Re: Informix 2.10 Splitting Data Across File Systems?

>From: jh@cadre.com (Joe Hartley) >Subject: Re: Informix 2.10 Splitting Data Across File Systems? >Date: Thu, 29 Sep 1994 13:13:46 GMT >X-Informix-List-Id: <news.8963> > >We know now that there are two ways to split a DB across file systems. >It seems to me that Jon's approach is easier from an implementation >point of view than Alan's. Are there any drawbacks to either method? >does the symbolic link pose any hits on performance? It seems that >the 2 different approaches would work equally well. Any disagreements? "Alan's" approach (using CREATE TABLE TabName ( ... ) IN "/some/where";) has the merit of being officially supported, and makes it obvious when you look at the system catalogue that the actual data is not stored in the database directory proper. However, it does not allow you to separate the index and the data files -- they have to live in the same directory. "Jonathan's" approach (using symbolic links) allows the data and index files to be stored separately, but could be combined with Alan's approach anyway. There is a marginal overhead in tracing a symlink when the file is opened, but on a scale of 0-100 we're looking at 0.003, or some equally preposterously small figure. Let's put it like this: I can't conceive of a situation where the symlink could cause the application to slow down noticably. So, the basic answer is: Yes, the two different approaches work equally well. One word of warning -- you should not try to use a symlink to NFS-mounted resources. NFS is a stateless file system, and locking and updates are state-full operations, so the product doesn't allow it. Yours, Jonathan Leffler (johnl@informix.com) #include <disclaimer.h>