My original question was:
>What is the experience of users with setting the minfree parameter
>(i.e. -m) in newfs to less than 10%? At what point do you see
>performance degradation? I am installing a 4GB RZ29B.
>Many thanks.
A rough summary of the responses is that -m 5 and even lower than 5%
appears to not cause any problem. I am currently using -m 3 on an RZ29B.
Below are the responses.
------------------------------------------------------------------
>From rem_at_MATH.AMS.ORG Tue Apr 23 13:43:28 1996
I have recently added an external 4gb disk to an
alpha 1000 and used newfs -m 5 to create the filesystem
on it. I have not noticed any degradation in performance
on that so far.
-------------------------------------------------------------------
>From RJONES_at_hbsi.com Tue Apr 23 14:02:13 1996
it's not a performance issue, it's a safety issue. On file systems which
are fixed in size (db directories, etc) I use -m0.
--
rjones_at_hbsi.com, rjones_at_wicker.com (
http://www.wicker.com)
-------------------------------------------------------------------
>From alan_at_nabeth.cxo.dec.com Tue Apr 23 14:09:41 1996
Making an educated guess, I'd say that 3-4% is probably
safe. At 16 cylinders per group this allows an average
of ~512 KB/per group. This is consistent with 10% on
the disks that BSD had available at the time that the
10% default was chosen.
I also have to wonder at how much you'd notice the time
spent going to a linear search on systems that are over
200 time faster than the ones Berkeley was using when
minfree was implemented. Reading back poorly allocated
files may be a bigger problem, but if the file is read
so frequently that it is a major problem, then buffer
cache is doing a poor job...
------------------------------------------------------------------------
>From shollen_at_valhalla.cs.wright.edu Tue Apr 23 14:10:00 1996
We are routinely setting minfree to 5% or less. I took a UNIX internals
class with Kirk McKusick (one of the original authors of the UNIX
file system code) and he told us that 10% was basically pulled out of
the air as something that they thought would work. On extremely
large disks (4-9GB) we are going to 1-2% without noticeable problems.
Certainly when the code was written, disks were much smaller than
they are now. You may want to take into consideration the size of the
files normally created on your system. If they are very large, and
especially if your disks routinely run near capacity, you may want to
stay with a larger number. But we have not noticed any problems going
with smaller numbers.
--Sheila
---------------------------------------------------------------------------
>From aad_at_olympus.nwnet.net Tue Apr 23 15:15:27 1996
BSDI (and probably other 4.4 descendants) default it to 5%, claiming
that the usage patterns have changed since the days when 10% was
decided. I still like to run my filesystems <85% full, though.
---------------------------------------------------------------------------
Received on Thu Apr 25 1996 - 22:19:01 NZST