file, 'iozone.tmp'...1.335938 seconds > Reading the file...0.421875 seconds > IOZONE performance measurements: > 3139595 bytes/second for writing the file > 9942053 bytes/second for reading the file FreeBSD read and wrote it a lot faster.... but 4MB is a test of cache code only.. > > Linux > time tar xfz gcc-2.6.3.tar.gz > 11.740u 10.630s 1:06.79 33.4% 0+0k 0+0io 81pf+0w > > FreeBSD > time tar xfz gcc-2.6.3.tar.gz > 11.626u 9.109s 1:05.82 31.4% 172+543k 155+4015io 0pf+0w > > > Linux > time /bin/rm -rf gcc-2.6.3 > 0.020u 0.660s 0:09.76 6.9% 0+0k 0+0io 25pf+0w > > FreeBSD > time /bin/rm -rf gcc-2.6.3 > 0.016u 0.907s 0:27.86 3.2% 151+304k 119+1836io 0pf+0w I believe this may be becuse of teh synchronous metadata updates needed for deleting >
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199504262340.QAA13778>