On Fri, 19 Mar 1999, Bill Studenmund wrote:

> On Fri, 19 Mar 1999, Paul H. Anderson wrote:
> 
> > The script uses a program 'maptest' which mmaps all listed files and
> > touches each page, forcing it to be pages into vm.  The directory that I
> > use for testing had 16 256MB files in it, so it tries to map 4GB of data.
> > 
> > I ran a test without the maptest, and it paniced.  I put it in thinking it
> > might make it panic sooner.  Giving you the source that is useful is
> > harder, cause it is part of my software, and I need to split out the
> > components that are essential for compilation.
> 
> I'm confused. The script causes a panic even w/o the maptest program? Then
> for now it's not needed. :-)

I'm not sure if it does or not.  I can't precisely remember under what
circumstances I saw the first panic (I know I was running the script with
just background tar files).  I do believe that with maptest, it may panic
a whole lot sooner.

I'll start working on a version which is more self-contained, source code
wise...

> Though a utility which mmaps a found file
> could be cool. :-)

Dunno what you mean by "mmaps a _found_ file" - maptest just takes the
filenames argv[1..n], mmaps them, touches all the pages, then dumps per
file paging and resource use - I was testing to make sure the VM system
was properly caching large amounts of data (1GB machine should cache >800
or so MB for mmap(), which alpha/netbsd now does, thankfully ).

> How big is the tar file again?

50MB or so - it is the entire netbsd source tree tar file.  I guess it
exands to a few hundred meg at least.  Suffice it to say, lots of files
get created.

BTW, after the panic, the filesystem is seriously corrupted... (*:

Paul

+------------------------------------------------------+
| Paul Anderson           Public Data Queries, Inc.    |
| pha@pdq.com             734-213-4964(W) 994-3734(H)  |
+------------------------------------------------------+