[Starkit] Extraneous benchmark between Red Hat Lynux and UnixTru64 platforms

[Jeff Hobbs]

> Eduardo Rodriguez (Gerencia de Desarrollo) wrote:
> >  Hi everybody, I did a test consisted to update 10000 records on two
> > Metakit datafiles . The test was ran on two servers, a P4  with Lynux 
> > RedHat release 9 and on Alpha Server with Unix Tru64 V5.1. As result 
> > of running on both of them the times executing were a big different, 
> > but to my surprise significantly smaller on P4.
> >  Some data about environment on each platform:

> > Linux :
> > real    3m19.413s
> > user    3m19.070s
> > sys     0m0.150s
> >
> > Tru64:
> > real    9m48.80s
> > user    9m48.43s
> > sys     0m0.13s
> 
> It's hard to say what causes this.  One explanation for the very large 
> difference could be that memory-mapped files are not enabled on Tru64 
> (whether by accident or because it's not supported).  One way to find 
> out is to build Mk4tcl on Tru64 and check the HAVE_MMAP #define in the 
> generated config.h (or just look at the "configure" transcript).

Another important aspect is whether both tests were run on
local disks (and how fast those disks were).  For example,
the ActiveTcl windows installer is a metakit-based starpack.
If you run it from your local fs, it opens (starts up)
zippity-quick.  If you have it on a remote file system, it
can take a LONG time to start up (difference between like
5 seconds and up to 3 minutes).

Jeff