Andrea Franceschini writes:
> Simon Burge wrote:
> > 
> > Andrea Franceschini wrote:
> > 
> > > On raid5 Partition :
> > >
> > >               -------Sequential Output-------- ---Sequential Input-- --Ra
> ndom--
> > >               -Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Se
> eks---
> > > Machine    MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU /sec
>  %CPU
> > >           100  2504 15.4  2468  6.8  2749  7.8 21473 93.2 86272 99.9 2311
> .4 27.9
> > >                ^^^^
> > > On non-raid partition :
> > >
> > > Machine    MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU /sec
>  %CPU
> > >           100 12802 96.0 15757 74.2 15847 73.0 23237 99.9 86406 100.0 852
> 3.6 100.1
> > >               ^^^^^
> > > [ ... ]
> > >
> > > NetBSD 1.5R (NETAPP) #1: Thu Feb  8 05:29:20 UTC 2001
> > >     root@netapp:/usr/src/sys/arch/i386/compile/NETAPP
> > > cpu0: AMD K6-2 (586-class), 501.16 MHz
> > > cpu0: features 8021bf<FPU,VME,DE,PSE,TSC,MSR,MCE,CX8>
> > > cpu0: features 8021bf<PGE,MMX>
> > > total memory = 255 MB
> > > avail memory = 234 MB
> > > using 3296 buffers containing 13184 KB of memory
> > 
> > One quick comment - it's no use testing disk I/O with bonnie if you are
> > testing a file size less than the buffer cache size.  That's why you are
> > seeing 80MB/s in some cases.  With -current and a box with 256MB of RAM,
> > you'd be wanting to use a 256MB test as an absolute minimum - larger
> > would be better.
> > 
> > The "using 3296 buffers containing 13184 KB of memory" line refers to
> > buffers for metadata - things like inode info and so on.  Buffers for
> > actual file data use (pretty much) any free memory of your 234MB.
> > 
> Thanks for the comments.
> 
> Yes, i know that bonnie wasn't such a good tool for disk speed
> benchmark,but as you can
> argue from my post,the problem is focused on write speed.
> 
> I don't know if it is a speed issue at all.... i mean ... the loss of
> speed  was a
> conseguence of something else .....a raidframe problem for istance..

What sort of speed were you seeing before?  There havn't been any changes in 
the RAIDframe code from 1.5 which would cause a huge performance drop.

> Someone has a similar configuration (4-IDE/RAID5)?It works?

It works, but having each disk on it's own channel is better, and having 5 
disks instead of 4 is better still.. You also want to tweak things a bit, but 
if you do, you can get reasonable perf:

     -------Sequential Output-------- ---Sequential Input-- --Random--
     -Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks---
  MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU  /sec %CPU
2000  8139 89.3 14519 79.7  3024 19.4  7792 97.6 23553 91.1  62.3  4.6

(RAID 5 set, PII-something, w/ 5 60GB Maxtors, each on their own channel
of a Promise Ultra66 controller.. OS was an early 1.5_alpha)

Also: Try doing 4 simultaneous bonnies (one on each disk, all at the same 
time), and see what you get for overall performance.

Later...

Greg Oster