Hi Eric

It may become a limiting factor. Frankly, I have never done this type of
test. Of course, you should use -e. With this option most of word's data
is swaped to disk but the words an other info may still remain in
memory.

I know that this is not a clear answer but it is what I am able to say
at this moment.

Jose

On 30 Oct 2003 at 9:44, Ziegenhorn, Eric wrote:

> Awesome.  Do you know if it is likely that having 1.5 GB of RAM may
> become a limiting factor before the size of the index file with 64-bit
> support?
> 
> Thanks!
> Eric
> 
> -----Original Message-----
> From: jmruiz@boe.es [mailto:jmruiz@boe.es]
> Sent: Thursday, October 30, 2003 9:33 AM
> To: Ziegenhorn, Eric
> Cc: swish-e@sunsite.berkeley.edu
> Subject: Re: [SWISH-E] Re: A swish-e performance question
> 
> 
> 
> Hi Eric,
> 
> I am working now in 64 it support.
> 
> Jose
> 
> On 29 Oct 2003 at 10:09, Ziegenhorn, Eric wrote:
> 
> > On 17 Oct 2003 at 15:33, Patrick O'Lone wrote:
> > 
> > > All,
> > > 
> > > > 
> > > > > There is a 2GB limit index file. So I think that you will
> > > > > reach the limit of swish easily. But, the solution must be
> > > > > easy. You can hack the source the code and make it 64 bit. The
> > > > > trick is easy. Look for fopen, fclose, fseek, fwrite...
> > > > > routines in the code, mainly in db_native.c
> > > > >
> > > > > I have seen this question before. Anyone in the list has made
> > > > > anything? If not, and if I find some time I will try to add
> > > > > this feature. Now, I am fighting against the PHP module.
> > > > >
> > > 
> > > Can't you just compile with the -D_FILE_OFFSET_BITS=64
> > > -D_LARGEFILE_SOURCE? Shouldn't this just make it 64-bit assuming
> > > the file system the index resides on is LFS-aware? You shouldn't
> > > have to go through the code and convert fopen() to fopen64().
> > > 
> > 
> > >Nope. There must be also changed the offset types. fseek and ftell
> > >uses a long (usually 32 bit). 
> > >
> > >Basically, fseek anf ftell should be moved to fseeko and ftello.
> > >This uses an off_t type instead of a long type. So there are more
> > >things to change to allow swish internal data to hold the 64 bit
> > >offsets.
> > >
> > >I am working on this issue now. As Bill points, this may be a
> > >feature for a future 2.4.1.
> > >
> > >This is true for UNIX/LINUX system but Ms Win32 is a different
> > >story as Dave told me.
> > >
> > >cu
> > >Jose
> > 
> > 
> > Does anyone know that if Swish-e was changed to handle large files
> > over 2GB if memory usage would likely still be a limiting factor? 
> > I'm on a machine with 1.5GB RAM, but I'm trying to index 60GB of
> > files. Anyone know if this will be possible in the near future or
> > how I could make it possible?  Any guesses on a release date for
> > 2.4.1 with 64-bit file support?  I may be able to work on it if it
> > seems acheivable in the near term.
> > 
> > Thanks,
> > Eric
>