Re: [Announce] XFS 1.1 Prerelease 2 available for testing

[Warren Stockton <wns@slc.unisys.com>]

On Friday 22 March 2002 13:53, Seth Mos wrote:
> At 13:49 22-3-2002 -0700, Warren Stockton wrote:
> >Well done....
> >
> >Stability has improved tremendously from the earlier version and
> >it is far more econonical on memory.
>
> sarcasm ?
> If so in contrast to what?
No.  I had taken the XFS code from 2.4.9-6SGI-XFS1.0.2 and the XFS CVS updates
and merged it with RH 2.4.9.-21 and built with gcc3-3.0.1-3.    I was still 
working some issues when I saw this announcement yesterday.   

My hacked code would have 98% of memory allocated to cache within 3 minutes 
of starting the test. (Beyond this point it was hard-hats only and falling 
machine parts)

-31XFS1.1-pre2 running the same test has the cache somewhere between 25% to 
75% of physical memory, depending on where the testsuite is in its cycle.

So far I tested 70 drives active where total active open files = 50% of 
physical memory (16G)  (Test ran about 7hours with no errors or data 
corruption before I stopped it.)  IO stats showed numbers ext2 can only dream 
about.

Currently testing 70 drives active where total active open files = 3x 
physical memory.  Has been running 5hours with no lock-ups, errors or data 
corruption and the 32 copies of setiathome are still making good progress and 
the IO stats are literally 100x better that I ever got with ext2!!!

Time to hand over the the benchmark guys!!!

>
> >Now I just get a few of these show up intermittently while running
> >on 70 fibre drives on 5 qla2200 controllers.
>
> That doesn't seem very healthy. Is this compiled with kgcc, 2.96 or gcc3?
gcc-2.96-101
>
> Is this on ia64 or ia32?
ia64 C1 steppings
>
> Cheers