On 8/24/06, Quentin Garnier <cube@cubidou.net> wrote:
> Hi,
>
> I'm trying to understand the results of a run of bonnie++.  There is
> machine A, which happens to be my desktop station, running a Xen2 dom0
> kernel, with ~750MB of RAM and a wd(4) drive attached to a piixide(4)
> controller under UDMA5.  So far so good.
>
> There is machine B, which is a newish Dell running i386 with 4GB of
> memory (well, 3GB under NetBSD/i386).  I started doing tests on a disk
> attached to a mpi(4) device, but I switched to doing tests on a wd(4)
> drived attached to a piixide(4) controller under UDMA6.
>
> Note that softdeps are turned on for the considered mount points, and
> that both machines run -current from this week.
>
> IO throughput is normal in both cases.  Both seems pretty fast.  Of
> course they really haven't the same amount of memory so I used different
> file sizes.
>
> Version  1.03       ------Sequential Output------ --Sequential Input- --Random-
>                     -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
> Machine        Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP  /sec %CP
> A                1G 41420  50 39275  26 +++++ +++ 41311  43 42128  13 369.2   1
>
> Machine        Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP  /sec %CP
> B                8G 61225  62 61574  40 30519  17 64889  87 65111  21 161.6   3
>
> Directory operations are another story:
>
>                     ------Sequential Create------ --------Random Create--------
>                     -Create-- --Read--- -Delete-- -Create-- --Read--- -Delete--
> Machine       files  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP
> A               100 10195  26 92042  96 28823  54 14703  39 79685  96 29397  53
>
> Machine       files  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP
> B               100   590  74 +++++ +++  2145  55   596  74  1013  99  1088  69
>
> Only the sequential stats match the performance of machine A.  It shows
> up as +++++ here but I got a number a couple of times and it was the
> same order as machine A.
>
> But how do I explain the rest?  I mean, "random read".  That's almost 80
> times slower!
>
> Partitions info:
> A -> g: 122880240  24289776     4.2BSD   2048 16384   328
> B -> e:  31461696  33559800     4.2BSD   2048 16384 28104
>
> What's the exact meaning of the last field?  Is it relevant to the
> issue?  I simply newfs'd the partition.  The partition on machine A was
> newfs'd a long, long time ago.
>

Note that incerasing the file size 8 times increases the seek times
dramatically which _should_ result in poor random read/write
performance.

Thanks

Michal